Article | Published:

# Direct air capture of CO2 via aqueous-phase absorption and crystalline-phase release using concentrated solar power

## Abstract

Using negative emissions technologies for the net removal of greenhouse gases from the atmosphere could provide a pathway to limit global temperature rises. Direct air capture of carbon dioxide offers the prospect of permanently lowering the atmospheric CO2 concentration, providing that economical and energy-efficient technologies can be developed and deployed on a large scale. Here, we report an approach to direct air capture, at the laboratory scale, using mostly off-the-shelf materials and equipment. First, CO2 absorption is achieved with readily available and environmentally friendly aqueous amino acid solutions (glycine and sarcosine) using a household humidifier. The CO2-loaded solutions are then reacted with a simple guanidine compound, which crystallizes as a very insoluble carbonate salt and regenerates the amino acid sorbent. Finally, effective CO2 release and near-quantitative regeneration of the guanidine compound are achieved by relatively mild heating of the carbonate crystals using concentrated solar power.

## Main

Sustained burning of fossil fuels over the past one and a half centuries led to an increase in the atmospheric CO2 concentration by more than 45%, from about 280 to more than 406 ppm1. As a result, the global temperature now exceeds +1 °C relative to the pre-industrial era2. Recent studies indicate that limiting global warming below 2 °C by the end of this century will probably require large-scale deployment of ‘negative emissions’ technologies (that is, removal of CO2 from ambient air)3,4. Indeed, as the Earth is out of energy balance with the current atmospheric composition, more warming is in the pipeline even if fossil-fuel emissions were to suddenly stop5. Furthermore, it has recently been suggested that ideally the atmospheric CO2 concentration should be below 350 ppm, to bring the global temperature back within the optimum range of the pre-industrial Holocene period3. Meeting this ambitious goal would require the removal of at least 550 GtCO2 (550 billion tons of CO2) out of the atmosphere by the end of this century3.

Direct air capture (DAC) of carbon dioxide from ambient air by engineered chemical reactions6,7,8 represents a distinct category of negative emissions technologies among other more ‘natural’ approaches to negative emissions, including bioenergy with carbon capture and storage, afforestation and reforestation, and enhanced weathering of minerals9. DAC has the advantage of relatively low land and water requirements, and it was estimated that it could remove up to 12 GtCO2 per year10. However, living up to such great expectations requires sustained research efforts over the next few decades to improve the existing DAC technologies or develop completely new ones that are economical and can be deployed on a large scale.

The low concentration of CO2 in the air (~400 ppm) and the inherently open nature of the DAC process impose some constraints on the type of sorbents that can be used. First, the CO2 binding has to be relatively strong and selective against other atmospheric components (especially water), which disqualifies most physisorbent materials. Amine sorbents, the workhorse of industrial CO2 scrubbing, bind CO2 strongly, but their volatility and toxicity are incompatible with large-scale deployment in open spaces. To date, there are two classes of sorbents that have been extensively investigated for DAC applications: aqueous alkaline sorbents (that is, NaOH, KOH and Ca(OH)2)11,12,13,14 and porous solid-supported amines15,16,17,18,19. The aqueous alkaline sorbents have the advantage of ready availability and relatively fast sorption kinetics, but are highly corrosive and the sorbent regeneration is energy intensive, requiring very high temperatures of ~900 °C. Solid-supported amines have lower regeneration energies and temperatures but tend to have slower sorption kinetics and their optimum performance requires the maintenance of a high surface area over multiple cycles and preventing water condensation in the pores20,21. They also tend to chemically and thermally degrade over time, especially when heated in open air. Anion-exchange resins acting as moisture-swing CO2 absorbents have also been employed for DAC, though the partial pressure of the released CO2 is relatively low, requiring additional concentration steps before storage22,23. Thus, developing new concepts and materials that combine the best attributes of liquid and solid sorbents, and are energy-efficient and cost-effective, remains a high priority in DAC research.

Here, we report a proof-of-concept small-scale DAC system involving aqueous solutions of glycine and sarcosine amino acids as non-volatile, environmentally friendly and readily available sorbents, and employing a household humidifier as a simple air–liquid contactor. The CO2-loaded amino acids were subsequently reacted with a guanidine compound24, leading to crystallization of a guanidinium carbonate salt of very low aqueous solubility (on par with CaCO3) and regeneration of the amino acid sorbent. The CO2 was released from the isolated carbonate crystals by relatively mild heating using concentrated solar power, which regenerated the guanidine compound in near-quantitative yield.

## DAC system design

Preliminary results indicated that aqueous 2,6-pyridine-bis(iminoguanidine) (PyBIG) captures CO2 from ambient air and binds it as a crystalline tetrahydrated carbonate salt (PyBIGH2(CO3)(H2O)4). The CO2 can be released by heating the carbonate crystals at relatively mild temperatures of 80–120 °C, which regenerates PyBIG quantitatively (Fig. 1a)24.

The elementary steps involved in the CO2 absorption and the overall reaction are represented by equations (1)–(7) in Fig. 2. First, PyBIG dissolves into water (equation (1)), and then the two guanidine groups become protonated by water molecules, generating the dicationic form of the compound (PyBIGH22+) and HO (equation (2)). The hydroxide anions then react with the CO2 absorbed from air (equation (3)) and generate bicarbonate (equation (4)) and carbonate anions (equation (5)). Finally, the PyBIGH22+ and CO32 ions crystallize with water into crystalline PyBIGH2(CO3)(H2O)4 (equation (6)). The net reaction, shown in equation (7), corresponds to crystalline PyBIG converting into crystalline PyBIGH2(CO3)(H2O)4, in the presence of CO2 and water, through dissolution/recrystallization.

