Direct Observation and Quantification of CO2 Binding Within an Amine-Functionalized Nanoporous Solid

See allHide authors and affiliations

Science  29 Oct 2010:
Vol. 330, Issue 6004, pp. 650-653
DOI: 10.1126/science.1194237

Designing Carbon Dioxide Traps

One widely discussed means of stemming the rise in atmospheric carbon dioxide concentration is to capture the gas prior to its emission and then bury it. The materials currently known to best adsorb CO2 for this purpose tend to involve amine groups; however, the precise molecular details of adsorption often remain murky, and rational improvement of sorbent properties by structural modification has been challenging. Vaidhyanathan et al. (p. 650; see the Perspective by Lastoskie) have crystallographically resolved the binding motifs of CO2 in an amine-bearing metal-organic framework solid. Accompanying theoretical simulations matched the experimental observations.


Understanding the molecular details of CO2-sorbent interactions is critical for the design of better carbon-capture systems. Here we report crystallographic resolution of CO2 molecules and their binding domains in a metal-organic framework functionalized with amine groups. Accompanying computational studies that modeled the gas sorption isotherms, high heat of adsorption, and CO2 lattice positions showed high agreement on all three fronts. The modeling apportioned specific binding interactions for each CO2 molecule, including substantial cooperative binding effects among the guest molecules. The validation of the capacity of such simulations to accurately model molecular-scale binding bodes well for the theory-aided development of amine-based CO2 sorbents. The analysis shows that the combination of appropriate pore size, strongly interacting amine functional groups, and the cooperative binding of CO2 guest molecules is responsible for the low-pressure binding and large uptake of CO2 in this sorbent material.

The capture and storage of CO2 emitted from industrial processes are global challenges. In many industrial processes, CO2 is present at low partial pressures among other gases that ideally should be recycled. The currently employed capture method involves alkanolamine-based solvents that act as CO2 scrubbers (1, 2) by chemisorptive formation of N-C bonded carbamate species (bonding energies are typically 100 kJ mol−1). Regeneration of the amine requires cleavage of this covalent bond by heating (at 100° to 150°C) to release CO2. Major drawbacks of this process include the corrosive nature and volatility of the amines, their occasional decomposition, and most prominently, the high energy cost of their regeneration (1, 2). The challenge is thus to couple efficient CO2 capture with facile release in a sorbent material.

Porous systems, including zeolitic/zeotypic materials (3, 4), mesoporous silica (58), porous carbon (9), and, more recently, metal-organic frameworks (MOFs) (1019), have been investigated for CO2 storage. Merging the inherent sorptive behavior of porous solids with less-basic amines offers a route to the sort of easy-on/easy-off materials described above. Less-basic amines would favor physisorption over chemisorption of CO2, thus greatly reducing the energy of regeneration. This prospect has prompted research on many amine-functionalized solid materials that on the whole demonstrates that amines can enhance CO2 uptake (58, 1719). Despite this conclusion, experimental insights at a molecular level on the nature of the NH2·⋅·CO2 interaction are lacking (20). For materials such as silica and carbon, a combination of factors (lack of order, large voids, flexible amine groups, and random adsorption sites) makes the study of individual sorptive interactions virtually unfeasible. In contrast, the crystallinity of MOFs enables diffraction experiments to study structure at a molecular level. Beyond characterization of the framework, in exceptional cases, x-ray or neutron diffraction can allow direct visualization of gases within pores (2127) to elucidate specific binding interactions and enable better sorbent design. Locating gas molecules in a MOF is challenging, but the systems typically display strong confinement effects on the guest molecules and/or specific sites of strong interaction (such as bare metal sites) that can serve as excellent models for understanding the interactions of gases in all porous systems.

We previously noted (28) that the MOF Zn2(Atz)2(ox) (1) (Atz, 3-amino-1,2,4-triazole; ox, oxalate) showed CO2 uptake at low pressures with an initial heat of adsorption (ΔHads) of ~40 kJ mol−1 (Fig. 1 and fig. S4). After manifesting the expected trend of decreased ΔHads with increased coverage, the material showed a subsequent increase in ΔHads, which remained over 35 kJ mol−1 during continuous gas exposure, suggesting a cooperativity effect in the binding mechanism. Here we describe a detailed study, via a combination of crystallographic and computational methods, of the nature of CO2 binding in 1. In a crystal structure of 1 loaded with CO2 molecules, the CO2 binding sites are readily identified, even from room temperature diffraction data. The characteristics of CO2 uptake in 1, including isotherm, heat of adsorption, and location of CO2 molecules, are modeled with high accuracy via a combination of classical grand canonical Monte Carlo (GCMC) simulations, molecular dynamics (MD) simulations, and periodic density functional theory (DFT) calculations. Thus, 1·(CO2)1.3 serves to calibrate these methods for modeling gas sorption in MOFs. The modeling enables the partitioning of CO2 binding in 1 into components based both on neighboring groups and the nature of interaction (electrostatics/dispersion).

