Electrifying amine carbon capture with robust redox-tunable acids

The urgent need to address climate change has sparked a growing interest in carbon capture technologies, which play a crucial role in reducing greenhouse gas emissions. Currently, the most mature method for point source carbon capture is amine scrubbing, which employs aqueous amines as absorbents and thermal swing processes to regulate CO₂ separation (Fig. 1a)^1,2,3,4. In this approach, amines are regenerated by stripping with water vapor at 100–120 °C, leading to a multitude of challenges, including low energy efficiency, fast absorbent loss via evaporation or degradation, equipment corrosion, and complexities in integrating this technology with existing infrastructures. Recent efforts in developing water-lean CO₂ absorbents are attractive as they can enhance CO₂ capacity through different binding mechanisms from aqueous systems⁵ and mitigate the evaporative and sensible heat losses by utilizing non-volatile organic solvents^6,7,8. Nevertheless, these nonaqueous absorbents still require thermal regeneration.

a Conventional amine scrubbing via thermal-swing processes and previous aqueous/nonaqueous electrochemical amine regeneration based on heterogeneous electrochemistry involving heavy metal species. b Chemical versatility of redox-tunable Brønsted acids (RAs) and schematic showing the mechanism of redox-tunable Brønsted acids mediated amine regeneration (RAMAR).

Full size image

The increasing availability of renewable electricity has created a compelling opportunity to enhance the efficiency and sustainability of carbon capture through electrification^9,10, as highlighted in recent reviews^{11,12,13,14,15,16,17,18,19}. Electrochemically meditated carbon capture (EMCC) can be achieved via various mechanisms, such as redox-driven reversible CO₂ complexation using organic sorbents^{20,21,22,23,24} or pH swing^{25,26,27,28,29}. Unlike conventional thermochemical methods, EMCC is compatible with intermittent renewable energy and can operate under mild, isothermal conditions. Moreover, the modular nature of EMCC allows the development of plug-and-play systems, enhancing their adaptability to the diverse and multiscale demands of carbon capture. However, current redox-active materials used in EMCC face fundamental limitations. Specifically, existing redox-active CO₂ sorbents and pH mediators are often sensitive to molecular oxygen (O₂) in their activated (reduced) form^22,30,31, and parasitic reactions during repeated capture-release cycles lead to the accumulation of unstable species and rapid sorbent degradation^21,23. Additionally, costly synthetic procedures limit the scalability of current EMCC materials for scalable implementations.

To capitalize on the exceptional CO₂ removal efficiencies and optimized process equipment of amine absorption while reducing parasitic energy and capital costs, we present a scale-transferrable approach to electrify amine scrubbing, termed redox-tunable (Brønsted) acid mediated amine regeneration (RAMAR). RAMAR leverages Wurster-type compounds as redox-tunable Brønsted acids (RAs) to modulate CO₂ capture and release across a broad spectrum of classic amines (Fig. 1b). In their oxidized state, RAs function as strong Brønsted acids (protonated iminium), facilitating CO₂ release by protonating amine-CO₂ adducts. Conversely, in the reduced state, RAs transition into potent Brønsted bases (deprotonated azanide), effectively regenerating amines for CO₂ capture by abstracting protons from ammoniums. At the molecular level, RAs exhibit highly reversible pK_a modulation spanning over 20 units in nonaqueous solvents through a redox cycle, which is the largest tunability reported among known stimuli-responsive acids/bases to our best knowledge^32,33,34. This wide dynamic pKa range is the key enabler for electrochemical amine regeneration. Moreover, RAs demonstrate exceptional redox stability and molecular designability and are synthesized and isolated in pure form by facile chemical methods. At the process level, unlike previous attempts at electrifying amine carbon capture, such as those using copper redox^{35,36,37,38,39} and dual-salt cation swing⁴⁰ (Fig. 1a), RAMAR eliminates the need for heavy metals and poorly reversible electroplating/intercalation chemistries. Importantly, all species involved in the RAMAR process are thermodynamically stable in ambient conditions (especially O₂), thereby offering opportunities for practical implementation in complex gas environments.