Though PyBIG can capture CO2 from ambient air according to equation (7), the reaction is too slow for practical considerations. The kinetics of CO2 absorption from air by PyBIG is limited by a number of factors. First, the atmospheric CO2 concentration is very low (~0.04%). For aqueous alkaline solutions, such as NaOH, the CO2 absorption rate has been found to be limited by a combination of the CO2 diffusion into the aqueous solution (equation (3)), and the reaction of CO2 with HO (equation (4))11,14. Thus, the rate of CO2 absorption is controlled by the surface area of the air–liquid interface and the solution alkalinity. The relatively low aqueous solubility of PyBIG (~10 mM) limits the solution alkalinity (pH ≈ 10) and therefore the reaction rate with CO2. Another constraint is that, in a typical crystallization set-up, the air–liquid contact area is relatively small, which further limits the CO2 absorption rate.

One possible solution to the slow CO2 sorption problem is to combine the PyBIG crystallization with a traditional aqueous sorbent that absorbs atmospheric CO2 relatively fast and converts it into carbonate. The carbonate-loaded solution is subsequently reacted with PyBIG to crystallize PyBIGH2(CO3)(H2O)4 and regenerate the sorbent. Finally, the carbonate crystals are filtered out of solution and heated in the solid state to release the CO2 and regenerate the PyBIG compound, which can then be reused in another cycle (Fig. 1b). The advantage of such a hybrid approach to DAC, which combines room-temperature absorption in the liquid phase with CO2 release in the solid state, is that it benefits from the fast sorption kinetics of an aqueous sorbent while avoiding the energy penalty associated with heating aqueous solutions during regeneration. Furthermore, sorbent loss through evaporation and thermal degradation is minimized.

After considering several potential sorbents, we selected two simple amino acids, glycine and sarcosine, for our DAC system. Aqueous amino acids have a number of positive attributes that make them promising candidates for DAC. They have fast reaction rates with CO2 (refs 25,26), on par with or surpassing more traditional sorbents such as monoethanolamine or NaOH. Amino acids are non-volatile, non-corrosive, environmentally friendly and relatively inexpensive. They are also less susceptible to oxidation than amines. While amino acid sorbents have been employed in CO2 scrubbing from flue gas27, their deployment in DAC remains largely unexplored.

The chemical reactions involved in the CO2 absorption with amino acids and in the sorbent regeneration with PyBIG are depicted in equations (8)–(10) in Fig. 2, using glycine as a representative example. First, the anionic form of glycine (glycinate) reacts with CO2 and generates the corresponding carbamic acid, which is deprotonated by a second equivalent of glycine to generate the carbamate and the zwiterionic glycine (equation (8)). The carbamate is subsequently hydrolysed to glycinate and bicarbonate (equation (9)). Finally, PyBIG grabs two protons and a CO32 anion from solution and crystallizes as PyBIGH2(CO3)(H2O)4 (equation (10)), which in the process leads to deprotonation of the zwiterionic glycine and regeneration of the glycinate (nominally the two protons are abstracted from HCO3 and the glycine zwitterion). Note that adding the three reactions together leads to the same overall reaction represented by equation (7). However, the kinetics of the amino-acid-mediated DAC process are expected to be significantly faster than using PyBIG alone.

In addition to fast kinetics for the CO2 reaction with the sorbent, a practical DAC system requires effective mass transfer of CO2 from air into the sorbent solution, which in turn requires an efficient contactor that maximizes the air–liquid interfacial area. Unlike CO2 capture from flue gas, which is typically performed in a pressurized absorption column that is designed to operate at high liquid/gas ratios with a high degree of CO2 removal, DAC is more suitably performed in an open system with contactors that are optimized to ingest large volumes of ambient air, in many ways similar to large-scale cooling towers14. For the purpose of this study, we opted for a household humidifier as our air–liquid contactor (Supplementary Fig. 1), replacing the water with amino acid solutions.

## Thermodynamic analysis

The thermodynamics of the CO2 absorption and desorption, as well as of sorbent regeneration, were determined in the next step to define the energy and efficiency boundaries for our DAC system. The enthalpy of the CO2 absorption (equation (7)) can be obtained by adding up the enthalpies of the elementary reactions represented by equations (1)–(6). The enthalpies of CO2 hydration (equation (3)), of reaction with HO to generate bicarbonate (equation (4)) and of bicarbonate deprotonation by HO to generate CO32 (equation (5)) were obtained from the literature28,29,30,31. The remaining enthalpies for the reactions involving PyBIG and PyBIGH2(CO3)(H2O)4 were determined as part of this study. The enthalpies of PyBIG dissolution (equation (1)) and PyBIGH2(CO3)(H2O)4 crystallization (equation (6)) were obtained by variable-temperature solubility measurements of the two solids and van’t Hoff analyses (Supplementary Tables 1 and 2 and Supplementary Figs. 2 and 3). Finally, the enthalpies of PyBIG protonation (equation (2)) were obtained from variable-temperature pKa measurements by potentiometric titrations and van’t Hoff analyses (Supplementary Table 3 and Supplementary Fig. 4). The corresponding ΔH values for reactions (1)–(6) are listed in Table 1. Adding up these values results in an overall enthalpy of CO2 absorption by PyBIG of −70.7 kJ mol−1. By comparison, amine sorbents have CO2 absorption enthalpies in the range of 60–80 kJ mol−1. However, unlike primary or secondary amines, PyBIG does not bind directly to CO2. Instead, the actual CO2 binding is performed by the hydroxide ions to generate carbonate ions (equations (3)–(5)), with a combined enthalpy of −109.8 kJ mol1. This is essentially the same as the enthalpy of CO2 absorption by aqueous NaOH (ref. 11). The crystallization step (equation (6)) then adds −47 kJ mol−1, to make the overall absorption process a combined −156 kJ mol−1, which is about twice as exothermic compared with the CO2 absorption by a typical amine sorbent. However, dissolution of PyBIG (equation (1)), which is endothermic (+42.5 kJ mol−1) and essentially offsets the exothermic crystallization in step (6), and proton transfer from water to the guanidine groups (equation (2)), which is also endothermic (+43.6), bring the overall enthalpy for CO2 capture by PyBIG (equation (7)) to −70.7 kJ mol−1.