Fig. 1

(A) Structure of the Zn-Atz layer in 1 (Zn, cyan; C, black; N, blue; H, not shown). (B) Three-dimensional structure of 1, wherein the Zn-Atz layers are pillared by oxalate moieties (O, red) to form a six-connected cubic network shown as green struts. (C) Adsorption isotherms for different gases carried out using 1. The inset shows heat-of-adsorption data calculated with the CO2 isotherms measured at 273 and 293 K (fig. S5). The zero-loading heat of adsorption was calculated, using a model based on the virial equation (figs. S6 and S7 and table S2), to be 40.8 ± 0.8 kJ mol−1.

Single crystals of 1 were prepared via a procedure modified from that previously reported (28), for the phases with hydrated and evacuated pores (29). X-ray crystallography of evacuated 1 showed no electron density in the voids, thus confirming the effectiveness of the activation procedure and the stability of the crystal. CO2 was then loaded into evacuated crystals of 1, and x-ray diffraction experiments conducted at 123, 173, 195, and 293 K all yielded refinable data. The CO2 molecules could be located within the pores in all cases and were ordered except at 293 K, where the disorder could be modeled. The 173-K data set yielded the best refinement parameters [refinement factor (R) = 2.7%, weighted R (Rw) = 6.5%] and will be used for structural discussions.

The 173-K structure refinement gave a formula of 1·(CO2)1.30, which agrees well with the calculated loading of 1.35 CO2 from the adsorption isotherms at 840 mbar and 293 K (which are comparable conditions to those of single-crystal experiments). In the lattice Zn, aminotriazolate layers are pillared by oxalate ions, with free amine groups lining the pores (Fig. 1).

Within each pore, two independent CO2 binding sites were located: CO2-I [O(100)-C(100)-O(101)] and CO2-II [O(200)-C(200)-O(201)] (Fig. 2 and fig. S1). CO2-I, near the free amine group, was 80% occupied, and CO2-II, closer to the oxalates, was 50% occupied; in filled pores, neither CO2 molecule showed positional disorder. CO2 could interact with an amine via N-H·⋅·O hydrogen bonding or via an interaction between the N lone pair and the C atom of CO2. The H atoms of the amine groups were readily located in the x-ray structure of 1·(CO2)1.30, enabling direct visualization of the H-bonding. CO2-I was adjacent to the amine, with its electropositive C atom oriented toward the electronegative N atom [C(δ+)·⋅·N(δ–) = 3.151(8) Å; the C-N of monoethanolamine carbamate was modeled as 1.45 Å (30) and factoring C and N van der Waal radii = 3.25 Å]. The bowing of the protons on the N atoms (~21° from the mean triazole plane) confirms that the lone pair is not delocalized into the triazole ring (Fig. 2). Both O atoms of CO2-I are within the range of longer H-bonds to the amine [N4-H·⋅·O100 = 3.039(4), O101 = 3.226(9) Å], with angles also corroborating very weak interactions [∠ N-H·⋅·O: O100 = 97.486(1)°; O101 = 95.822(1)°]. The C atom of CO2-I also interacts with an oxalate O atom lining the pore [C(δ+)…O(δ–) = 3.155(8) Å]. The amine undergoes stronger H-bonding with oxalate O atoms [N4-H·⋅·O2 = 2.888(7), O4 = 2.122(2) Å, ∠ N-H·⋅·O = 154.718(1)° and 137.843(1)°, respectively]. CO2-II was located between oxalate groups along the b axis. The O(δ–) of CO2-II interacts with the C(δ+) of the oxalate [O(δ–)(CO2)…C(δ+)(Ox) = 2.961(5) Å]. In contrast to CO2-I, CO2-II interacts with a proximal amine via only a single H-bond [(N8-H·⋅·O200 = 2.783(8), ∠ N-H·⋅·O = 101.953(1)°]. There is an interaction between CO2 molecules as one O atom of CO2-I forms a contact with the C of CO2-II [(C(δ+)…O(δ–) = 3.023(7) Å, C200·⋅·O100]. This CO2-CO2 interaction is highly relevant because it is most likely the origin of the observed increasing ΔHads with loading.