Herein, RAMAR is evaluated in symmetric, flow-based carbon capture prototypes, where the electrochemical cell design can enhance the energy efficiency and cycling stability compared to many previous EMCC configurations²⁹. Using the readily available N¹,N⁴-diphenylbenzene-1,4-diamine (RA 1), we showcase continuous CO₂ capture-release operation over 80 cycles (400 h) under realistic aerobic conditions (10% CO₂ and 21% O₂) with near-unity electrochemical efficiency and nearly half the energy consumption compared to previous EMCC processes. The RAMAR mechanism lays the foundation for developing efficient EMCC systems and can potentially serve as a drop-in replacement for amine scrubbing to accelerate the widespread adoption of carbon capture technologies.

Validating and understanding redox-tunable Brønsted acids

Governed by the hard-soft acid-base theory, alkaline CO₂ sorbents, such as sp³-N amines, form strong adducts with CO₂ via acid-base interaction, thus necessitating substantial energy input to break the N–C bond for CO₂ release. In principle, chemical acidification and subsequent basification could liberate CO₂ and regenerate amines for reversible CO₂ separation at ambient temperatures. To avoid introducing non-recyclable chemical reagents and additional separation steps, pH swing strategies employing redox-active mediators, which can reversibly bind and release protons at distinct electrochemical potentials, present a promising avenue for electrified amine scrubbing. Specifically, the mediators should have much lower pK_a values than amines in their oxidized state while exhibiting much higher pK_a values than amines in their reduced state. Nonetheless, reported redox-active pH mediators do not have the required dynamic pK_a range^28,29,41,42. To bridge this knowledge gap, we judiciously explore a class of Wurster-type compounds to investigate their potential as redox-tunable proton acceptors/donors in the presence of amines. These compounds are secondary amines/azanides (proton acceptors) in the reduced state and iminiums/imines (proton donors) in the oxidized state, which are considered strong bases and acids, respectively. Moreover, RAs are stable and isolatable in both reduced and oxidized forms via straightforward synthetic steps (see Synthetic Procedures in Methods and Supplementary Figs. 1–8). We postulate that oxidizing RAs in the presence of CO₂-loaded amines can result in the donation of protons to amines for CO₂ release, whereas subsequent reduction of oxidized RAs can attract protons from ammonium to regenerate amines for CO₂ capture (Fig. 2a).

a Reversible proton-coupled oxidation/reduction reactions between 1 and 1-ox in the presence of MEA. b Crude ¹H NMR of 1 and 1-ox after bulk electrolysis in the presence of MEA under 10% CO₂ atmosphere. Bulk electrochemical oxidation of 1 affords 1-ox identical to chemical oxidation methods with negligible impurities. c CV curves of 1 and 1-ox under various conditions. 1 or 1-ox (5 mM) was measured in DMSO saturated with N₂ and with tetrabutylammonium hexafluorophosphate (NBu₄PF₆, 100 mM) as the supporting salt. 1 and 1-ox were scanned anodically and cathodically at a rate of 50 mV s⁻¹, respectively. Source data for Fig. 2 are provided as a Source Data file.

Full size image

To validate the proposed chemistry, we conducted bulk electrolysis of compound 1 in deuterated dimethyl sulfoxide (DMSO-d₆) in the presence of two molar equivalents of monoethanolamine (MEA) under 10% CO₂ (Fig. 2b and Supplementary Figs. 9 and 10). The crude proton nuclear magnetic resonance (¹H NMR) spectrum after bulk electrochemical oxidation of 1 suggests the full conversion to N¹,N⁴-diphenylcyclohexa-2,5-diene-1,4-diimine (1-ox) with negligible impurities, as compared to the ¹H NMR of 1-ox derived from the chemical oxidation method⁴³. Besides, the formation of monoethanolammonium indicates successful proton transfer from 1 to MEA after oxidation (see Supplementary Figs. 11–13 for the NMR spectra of pristine MEA, its protonated form, and carbamic acid/carbamate form). Furthermore, the reaction exhibits high chemical reversibility, as suggested by the complete regeneration of 1 and carbamic acid/carbamate in the crude ¹H NMR after subsequent bulk electrochemical reduction of 1-ox.