Once the CO2 loading is complete, the crystalline PyBIGH2(CO3)(H2O)4 is removed from solution and heated in the solid state to release the CO2 gas and the water vapours. Given the completely different conditions involved, such a solid–gas–solid process must have different energetics compared to the gas–liquid–solid process involved in the CO2 absorption. To determine the enthalpy of CO2 release from PyBIGH2(CO3)(H2O)4, we employed differential scanning calorimetry (DSC). The obtained DSC curve (Supplementary Fig. 5) shows a series of endothermic events between 80 and 140 °C, corresponding to the release of water and CO2 as previously found by thermogravimetric analysis (TGA)24. Unfortunately, the extensive overlap between the peaks prevented us from obtaining the heats associated with each thermal event. Instead, we integrated all of the peaks together to obtain the overall enthalpy of 223 ± 4 kJ mol1. While the measured enthalpy of reaction is highly endothermic, it must be taken into account that, for each CO2 molecule released from PyBIGH2(CO3)(H2O)4, there are four water molecules of hydration that are desorbed (plus a fifth water molecule that is a by-product of the carbonate decomposition). To determine the fraction of the reaction enthalpy corresponding to the CO2 release from the carbonate, we prepared the anhydrous PyBIGH2(CO3) by overnight vacuum-pumping of PyBIGH2(CO3)(H2O)4 at 30 °C. The enthalpy of the CO2 release from the anhydrous carbonate salt, measured by DSC (Supplementary Fig. 5), is 75 ± 6 kJ mol−1. The difference between the measured endotherms of PyBIGH2(CO3)(H2O)4 and PyBIGH2(CO3), of 148 kJ mol−1, can be assigned to dehydration of the carbonate salt. This corresponds to an average of 37 kJ mol1 per water molecule of hydration, which is close to the enthalpy of vaporization for liquid water, of 40.65 kJ mol1. While a large fraction of the heat required to release the CO2 is spent on water evaporation, part of the waste heat could be recovered by process modifications that involve water condensation steps in which the resulting heat is collected and reused via heat exchangers32. Alternatively, vacuum-drying operations that typically require less energy than heat drying33 can be employed for dehydration of PyBIGH2(CO3)(H2O)4 before CO2 release. We are also considering the use of renewable sources of energy, such as concentrated solar power (see below), to increase the sustainability of the overall DAC process.

In addition to the enthalpies of CO2 absorption and release, another important thermodynamic parameter is the equilibrium constant for the amino acid regeneration reaction (equation (10)), which defines the efficiency limit for sorbent regeneration. In the regeneration reaction, the solid PyBIG has to dissolve into the sorbent solution, deprotonate the amino acid and the bicarbonate ion, and crystallize as PyBIGH2(CO3)(H2O)4. In the case of the glycine sorbent, the equilibrium constant for the regeneration reaction (logKreg) is defined by equation (11):

$$\begin{array}{c}\\ {\rm{log}}{K}_{{\rm{reg}}}=\;{\rm{log}}{K}_{{\rm{sp}}}\left({\rm{PyBIG}}\right)-{\rm{log}}{K}_{{\rm{sp}}}\left({{\rm{PyBIGH}}}_{2}\left({{\rm{CO}}}_{3}\right){\left({{\rm{H}}}_{{\rm{2}}}{\rm{O}}\right)}_{4}\right)\\ -{\rm{p}}{K}_{{\rm{a}}}\left({\rm{Gly}}\right)-{\rm{p}}{K}_{{\rm{a}}}\left({{\rm{HCO}}}_{{\rm{3}}}^{-}\right)+{\rm{p}}{K}_{{\rm{a1}}}\left({\rm{PyBIG}}\right)+{\rm{p}}{K}_{{\rm{a2}}}\left({\rm{PyBIG}}\right)\end{array}$$
(11)

Thus, the amino acid regeneration is driven by the difference in solubility between PyBIG and PyBIGH2(CO3)(H2O)4, as well as by the difference in basicity between glycine and bicarbonate on one hand, and PyBIG on the other hand. The acid dissociation constant (pKa) values at 25 °C for the two guanidinium groups of PyBIG, determined by potentiometric titration (Supplementary Table 3), are 7.6 and 8.7. On the other hand, the pKa values of glycine and bicarbonate are 9.5 and 10.3, respectively34. Thus, PyBIG is not sufficiently basic to drive the regeneration equilibrium to the right, and therefore the main driving force has to come from the solubility difference between PyBIG and PyBIGH2(CO3)(H2O)4. We have previously reported a preliminary estimated value of 1.0 × 10−8 for the solubility product (Ksp) of PyBIGH2(CO3)(H2O)4 (ref. 24). As part of this study, determination of a more accurate value became possible as the exact speciation of PyBIG in solution could now be obtained on the basis of the measured pKa values of PyBIG, resulting in a revised Ksp for PyBIGH2(CO3)(H2O)4 of 1.0 ± 0.3 × 10−9 at 25 °C. This corresponds to a very insoluble carbonate salt, on par with calcite mineral (Ksp = 3.3 × 10−9). By comparison, the measured Ksp of PyBIG at the same temperature is 1.0 ± 0.3 × 10−2. Thus, under ideal conditions, logKreg = 3.5, which predicts a very efficient amino acid regeneration with PyBIG. However, under realistic conditions involving high-ionic-strength solutions that can significantly impact the solubilities of the various species involved through ion pairing, salting out and so on, the observed regeneration efficiency may actually be less than ideal.