Fig. 2

X-ray structure of CO2 binding in 1·(CO2)1.3 at 173 K. (A) The role of the amine group of Atz in binding CO2-I is depicted. The H atoms of the amine group (located crystallographically) H-bond to oxalate O atoms, directing the N lone pair toward the C(δ+) atom of the CO2 molecule. H-bond distances shown are for H-acceptor interactions. (B) Both crystallographically independent CO2 molecules are shown trapped in a pore, showing the cooperative interaction between CO2-I and CO2-II molecules. The CO2…NH2 interaction is represented as a dotted purple bond, and the CO2…CO2 interaction is indicated as a dotted yellow bond. (C) This panel shows the other interactions present. The CO2-I…Ox interactions are shown in orange, and the CO2…NH2 hydrogen bond interactions are shown in green. For clarity, H atoms are shown in purple.

Regarding the geometries of the CO2 molecules, both appear, with sizeable uncertainties, slightly bent [∠O-C-O: CO2-I, 175.72(1.01)°; CO2-II, 177.15(1.66)°]. Given that a 3σ range of angles approaches or includes linearity and the fact that neither chemical intuition nor the modeling studies support a nonlinear structure, the apparent bend of the CO2 molecules probably arises from their positional distributions rather than any true distortion in bonding. The C-O bond lengths [CO2-I: 1.137(8), 1.079(9) Å; CO2-II: 1.141(13), 1.125(13) Å] were slightly shorter than reported [1.155(1) Å] in a 150-K/ambient pressure structure of pure CO2 (31). Variable-temperature x-ray crystallography showed that although the CO2 bond lengths and angles varied slightly with temperature, their occupancies and orientations did not appreciably change (table S1). A decrease in total gas uptake would be expected with increasing temperature; however, for 1, this did not change between 195 and 273 K. This observation, coupled with the order of the gas molecules, reinforces the fact that the collective interactions are highly favorable for CO2 binding. Insight regarding the specific interactions was gained through computational modeling.

To investigate the nature of the CO2 interactions with 1 and the cooperative guest-guest binding effects, we used a combination of dispersion-corrected (32) periodic DFT calculations and classical GCMC simulations and MD simulations (33), in which the partial charge parameters were derived from the periodic DFT calculations by means of the REPEAT method (34). Figure 3A displays the excellent agreement between the experimental and GCMC-simulated CO2 adsorption isotherms of 1 at 273 K. The parameters associated with the CO2-1 intermolecular potential were not adjusted to obtain this quality of fit. Figure 3B shows a center-of-mass probability density plot from a GCMC simulation of 1 (at 850 mbar and 273 K), where darker regions reveal a greater probability of finding CO2. Even at 273 K, binding sites are well localized, and the symmetry of the probability clouds suggests two distinct binding sites, corroborating the crystallography. To compare the CO2 binding sites determined computationally and crystallographically, the experimental CO2 positions are superimposed on three-dimensional isosurfaces of the probabilities for C and O in Fig. 3C. The remarkable agreement between the simulated and experimental CO2 positions suggests that GCMC simulations that are often used to study gas adsorption in MOFs (33, 35) not only reproduce adsorption isotherms but can also accurately reproduce specific binding sites.

Fig. 3

(A) Comparison of the simulated (red) and experimental (blue) CO2 gas adsorption isotherm of 1 at 273 K. The inset shows a comparison of experimental and simulated heats of adsorption as a function of guest loading. (B) Center-of-mass probability density plots of CO2 molecules in 1 from a GCMC simulation at 850 mbar and 273 K. Probabilities along the b axis are summed and projected onto the ac plane. (C) Comparison of the location of the CO2 binding sites in 1 obtained from x-ray analysis and those predicted by simulation. Probability isosurface plots of the CO2 O atoms are shown as transparent red surfaces and those of the C atoms as transparent cyan surfaces. The 16 experimentally determined CO2 molecule positions are shown as tubes. An isosurface value of 0.15 Bohr−3 is used. (D) Trace of a single CO2 molecule during a MD simulation of 1 with five CO2 molecules per unit cell or a loading of 1.6 mmol g–1. Shown are 23 successive snapshots, separated by 62.5 ps. Other CO2 molecules present in the simulation are not shown. In (B) and (D), the approximate locations of binding sites I and II in a single channel are shown with dotted ellipses (site I, black; site II, red). In (B) and (C), a single unit cell of MOF 1 is shown, whereas in (D) a 2 × 1 × 1 cell is depicted. In (B) to (D), the MOF framework is shown as lines and the CO2 molecules are shown as tubes (C, light blue; O, red; N, dark blue; Zn, not shown).

