Unravelling the structure of chemisorbed CO2 species in mesoporous aminosilicas: a critical survey

Chemisorbent materials, based on porous aminosilicas, are among the most promising adsorbents for direct air capture applications, one of the key technologies to mitigate carbon emissions. Herein, a critical survey of all reported chemisorbed CO2 species, which may form in aminosilica surfaces, is performed by revisiting and providing new experimental proofs of assignment of the distinct CO2 species reported thus far in the literature, highlighting controversial assignments regarding the existence of chemisorbed CO2 species still under debate. Models of carbamic acid, alkylammonium carbamate with different conformations and hydrogen bonding arrangements were ascertained using density functional theory (DFT) methods, mainly through the comparison of the experimental 13C and 15N NMR chemical shifts with those obtained computationally. CO2 models with variable number of amines and silanol groups were also evaluated to explain the effect of amine aggregation in CO2 speciation under confinement. In addition, other less commonly studied chemisorbed CO2 species (e.g., alkylammonium bicarbonate, ditethered carbamic acid and silylpropylcarbamate), largely due to the difficulty in obtaining spectroscopic identification for those, have also been investigated in great detail. The existence of either neutral or charged (alkylammonium siloxides) amine groups, prior to CO2 adsorption, is also addressed. This work extends the molecular-level understanding of chemisorbed CO2 species in amine-oxide hybrid surfaces showing the benefit of integrating spectroscopy and theoretical approaches.

The solution was then stirred at 35 °C for 24 h and subsequently heated at 100 °C for 24 h, under static conditions. The obtained solid was filtered and dried in air. After, the solid was calcined at 550 °C for 5 h with a ramp of 1 °C/min. The resulting material (SBA-15) was stored in a desiccator for further use.
Dryness conditions of the reaction media and material are of paramount importance to prevent extension of lateral silane polymerization within the materials and to allow an efficient silane functionalization. To achieve this, typically 2 g of SBA-15 were introduced in a closed reflux apparatus connected to a vacuum line and heated overnight at 140 °C. After cooling, the nitrogen was introduced into the system prior to the opening of the reflux apparatus, and SBA-15 was refluxed with 100 cm 3 of dry toluene (Aldrich, 99.8 %) containing 9 mmol (high loading sample) of the 3aminopropyltriethoxysilane (APTES, Aldrich, >98%) for 24 h in a nitrogen atmosphere.
The resulting material was purified by Soxhlet extraction with dry toluene, to remove the unreacted amino-organosilanes, and finally dried under vacuum, at 120 °C for 24 h. The low amine loading sample was synthesized according to the literature 2 . Firstly, 0.35 g of SBA-15 was dried in the vacuum line by heating at 140 °C for 8 h. Next, SBA-15 was suspended in the mixture of 3.5 cm 3 of anhydrous toluene and 65 cm 3 of (3triethoxysilylpropyl)-tert-butylcarbamate (TESPtBC, Gelest). The suspension was heated at 80 °C, under reflux, for 12 h. Lastly, the material was filtered out and washed with toluene. To remove the organic part that protects the amine, the sample was placed in a zirconia NMR rotor, degassed and heated for 6 h at 250 °C in our sorption apparatus (fully described below).

Sorption Apparatus
The sorption apparatus comprised a laboratory made high vacuum line, connected to a turbomolecular pumping station (HiCube 80, Pfeiffer Vacuum), capable of vacuum greater than 10 −2 Pa. A borosilicate glass cell, adapted from the description in the literature, was connected to the vacuum line and served as an enclosure for an NMR rotor allowing to degas and heat zirconia NMR rotors up to 300 °C under vacuum. The heating was performed with a laboratory made oven connected to a power controller (