Cyclic voltammetry (CV) experiments were also conducted on 1 and 1-ox under various conditions (Fig. 2c, see Supplementary Fig. 14 for the corresponding chemical reaction under each condition). CV reveals an irreversible electrochemical reduction process of the imine compound 1-ox in the absence of free protons (curve i). The rather negative reduction potential of 1-ox indicates the strong basicity of 1’s conjugate base (deprotonated form). By adding two molar equivalents of superacidic bistriflimidic acid (HTFSI) to the 1-ox solution, the CV of 1-ox (cathodic scan, curve ii) becomes electrochemically reversible and mirrors that of 1 (anodic scan, curve iii). The results strongly evidence the reversible electrochemical interconversion between 1 and protonated 1-ox (iminium form). In the presence of a stoichiometric amount of MEA, the CV of 1 (anodic scan, curve iv) becomes electrochemically irreversible again, accompanied by cathodically shifted reduction and oxidation potentials compared to curve iii. This implies a proton-coupled electron transfer (PCET) process, where the basic MEA attracts protons from 1 via hydrogen bonding and facilitates its oxidation but impedes the proton transfer to 1-ox and thus negatively shifts its reduction potential. Similarly, the cathodic scan of 1-ox conducted with a stoichiometric amount of MEA and HTFSI (curve v) exhibits an identical CV behaviour to that observed in curve iv, further confirming the reversibility of the reaction shown in Fig. 2a.

We further compared the CV of 1 in the absence and presence of MEA under N₂ and CO₂ atmospheres (Supplementary Fig. 15a). As mentioned earlier, the addition of MEA breaks the redox reversibility of 1 through the introduction of a PCET process, with both cathodically shifted onset potentials for reduction (−0.9 V vs. Fc^+/0) and oxidation (−0.4 V vs. Fc^+/0). Interestingly, when the solution is saturated with pure CO₂, the oxidation onset potential returns to −0.08 V vs. Fc^+/0, identical to that in the absence of MEA. This indicates the consumption of free MEA by forming CO₂ adducts, which eliminates the hydrogen bonding that promotes oxidation. Alternatively, the presence of CO₂ anodically shifts the reduction potential of 1 with MEA compared to that under N₂, which we attribute to the amine-CO₂ binding that facilitates proton transfer from ammonium to 1-ox. The CV of 1 was also tested and analyzed at lower CO₂ partial pressures (10–20%) (Supplementary Figs. 15a and 16 and Supplementary Note 1).

To gain fundamental insights into the PCET process of 1 in the absence/presence of MEA in DMSO under N₂ and CO₂, we proposed the nine-state equilibrium chart and calculated the potential of each electron transfer step and the Gibbs free energy change of each proton transfer step using density functional theory (DFT, Fig. 3a). We consider three scenarios for the proton transfer steps (Supplementary Fig. 17), including proton transfer between different forms of 1 and the DMSO solvent molecule (reaction i), MEA (reaction ii), and MEA-CO₂ adduct (reaction iii), which represents PCET without amine, with amine under N₂, and with amine under CO₂, respectively. The blue-shaded region in Fig. 3a shows the most energetically favourable reaction pathway in the presence of MEA, where the two electron transfer steps occur at the same potential, corroborating our experimental CV observation. Besides, the proton transfer steps show the lowest free energy in the presence of MEA under N₂ (reaction ii), which also agrees well with the most negatively shifted CV oxidation potential. Furthermore, we calculated the pK_a of each state of 1, revealing a wide swing between −7.3 and 30.1 (Fig. 3a).

a PCET pathways under conditions without amine (i), with MEA and N₂ (ii), and with MEA and CO₂ (iii). The potentials (vs. Fc^+/0) for the electron transfer steps (E), the changes in Gibbs free energy for the proton transfer steps (∆G), and the pK_a values of each state are calculated by DFT and compared with experimental results if available. The most favourable pathway in the presence of proton donors with varying pK_a is shaded in pink (1), blue (2), green (3), and yellow (4), corresponding to the four different regions in the pK_a-potential diagram in (b). b Reduction potentials of 1-ox in the presence of proton donors with varying pK_a (−14.9 to 32 in DMSO). The data was obtained by the average of three independent CV measurements in the presence of proton donors under N₂. The diagram is divided into four regions, which undergo different reaction pathways highlighted in (a). Source data for Fig. 3 are provided as a Source Data file.