The nature of the CO2 binding was further studied with dispersion-corrected periodic DFT calculations. With all 16 binding sites in the unit cell occupied, the resulting fully optimized geometry is in excellent agreement with the x-ray structure, including the relevant CO2-amine and CO2-oxalate distances (fig. S8). The exception to this was the geometries of both CO2 molecules, which, unlike their geometry in the x-ray analysis, were optimized to linear configurations, revealing that there is minimal geometric distortion from interaction with the framework. Radial distribution plots extracted from the GCMC simulations are also in agreement with the geometric parameters determined by crystallography (fig. S10). Although specific binding sites were located experimentally, the CO2 binding is expected to be dynamic in nature. This prospect has been investigated via classical MD simulations, which show that, in a 1.4-ns time span, the CO2 molecule can hop between several of the binding sites I and II (Fig. 3D).

The CO2 binding energies were calculated at various occupancies at the DFT level (29). When 1 is empty, CO2-I has a binding energy of 39.6 kJ mol−1, which is in good agreement with the experimental zero-loading ΔHads (40.8 ± 0.8 kJ mol−1). We see a strong cooperative enhancement of CO2 binding that increases with loading. Specifically, the binding energy of CO2-II increases by 4.6 kJ mol−1 to 37.0 kJ mol−1 when an adjacent site I is occupied. When 1 is fully occupied less one binding site, the binding energy for CO2-II increases to 38.1 kJ mol−1, whereas that of CO2-I is 44.2 kJ mol−1. Further insight into the nature of the CO2 binding can be gained by partitioning the total binding energy. This analysis reveals that 82% of the binding energy of site I is due to dispersion, whereas 18% results from electrostatics (29). This contrasts with site II, where the binding energy is almost entirely (99%) due to dispersion interactions.

The cooperative binding in 1 can be attributed to a combination of dispersion and electrostatic interactions between CO2 molecules, which can be quantified by using DFT to evaluate the interaction energy between two CO2 molecules at sites I and II in a vacuum (29). This calculation gives a CO2-CO2 interaction energy of 3.9 kJ mol−1, which accounts for most of the observed 4.6–kJ mol−1 binding enhancement. Of the 3.9–kJ mol−1 interaction between CO2 molecules, 66% can be attributed to dispersion and 34% can be assigned to electrostatics. The same analysis, with the MOF fully loaded, attributes 61% of the interaction to dispersion and 39% to electrostatics, suggesting similar cooperativity at higher loadings.

Several key insights obtained from the detailed analysis of the binding interactions in 1 have implications for the design of future materials used for the physisorption of CO2. The relatively high binding energies observed in the material are dominated by dispersion interactions. Because CO2 has a substantial quadrupole moment, there is opportunity to further increase the binding energies through appropriately designed binding sites that maximize the electrostatic interactions with CO2. The importance of the cooperative guest binding to the uptake of CO2 in 1 is another key insight. In particular, the proper mutual orientation and high density of binding sites can be used as a strategy to increase CO2 binding energies. Further, we believe that these strategies can be incorporated into materials with larger pores, in order to increase the overall CO2 uptake capacity and binding energies, thereby improving the uptake properties in the important low–partial-pressure regime. In the broadest sense, via detailed modeling, this study lays the groundwork for the design of easy-on/easy-off physisorptive materials for CO2 capture, designed from the characteristics of the guest molecules outward rather than from the host framework inward.

Supporting Online Material

Materials and Methods

SOM Text

Figs. S1 to S12

Tables S1 to S5


Movie S1

References and Notes

  1. Supporting material on Science Online includes complete details of the crystallography (crystallographic information files for all temperatures), gas sorption experiments, and computational modeling (including a movie showing the correlation of experimental and modeled CO2 positions in 1).
  2. T.K.W. and G.K.H.S. thank the Natural Sciences and Engineering Research Council of Canada. G.K.H.S. thanks the Institute for Sustainable Energy, Environment and Economy at the University of Calgary and the Alberta Energy Research Institute for partial financial support of this work. T.K.W. and P.G.B. thank the Canada Research Chairs Program and the High Performance Computing Virtual Laboratory for financial support. Crystallographic data for the structures reported in this paper have been deposited with the Cambridge Crystallographic Data Centre (CCDC) under reference numbers CCDC 782638 to 782642. These data can be obtained free of charge via (or from the CCDC, 12 Union Road, Cambridge CB2 1EZ, UK).
View Abstract

Navigate This Article