## Direct air capture with aqueous glycine and sarcosine

The loading of CO2 with the humidifier using 1 M aqueous solutions of potassium glycinate and sarcosinate is shown in Fig. 3. The extent of CO2 absorption as a function of time was monitored in situ by pH measurements, and ex situ by ion chromatography (IC) and 1H NMR (Supplementary Figs. 6 and 7) to determine the amounts of carbonate and carbamate formed. The sorption experiments were run for 24 h and the results are summarized in Table 2. The final CO2 loading for glycine (0.76 ± 0.04 mol mol−1) was slightly higher than for sarcosine (0.69 ± 0.05 mol mol−1). While the reaction rates of CO2 with both sarcosine and glycine are expected to be high25,26, the relatively long times required to reach CO2 saturation with both sarcosine and glycine sorbents are due to the very low concentration of CO2 in air and the relatively small air–liquid interfacial area relative to the volume of the sorbent in the humidifier.

## Sorbent regeneration

The CO2-loaded sorbents were stirred with a suspension of PyBIG (0.5 moles per moles of CO2 absorbed) at room temperature, which resulted in crystallization of PyBIGH2(CO3)(H2O)4 and regeneration of the anionic forms of the amino acids, according to equation (10). The formation of crystalline PyBIGH2(CO3)(H2O)4 was confirmed by powder X-ray diffraction (PXRD), which revealed, after 24 h, a mixture of PyBIG and its carbonate salt (Supplementary Figs. 8 and 9). The concentrations of carbonate and carbamate in the amino acid solutions were monitored by IC and NMR, respectively, and the total amount of CO2 removed as a function of time is plotted in Fig. 4a. More than 90% of the total amount of CO2 removed was released from the sorbent within the first hour, and longer regeneration times led to only marginal improvements.

The average swing capacity, defined as the difference between the maximum CO2 loading observed after 24 h of absorption and the minimum CO2 loading measured after 24 h of regeneration (based on three separate runs), was 0.28 ± 0.06 and 0.31 ± 0.05 mol mol−1 for the glycine and sarcosine sorbents, respectively. The swing capacity remained relatively constant over three consecutive loading/regeneration cycles (Fig. 4b). However, many more cycles will have to be run to test the long-term robustness of this DAC system. Though the room-temperature crystallization-based sorbent regeneration employed here circumvents the heating of the amino acid sorbent, thereby extending its operational lifetime, the long-term stability of PyBIG that must be thermally regenerated remains a concern, as its degradation over time would lead to a gradual decrease in the cyclic capacity. To test the thermal stability of the guanidine compound, PyBIG was subjected to accelerated ageing by heating the carbonate crystals at 120 °C in open air for a week, which is the equivalent of 336 thermal regeneration cycles (based on a typical 30 min regeneration time). The TGA showed an initial weight loss of 35.3% corresponding to the removal of the CO2 and H2O, followed by 0.5% weight loss over the course of a week (Supplementary Fig. 10). The ‘aged’ PyBIG was subsequently used in a sarcosine regeneration cycle and the measured swing capacity of 0.26 ± 0.02 mol mol−1 was very similar to the corresponding values measured with fresh PyBIG (Fig. 4b), attesting to the thermal stability of the guanidine compound.

The regeneration of the amino acid sorbents was also performed conventionally by boiling the aqueous solutions under reflux. After 1 h of refluxing, the amounts of CO2 removed from the glycine and sarcosine sorbents were 0.31 and 0.30 mol mol−1, respectively, which are comparable with the swing capacities obtained with PyBIG. With longer refluxing times (4 h), the measured swing capacities were 0.47 and 0.45 mol mol−1 for glycine and sarcosine, respectively. However, such long refluxing times are expected to come at a cost in terms of energy consumption and sorbent degradation.

## CO2 release with concentrated solar power

Our preliminary study had indicated that the PyBIGH2(CO3)(H2O)4 crystals release the CO2 and the water on mild heating at temperatures of 80–120 °C, and regenerate the PyBIG compound quantitatively24. However, considering that this transformation is highly endothermic, we decided to explore the possibility of using concentrated solar power to increase the sustainability of the DAC process. For the initial small-scale proof of concept, we employed a solar oven (Supplementary Fig. 11) to heat the crystalline PyBIGH2(CO3)(H2O)4 samples (~38 mg, 0.1 mmol) at five different temperatures ranging between 120 and 160 °C. The extent of the reactions was determined by the samples’ weight loss (Table 3). Fourier-transform infrared (FTIR) analyses corroborated the release of CO2, most evidently noticeable by the disappearance of the strong peak at 1,361 cm1 corresponding to the stretching mode of the carbonate anion (Supplementary Fig. 12). For benchmarking, under similar conditions (2 h of heating at 160 °C), no CO2 loss was observed from CaCO3, consistent with the much higher temperatures (>800 °C) typically required to desorb CO2 from metal carbonates35.