Computational details
The clusters used to model the silica surface are based on the experimental crystallographic structure of alpha-quartz 3 using the atomic positions of the silicon and oxygen atoms. Dangling bonds at the edges of the cluster models due to elimination of Si atoms were saturated with H atoms along the O−Si directions of the perfect crystal and imposing an O−H distance equal to 0.96 Å 4 . Note that although the real surface of mesoporous silicas is amorphous, using a surface built from a crystalline structure should not dramatically influence the results from the calculations, as we are dealing with relatively small clusters.
The silylpropylamines were grafted (through optimisation) on the clusters where OH groups existed, each binding three surface OH groups. Subsequent optimisations of different species involved the relaxation of the alkyl chain (and the respective functional group at its end), water or CO2 molecules (when present), the SiO3 moieties binding the alkylamines, and the surface OH groups, while the remaining Si and O atoms were kept frozen at their crystallographic positions. The fixation of some atomic S5 positions provides a simple but effective representation of the mechanical embedding of the solid covalent oxide surface [5][6] . The absence of imaginary values in the frequencies involving the atoms optimised in the different structural models ensured that the structures are true minima on their potential energy surfaces. However, to model some specific structures, it was necessary to fix specific structural parameters at no optimal values which led to vibrational modes with associated imaginary wavenumbers.
Three different sized clusters ( Figure S1) were used in this work; a small cluster, with a single amine chain and a single surface silanol, a medium cluster, with two amine chains and a surface silanol, and a large cluster, with one amine chain and five surface silanols. The cluster models are based on six (small), nine (medium) and fourteen (large) Si atoms and, following the notation used by López et al. 5 will be named as N-T models (N=6, 9 or 14, respectively). The tests in our previous study 7 indicate that their sizes are adequate to prevent spurious border effects. To model a very specific case of an isolated amine, not able to form hydrogen bonds with the silica framework, we have also employed a (CH3)3SiCH2CH2CH2NH2 model, where terminating OH moieties were replaced by non-interacting methyl groups.
The M06-2X hybrid functional based on the meta-generalized gradient approximation of Truhlar and Zhao [8][9] and the standard 6-31G(d) basis set [10][11] , with a single polarisation function in all the atoms except hydrogen, as included in the Gaussian 09 software 12 , were used in all the structural optimisations, in the calculation of electronic energies or Gibbs energies at T=298.15 K, and vibrational frequencies. The combination of the M06 family of functionals and Gaussian-type orbitals basis sets with cluster models was found to provide geometries, energies, and frequencies in very good agreement with available experimental data for several systems 13-16 that are challenging for DFT approaches from the lower rungs of the Jacob's ladder of density functional approximations 17 . The M06-2X functional has been suggested for geometry optimization in a guide to small-molecule structure assignment through computation of some NMR chemical shifts 18 , so it has been the default choice in our work with amine-functionalised silicas 7,19 . In all calculations, the default integration grids and convergence thresholds in the Gaussian 09 software were employed 12 . As common practice, the calculated wavenumbers were multiplied by a scale factor; the value used (0.947) was optimised for the M06-2X/6-31G(d) approach and was taken from ref. 20 .
NMR shielding tensors of the optimized geometries have been computed with the GIAO method [21][22] , also using the M06-2X functional and the 6-31G(d) basis set.
These conditions typically create relatively small root-mean-square errors of the calculated 13 C chemical shifts (cf. 3.2 ppm) 23 . No such reference result exists for 15 N, computed at the M06-2X/6-31G(d) level of theory, so we calculated it ourselves for a set of experimental chemical shifts of aqueous alkylamines 24 , having obtained a rootmean-square error of 4.9 ppm using an implicit solvent method, the polarisable continuum model (PCM) 25 (Table S25). The isotropic magnetic shielding tensors calculated for the clusters were subtracted from those calculated for a reference, to obtain the chemical shift relative to such reference. The reference for 13 C chemical shifts was gas-phase tetramethylsilane (at 0 ppm, as in our previous work 7, 26   1.5. NMR measurements 13 C NMR spectra were acquired at room temperature on a Bruker Avance III 700 spectrometer operating at B0 field of 16.4 T with 13 C frequency of 176.1 MHz. All experiments were performed on a double-resonance 4 mm probe and samples were packed into ZrO2 rotors with Kel-F (4 mm) caps. Spinning rate of 12 kHz was employed to record all spectra. 13 C chemical shifts are quoted in ppm from TMS (0 ppm) and αglycine (secondary reference, C=O at 176.03 ppm). 13 C CPMAS spectra were acquired under the following experimental conditions: 1 H 90° pulse was set to 3.0 µs corresponding to a radio-frequency of 83 kHz; the CP step was performed with a contact time of 2000 µs using a 50−100% RAMP shape on the 1 H channel and a radiofrequency of 77 kHz and a 50 kHz square shape pulse was used on the 13 C channel.
Recycle delay was 4 s for all measurements. During the acquisition, a SPINAL-64 decoupling scheme was employed using a pulse length for the basic decoupling units of 5.6 µs at rf field strength of 83 kHz. Table S1. Calculated 15 N chemical shifts for different amine and ammonium species (based on Tables S2-S4) together with the experimental values from Chen et al. 28 .

Chemical shifts and lowest energy models for amine/ammonium species
Corresponding structures shown in Figure S2, using the same labels.

Optimised structural models
3.1. Amine Table S2. Results of optimisation trials of the aminesilanol cluster. Energy differences to the most stable structure (10) and corresponding 15

Single-tethered amine and ammonium
Compared structures correspond to Structure xvii of Table S2 for the amine (10, Figure   S2), and Structure xvi of Table S3 for the ammonium (11, Figure S2). Table S15. Comparison of Gibbs energy of formation of singletethered amine (10) and ammonium (11).

Carbamic acid and ammonium carbamate
Compared structures correspond to Structure xxi of Table S6 for ammonium carbamate (1, Figure 1), and Structure xxv of Table S7 for carbamic acid (2, Figure 1). representing carbamic acid and adsorbed CO2. However, silylpropylcarbamate has one less oxygen atom and two less hydrogen atoms. Therefore, a single water molecule was added to the model (and then optimised) to allow energetic comparisons with adsorbed CO2 and carbamic acid.
We have chosen to not present the optimisation trials tables for the structures of silylpropylcarbamate and amine with adsorbed CO2 but three-dimensional representations of the most stable structures are shown in Figure S3.  Figure S3), amine with adsorbed CO2 (14, Figure S3) and carbamic acid (4, Figure 1). It was necessary to add a water molecule to the silylpropylcarbamate cluster, so a comparison could be made with the other two structures (i.e., for balancing the number of atoms).

Species ∆G (kJ×mol -1 )
Silylpropylcarbamate 30 Adsorbed CO2   As in the previous section, we have chosen to not present the full optimisation tables for each species. However, three-dimensional representations of the most stable structures are shown in Figure S4.  Figure S4), bicarbonate (16, Figure S4), and carbamic acid (17, Figure S4). It was necessary to add a water molecule to the carbamic acid and adsorbed CO2 clusters, so a comparison could be made with bicarbonate (i.e., for balancing the number of atoms).