Full size image

To experimentally estimate the pK_a values of 1 and the conjugate acid of 1-ox, we measured the CV of 1-ox in DMSO under N₂ in the presence of two molar equivalents of acids (proton donors) with various pK_a (−14.9 to 31.3 in DMSO)^44,45. The pK_a value of each proton donor used is shown in Supplementary Fig. 18. The reduction potential-pK_a diagram reveals two pK_a-independent regions and two pK_a-dependent regions in between (Fig. 3b). The independent regions indicate the upper and lower bound pK_a of the system as 1.6 (pK”_a2) and 21.1 (pK_a2), respectively (see Fig. 3a for the pK_a labels of the different forms of 1). The pK_a-dependent regions exhibit two different slopes of ~113.2 mV pK_a⁻¹ (1.6 to 7.47) and ~45.4 mV pK_a⁻¹ (7.47 to 21.1), corresponding to two protons and one proton transfer per electron transfer, respectively. The deviation from the idealized Nernstian slope is likely caused by homoconjugation of the acids in proton-accepting DMSO^44,45. The pK_a tunability range of RA 1 is, to the best of our knowledge, the widest among known stimuli-responsive acids and bases. In comparison, typical photoacids exhibit pK_a shifts of only 1–3 units³³, while photobases show shifts in the range of 1–9 units³².

Based on the reduction potential-pKa diagram, the PCET pathways of 1/1-ox under different pK_a environments are proposed in Supplementary Fig. 18. For proton donors with pK_a below 1.6 or above 21.1, no proton transfer is involved in the redox process. This is because protons cannot dissociate from 1-ox ([1-ox + 2H]²⁺, iminium form) under strongly acidic conditions, while protons cannot bind to 1 ([1 − 2H]²⁻, azanide form) under strongly basic conditions. In the region of 1.6 <pK_a < 7.5, one electron reduction of 1-ox can result in doubly protonated [1’ + H]⁺ due to the relatively strong acidic environment and the strong basicity of 1’. In the region of 7.5 <pK_a < 21.1, an overall two-proton-two-electron-transfer process dominates the redox reactions between 1 and 1-ox.

With an onset reduction potential of −0.9 V (vs. Fc^+/0), MEA falls into the pK_a region between 7.5 and 21.1, implying that CO₂ separation using 1 and MEA operates under a two-proton-two-electron mechanism. Furthermore, Fig. 3b provides a guideline for pairing suitable amines with RAs by matching their pK_a values. For energy efficiency considerations, it is ideal to drive amine regeneration under a one-electron-two-proton mechanism. This requires selecting an amine with lower Brønsted basicity (low pK_a for its conjugate acid) but sufficient nucleophilicity for CO₂ absorption. However, such amines rarely exist, as strong nucleophiles are usually strong bases. To circumvent this challenge, molecular engineering of RAs may potentially shift the one-electron-two-proton region to larger pK_a values to encompass the common amine absorbents.

Expanding the chemical space of RAs and candidate amines

With the RAMAR mechanism validated using model compound 1 and MEA, we further expanded the chemical space of RAs to tailor their physicochemical properties. As shown in Fig. 1b, the Wurster-type structural motif serves as a versatile platform for molecular engineering. Specifically, we demonstrate three additional structures with the ease of encoding functionality at N-substitutes and side chains, which are N¹,N⁴-di(naphthalen-2-yl)benzene-1,4-diamine (2), N¹,N⁴-di-sec-butylbenzene-1,4-diamine (3), and diethyl 2,5-bis(phenylamino)terephthalate (4). These compounds are either commercially available or can be obtained via facile synthetic steps from simple precursors (see Synthetic Procedures in Methods).

The redox potential and the reversibility of RAs in the presence of amine can be tuned by induction and resonance effects through structural modification, respectively (Supplementary Fig. 15 and Supplementary Table 1). Specifically, electron-donating or electron-withdrawing groups (EDGs or EWGs) can be introduced into the Wurster-type motif, shifting the redox potential cathodically or anodically governed by the free-energy relationship in the Hammett equation⁴⁶. Interestingly, the redox reversibility (the voltage gap between oxidation and reduction) can be enhanced by adding EDGs or elongating RA conjugation, thereby lowering the theoretical work of CO₂ separation. In a practical RAMAR process, RAs are oxidized under a CO₂-rich atmosphere to desorb concentrated CO₂ and reduced under a CO₂-lean atmosphere to regenerate amines. In the presence of MEA, the CV peak difference between oxidation under 100% CO₂ and reduction under N₂ are 0.76, 0.68, 0.57, and 0.96 V for 1, 2, 3, and 4, respectively (Supplementary Fig. 15 and Supplementary Table 1). A smaller voltage gap via molecular design corresponds to a lower theoretical energy consumption.