These results demonstrate that concentrated solar power can be effectively used for quick release of the CO2 from the carbonate crystals and regeneration of the PyBIG compound in near-quantitative (92–96%) yield. The regeneration of PyBIG requires relatively low temperatures that are easily attainable with simple parabolic-trough solar concentrators, in direct contrast to traditional alkaline sorbents that would require more elaborate solar reactor designs to achieve the much higher temperatures needed for regeneration35. Furthermore, solid-state regeneration of PyBIG is fast in comparison to aqueous sorbents, for which a substantial amount of energy needs to be expanded for heating and evaporating water. For example, under the same conditions (30 min, 120 °C), the regeneration yield of an aqueous K2CO3 sorbent was only 15% (Supplementary Methods), compared to the 92% regeneration yield for PyBIG.

## Discussion

In this proof-of-concept study, we demonstrated a laboratory-scale DAC system using simple, off-the-shelf equipment and readily available chemicals. This hybrid DAC approach combines the benefits of an aqueous sorbent, such as relatively fast CO2 absorption rates, easy handling and low maintenance, with the advantages of solid-state CO2 desorption, such as lower energy penalty, and minimal sorbent evaporation and degradation. Furthermore, the amino acid sorbents offer an environmentally friendly alternative to the toxic and corrosive amines and NaOH sorbents.

One important issue that requires future consideration is how to scale up the carbonate crystallization and handle large volumes of solid. Along this line, much can be learned from the CO2-capture technologies involving phase-changing amino acids (DECAB Plus)36 or aminosilicone sorbents37. In these processes, the solids are removed from solution using solid–liquid separators, such as cyclones or decanters, and then transported with the help of extruders in a continuous-feed mode. The effectiveness of such solid-handling processes depends to a large extent on a number of crystallization parameters, such as kinetics of crystallization, particle size distribution and crystal habit, which need to be optimized as part of the technology maturation.

Besides demonstrating the initial proof-of-concept, this study also identified a number of limitations for the current DAC system and provided guidelines for the design and optimization of future DAC technologies. First, while the rate constants for the reactions of the amino acids with CO2 are fast, the air humidifier used in this study is not optimized for DAC, as it provides a small air–liquid interfacial area relative to the volume of the sorbent, which limits the overall CO2 uptake rate. Moreover, as designed, the humidifier evaporates large amounts of water, which, in the case of DAC, is a disadvantage. We anticipate that combining the amino acid sorbents with better air–liquid contactors that optimize the interfacial area and minimize the water loss will lead to more efficient DAC systems. Second, while PyBIG regenerates the amino acid sorbents relatively well, with observed cyclic capacities of about 0.3 mol mol−1, the process would be significantly improved by replacing the PyBIG compound with a more soluble and alkaline analogue that could shift the regeneration equilibrium further to the right. The kinetics of crystallization and sorbent regeneration are also critical in achieving an efficient closed-cycle process that continuously circulates the sorbent between the sorption and regeneration units. Further work is needed to accurately measure the kinetics of carbonate crystallization and the influence of various parameters, such as temperature, particle size and stirring method on the rate of sorbent regeneration. Finally, though the CO2 desorption from crystalline PyBIGH2(CO3)(H2O)4 avoids much of the energy penalty associated with heating and evaporating aqueous solutions, the CO2 release is strongly endothermic, mainly due to the inclusion of water in the crystals. Engineering carbonate crystals that are anhydrous, or can release the CO2 at lower temperatures to avoid water vaporization, would greatly improve the energy efficiency of the DAC process. On the other hand, employing renewable energy sources, such as concentrated solar power, as demonstrated in this study, or low-grade waste heat, can alleviate much of this issue.

## Methods

### Materials and characterization

Common reagents, including glycine, sarcosine, KOH, NaOH and HCl, were purchased from commercial suppliers and used without further purification. All water used was deionized (18 mΩ). FTIR spectra were collected using a PerkinElmer Frontier FTIR spectrometer with a Universal ATR Sampling Accessory attachment. UV–vis spectra were measured in 10-mm-path-length quartz cuvettes using a Cary Varian 5000 spectrometer. NMR spectra were collected on a Bruker Avance III 400 spectrometer. PXRD measurements were performed with a Panalytical Empyrean diffractometer using Cu Κα radiation (λ = 1.5418 Å). pH measurements were conducted with a Thermo Scientific Orion Star A211 pH meter using an Orion 9107 APMD pH electrode for the CO2-loading experiments, or an Orion 8156 BNUWP pH electrode coupled with a Thermo Scientific ATC probe for the titrations. Ion chromatography measurements were performed on a Dionex ICS-5000+ Ion Chromatography System with an inline eluent generator. The column used was a Dionex Ion Pac AS11-SC IC with a 4 mm internal diameter and a 250 mm length. A Bore guard column of the same diameter but 50 mm long was placed before it as a sacrificial column. TGA was conducted under an ambient atmosphere using a TA Q5000 IR Instrument. DSC measurements were conducted under nitrogen on a TA Instruments Q20 in a temperature range of 25–200 °C and temperature ramp of 10 °C min−1. Data analysis was performed using the Advantage v5.4.0 (TA Instruments) software. The PyBIG compound was synthesized from pyridine-2,6-dicarbaldehyde and aminoguanidinium chloride, according to our previously reported method24, which was scaled up and optimized (see Supplementary Methods).

### Determination of PyBIG pKa values by potentiometric titrations

The variable-temperature titrations of PyBIG were carried out in a jacketed vessel maintained at the desired temperature using a circulating bath. The electrode was calibrated before use by a standard titration of NaOH (0.01 M) with HCl (0.01 M). In a typical experiment, a 50 ml solution containing PyBIG·2HCl (5 mM), HCl (6 mM) and NaCl background electrolyte (0.2 M) was titrated with a standard 0.1 M NaOH solution using a micropipette. The potential readings were recorded 7 min after each NaOH addition to allow the solution to equilibrate. For each temperature, the titrations were run in duplicate, and the reported pKa values are the average of the two measurements. The titration data were fitted using the Hyperquad 2013 program.

### Solubility measurements

The solubilities of PyBIG and PyBIGH2(CO3)(H2O)4 were determined by measuring the UV–vis absorption spectra of the corresponding saturated solutions as previously described for a similar bis-iminoguanidinium compound38. Saturated solutions were prepared by suspending an excess of the crystalline solids in 10 ml of H2O inside 15 ml polypropylene centrifuge tubes, and mixing the suspensions on a rotating wheel for 72 h at 60 r.p.m. inside an incubator set at different temperatures in the range of 10 to 35 °C. All measurements were run in triplicate, and the average solubility values with the corresponding standard deviations are reported in Supplementary Tables 1 and 2.

### Determination of solubility product of PyBIGH2(CO3)(H2O)4

The solubility product of PyBIGH2(CO3)(H2O)4 at 25 °C was determined using equation (12), where the activity coefficients (γ±) were estimated at 0.74 using the Debye–Huckel limiting law. The concentration of the PyBIGH22+ cation was determined taking into account the measured solubility of PyBIGH2(CO3)(H2O)4 of 0.0012 M, the pKa values of PyBIG (7.62 and 8.69) and the pH of the saturated solution of 8.35. Based on these parameters, the speciation of the ligand is: PyBIG 29%, PyBIGH+ 60% and PyBIGH22+ 11%. As a result, [PyBIGH22+] = 0.13 × 103. The concentration of the carbonate anion was determined to be 1.4 ± 0.2 × 105 M considering the pKa of HCO3 of 10.32 and the pH of the saturated solution of 8.35.

$$\begin{array}{c}\\ {K}_{{\rm{sp}}}{\rm{=}}{\left({{\rm{\gamma }}}_{\pm }\right)}^{2}\left[{{\rm{PyBIGH}}}_{2}^{2+}\right]\left[{{\rm{CO}}}_{3}^{2-}\right]\\ \\ {\rm{=}}{\left(0.74\right)}^{2}[0.13\times 1{0}^{-3}][1.4\times 1{0}^{-5}]\\ =1.0\pm 0.4\times 1{0}^{-9}\end{array}$$
(12)

### DAC of CO2 with glycine and sarcosine

The CO2 absorption from air was carried out with an Envion Humidiheat household air humidifier (Supplementary Fig. 1). The humidifier consists of a fan, a reservoir with a capacity of about 2 l and a rotating wick made of a synthetic fabric that spreads the aqueous solution into a thin film for better contact with the air. The reservoir was filled with 1.5 l aqueous solutions of glycine or sarcosine (1 M) and KOH (1 M) and the fan was run on a slow setting, corresponding to an air flow rate of 3.8 ± 0.2 m s−1 (measured with a Hold Peak HP-866B anemometer). The change in the solution pH was monitored in situ with a glass electrode. The CO2 capture experiments were run at ambient temperature (21 ± 1 °C). However, because of the water evaporation, the solution temperature was lower, averaging 16 ± 1 °C. To compensate for the evaporated water, the reservoir was replenished continuously with H2O using a mini-pump set at a flow rate of 2.2 ml min−1. The amino acid concentration was monitored by periodically withdrawing 300 µl samples and analysing them by 1H NMR spectroscopy using maleic acid as an internal standard. The amino acid concentration was found to remain relatively constant within 1.00 ± 0.05 M. The amount of CO2 absorbed was monitored by withdrawing 300 µl samples and analysing their carbonate and carbamate content by IC and 1H NMR spectroscopy, respectively. For NMR analyses, 900 µl of D2O was added to 100 µl of the samples, whereas for the IC analyses the samples were diluted 10–300-fold to bring the carbonate concentration to the 30–300 ppm range. The sorption experiments were run three times, and the average and standard deviation values for CO2 loading are reported in Table 2.

### Regeneration of the amino acid sorbents with PyBIG

All regenerations were carried out at room temperature. The amino acid solutions (5 ml) were placed in 20 ml vials and PyBIG·2.5H2O was added as a solid. The amount of PyBIG added varied with the CO2 loading of the solution; the optimum amount was found to be 0.5 molar equivalents relative to the CO2 absorbed (moles CO2/moles PyBIG = 2). The resulting suspensions were shaken at 1,000 r.p.m. on a vortex mixer, adding a Teflon-coated micro stirbar to each vial to aid in the mixing. Subsamples (50 µl) were withdrawn hourly for the first 8 h and then after 24 h, using a 0.22 μm syringe filter to remove solid particulates from solution before preparing samples for 1H NMR and IC. The subsamples were first diluted with 450 µl of D2O, and then left at room temperature for 24 h before they were analysed by NMR spectroscopy. For IC analyses, 20 µl of the solutions used for NMR analyses were diluted with 980 µl of deionized H2O. At the end of the regenerations, the final isolated solids were filtered and analysed by PXRD for phase identification (Supplementary Figs. 8 and 9). All regeneration experiments were run in triplicate, and the reported values and standard deviations are based on the corresponding average of the three measurements.

For the multi-cycle regeneration experiments, 5 ml of the CO2-saturated sarcosine solution (loaded for 24 h with the humidifier) and 0.5 molar equivalents of solid PyBIG were combined in a 20 ml scintillation vial and stirred as described above, for 2 h at room temperature. Subsequently, the suspension was sampled by NMR and IC using the same protocol as described above. The remaining suspension was filtered through a fine-frit (4.0–5.5 μm) Büchner funnel and the collected solid was placed in a temperature-controlled oven set at 120 °C for 2 h to regenerate the PyBIG compound, which was reused in the subsequent cycle. The regenerated sarcosine solution collected from the above filtration was combined with the bulk of the sarcosine solution (~1,000 ml) that was regenerated by boiling under reflux for 2 h, and reused in a subsequent loading–regeneration cycle.

### CO2 release and regeneration of PyBIG using concentrated solar power

The CO2 release from the PyBIGH2(CO3)(H2O)4 crystals was carried out by solar heating with a Gosun solar oven (Supplementary Fig. 11). The oven consists of a vacuum-insulated borosilicate tube placed in the focal point of two adjustable parabolic reflectors. The temperature inside the tube was monitored with a Fisher Scientific Traceable thermocouple. The PyBIGH2(CO3)(H2O)4 samples (36.4–45.0 mg, 0.095–0.118 mmol) were loaded in 1 ml glass vials, which were placed inside the oven tube. The solar oven was then placed in full sun and oriented to capture the maximum amount of sunlight. The temperature was ramped to the targeted values of 120 °C, 130 °C, 140 °C, 150 °C or 160 °C as fast as possible (typically within 3 to 10 min), and then held within ±2 °C by intermittently moving the oven out of the sunlight, or/and closing the reflectors. The samples were subsequently removed from the tube, allowed to cool to room temperature and weighed to determine their mass loss. The resulting yellow solids were analysed by FTIR to confirm the disappearance of the carbonate and water peaks. The solar regeneration experiments were repeated three times at each temperature, and the average values and standard deviations are reported in Table 3.

### Data availability

The data that support the plots within this paper and other findings of this study are available from the corresponding author upon reasonable request.

Publisher's note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

## References

1. 1.

Trends in Atmospheric Carbon Dioxide (Earth System Research Laboratory, Global Monitoring Division, NOAA, accessed September 2017); https://www.esrl.noaa.gov/gmd/ccgg/trends/global.html

2. 2.

Blunden, J. & Arndt, D. S. (eds) State of the climate in 2016. Bull. Am. Meteor. Soc. 98, Si–S277 (2017).

3. 3.

Hansen, J. et al. Young people’s burden: requirement of negative CO2 emissions. Earth Syst. Dynam. 8, 577–616 (2017).

4. 4.

Gasser, T., Guivarch, C., Tachiiri, K., Jones, C. D. & Ciais, P. Negative emissions physically needed to keep global warming below 2 °C. Nat. Commun. 6, 7958 (2015).

5. 5.

Mauritsen, T. & Pincus, R. Committed warming inferred from observations. Nat. Clim. Change 7, 652–655 (2017).

6. 6.

Keith, D. W. Why capture CO2 from the atmosphere. Science 325, 1654–1655 (2009).

7. 7.

Lackner, K. S. et al. The urgency of the development of CO2 capture from ambient air. Proc. Natl Acad. Sci. USA 109, 13156–13162 (2012).

8. 8.

Sanz-Perez, E., Murdock, C. R., Didas, S. A. & Jones, C. W. Direct capture of CO2 from ambient air. Chem. Rev. 116, 11840–11876 (2016).

9. 9.

Smith, P. et al. Biophysical and economic limits to negative CO2 emissions. Nat. Clim. Change 6, 42–50 (2015).

10. 10.

Psarras, P. et al. Slicing the pie: How big could carbon dioxide removal be? WIREs Energy Environ. 6, e253 (2017).

11. 11.

Zeman, F. Energy and material balance of CO2 capture from ambient air. Environ. Sci. Technol. 41, 7558–7563 (2007).

12. 12.

Baciocchi, R., Storti, G. & Mazzotti, M. Process design and energy requirements for the capture of carbon dioxide from air. Chem. Eng. Process. 45, 1047–1058 (2006).

13. 13.

Stolaroff, J. K., Keith, D. W. & Lowry, G. V. Carbon dioxide capture from atmospheric air using sodium hydroxide spray. Environ. Sci. Technol. 42, 2728–2735 (2008).

14. 14.

Holmes, G. & Keith, D. W. An air–liquid contactor for large-scale capture of CO2 from air. Phil. Trans. R. Soc. A 370, 4380–4403 (2012).

15. 15.

Didas, S. A., Choi, S., Chaikittisilp, W. & Jones, C. W. Amine-oxide hybrid materials for CO2 capture from ambient air. Acc. Chem. Res. 48, 2680–2687 (2015).

16. 16.

Gebald, C., Wurzbacher, J. A., Tingaut, P., Zimmermann, T. & Steinfeld, A. Amine-based nanofibrillated cellulose as adsorbent for CO2 capture from air. Environ. Sci. Technol. 45, 9101–9108 (2011).

17. 17.

Goeppert, A. et al. Carbon dioxide capture from the air using a polyamine based regenerable solid adsorbent. J. Am. Chem. Soc. 133, 20164–20167 (2011).

18. 18.

McDonald, T. M. et al. Capture of carbon dioxide from air and flue gas in the alkylamine-appended metal-organic framework mmen-Mg2(dobpdc). J. Am. Chem. Soc. 134, 7056–7065 (2012).

19. 19.

Lu, W., Sculley, J. P., Yuan, D., Krishna, R. & Zhou, H.-C. Carbon dioxide capture from air using amine-grafted porous polymer networks. J. Phys. Chem. C. 117, 4057–4061 (2013).

20. 20.

Socolow, R. et al. Direct Air Capture of CO 2 with Chemicals. A Technology Assessment for the APS Panel on Public Affairs (American Physical Society, 2011).

21. 21.

Wang, T. et al. Characterization of kinetic limitations to atmospheric CO2 capture by solid sorbent. Greenh. Gas. Sci. Technol. 6, 138–149 (2016).

22. 22.

Wang, T., Lackner, K. S. & Wright, A. B. Moisture swing sorbent for carbon dioxide capture from ambient air. Environ. Sci. Technol. 45, 6670–6675 (2011).

23. 23.

Wang, T., Lackner, K. S. & Wright, A. B. Moisture-swing sorption for carbon dioxide capture from ambient air: A thermodynamic analysis. Phys. Chem. Chem. Phys. 15, 504–514 (2013).

24. 24.

Seipp, C. A., Williams, N. J., Kidder, M. K. & Custelcean, R. CO2 capture from ambient air by crystallization with a guanidine sorbent. Angew. Chem. Int. Ed. 56, 1042–1045 (2017).

25. 25.

Xiang, Q., Fang, M., Yu, H. & Maeder, M. Kinetics of the reversible reaction of CO2(aq) and HCO3 with sarcosine salt in aqueous solution. J. Phys. Chem. A. 116, 10276–10284 (2012).

26. 26.

Guo, D. et al. Amino acids as carbon capture solvents: Chemical kinetics and mechanism of the glycine + CO2 reaction. Energy Fuels 27, 3898–3904 (2013).

27. 27.

Shariff, A. M. & Shaikh, M. S. in Energy Efficient Solvents for CO 2 Capture by Gas-Liquid Absorption (ed. Budzianowski, W. M.) (Springer, Cham, 2017).

28. 28.

Carroll, J. J., Slupsky, J. D. & Mather, A. E. The solubility of carbon dioxide in water at low pressure. J. Phys. Chem. Ref. Data 20, 1201–1209 (1991).

29. 29.

Wang, X., Conway, W., Burns, R., McCann, N. & Maeder, M. Comprehensive study of the hydration and dehydration reactions of carbon dioxide in aqueous solutions. J. Phys. Chem. A 114, 1734–1740 (2010).

30. 30.

McCann, N., Maeder, M. & Hasse, H. A calorimetric study of carbamate formation. J. Chem. Thermodyn. 43, 664–669 (2011).

31. 31.

Hale, J. D., Izatt, R. M. & Christensen, J. J. A calorimetric study of the heat of ionization of water at 25 °. J. Phys. Chem. 67, 2605–2608 (1963).

32. 32.

Le Moullec, Y. & Neveux, T. in Absorption-Based Post-Combustion Capture of Carbon Dioxide (ed. Feron, P. H. M.), Woodhead Publishing Series in Energy: Number 101 (Elsevier, 2016).

33. 33.

Parikh, D. M. Vacuum drying: basics and application. Chem. Eng. 122, 48–54 (2015).

34. 34.

Yang, N. et al. Protonation constants and thermodynamic properties of amino acid salts for CO2 capture at high temperature. Ind. Eng. Chem. Res. 53, 12848–12855 (2014).

35. 35.

Nikulshina, V., Gebald, C. & Steinfeld, A. CO2 capture from atmospheric air via consecutive CaO-carbonation and CaCO3-calcination cycles in a fluidized-bed solar reactor. Chem. Eng. J. 146, 244–248 (2009).

36. 36.

Sanchez Fernandez, E. et al. Conceptual design of a novel CO2 capture process based on precipitating amino acid solvents. Ind. Eng. Chem. Res. 52, 12223–12235 (2013).

37. 37.

Perry, R. J. et al. CO2 capture using phase-changing sorbents. Energy Fuels 26, 2528–2538 (2012).

38. 38.

Custelcean, R., Williams, N. J., Seipp, C. A., Ivanov, A. S. & Bryantsev, V. S. Aqueous sulfate separation by sequestration of [(SO4)2((H2O)4]4– clusters within highly insoluble imine-linked bis-guanidinium crystals. Chem. Eur. J. 22, 1997–2003 (2016).

## Acknowledgements

This research was supported by the US Department of Energy, Office of Science, Basic Energy Sciences, Chemical Sciences, Geosciences, and Biosciences Division.

## Author information

### Author notes

1. These authors contributed equally: Flavien M. Brethomé and Neil J. Williams.

### Affiliations

1. #### Chemical Sciences Division, Oak Ridge National Laboratory, Oak Ridge, TN, USA

• Flavien M. Brethomé
• , Neil J. Williams
• , Charles A. Seipp
• , Michelle K. Kidder

### Contributions

F.M.B. performed and analysed the CO2 absorption and sorbent regeneration experiments, and the potentiometric titration measurements. N.J.W. optimized and scaled up the synthesis of PyBIG, optimized the CO2 absorption and sorbent regeneration with PyBIG, and performed the solubility measurements. C.A.S. designed and synthesized the PyBIG compound. M.K.K. performed and analysed the DSC and TGA measurements. R.C. led the project, conceptualized the study, performed the measurements with concentrated solar power and wrote the manuscript. All authors contributed to discussions and manuscript reviews.

### Competing interests

A US patent application (no. 15/813,557), currently pending, has been filed, with R.C., C.A.S. and N.J.W. as inventors, covering the DAC system described in this manuscript.

## Supplementary information

1. ### Supplementary Information

Supplementary Figures 1–12, Supplementary Tables 1–3 and Supplementary Methods

### DOI

https://doi.org/10.1038/s41560-018-0150-z

• ### Organic agents offer innovation

• Christopher W. Jones

Nature Energy (2018)