For practical applications, RAs need to be stable against O₂, which is a common contaminant in the gas streams of interest for carbon capture. Most conventional redox-active CO₂ sorbents are O₂ sensitive in their reduced states, causing undesired parasitic reactions^21,29,31. In stark contrast, 1 exhibits high robustness towards O₂, suggested by CV under simulated flue gas, which contains 10% CO₂ and 3% O₂ (Supplementary Fig. 19). In a DMSO solution with 100 mM LiTFSI, pure O₂ exhibits an onset reduction potential of −1.2 V (vs. Fc^+/0), giving rise to LiO₂/Li₂O₂ products. In the presence of CO₂, the reduction potential of O₂ shifts anodically to −1.1 V (vs. Fc^+/0), probably due to the formation of peroxocarbonate or percarbonate species. This chemical property challenges the design of redox-tunable CO₂ sorbents for EMCC under practical operating conditions, as the O₂ reduction reaction may occur alongside or even prior to sorbent reduction/activation. Redox-active CO₂ sorbents can also act as redox mediators to further promote O₂ reduction⁴⁷. So far, sorbents with the most positive redox potentials are 4,4’-azopyridine (−1.1 V vs. Fc^+/0 in DMSO) and isoindigos (−1.2 to −0.9 V vs. Fc^+/0 in DMF)^23,24. To our delight, CV reveals that 1 possesses an onset reduction potential of −0.75 V (vs. Fc^+/0) in the presence of MEA in DMSO and has no redox mediation effect on O₂ reduction even with CO₂ present. Furthermore, 1 is thermodynamically robust against O₂. ¹H NMR shows 1 maintains its chemical integrity after being purged and stored in a pure O₂ atmosphere in DMSO after 7 days (Supplementary Fig. 20a).

As an additional note on sorbent stability, the RAMAR chemistry minimizes parasitic reactions by decoupling electron transfer and CO₂ complexation. That is, amines are exclusively responsible for CO₂ capture/release, while RAs serve as proton acceptors/donors modulated by electrochemical potentials. This is fundamentally different from conventional redox-active CO₂ sorbents^11,12,16, where electron transfer and CO₂ complexation occur at the same reaction site, causing sorbent reversibility and O₂ stability issues.

To facilitate efficient, flow-based EMCC technologies, RAs need to be highly soluble in the electrolyte solvent to ensure practical CO₂ separation capacity (Supplementary Fig. 20b). The solubility of different RAs is summarized in Supplementary Table 2. It is noteworthy that 1 possesses a high solubility of 2.68 M in pure DMSO, and the coexistence of supporting salts and amines has little impact on its solubility. For example, the solubility of 1 is maintained at 2.71, 2.43, and 1.40 M in DMSO solutions comprising 0.5 M LiTFSI, 0.5 M LiTFSI with 0.5 M MEA, and 0.5 M LiTFSI with 5 M MEA, respectively. The solubility of RAs can be further improved by molecular engineering; replacing phenyl groups at N-substitutes with sec-butyl groups gives rise to 3, which is a liquid at room temperature and fully miscible with the electrolyte.

In addition to RAs, optimizing the physicochemical properties of amine absorbents and their CO₂ adducts is equally important for RAMAR practicality. We expanded the chemical space of amine absorbents beyond MEA by measuring the CV of 1 in the presence of different amines, including diethanolamine (DEA), morpholine (Mph), tetrahydroisoquinoline (THIQ), piperazine (Pz), and 2,2′-(ethylenedioxy)bis(ethylamine) (DODA) (Fig. 4). To our delight, these amines exhibit CV curves similar to that of MEA, suggesting the generalizability of the RAMAR chemistry. The reduction/oxidation peaks shift slightly for each amine due to their different basicity, which are summarized in Supplementary Table 3.

a Monoethanolamine (MEA). b Diethanolamine (DEA). c Morpholine (Mph). d Tetrahydroisoquinoline (THIQ). e Piperazine (Pz). f 2,2′-(ethylenedioxy)bis(ethylamine) (DODA). CV of 1 (5 mM) was measured in DMSO containing amine (10 mM), NBu₄PF₆ (100 mM) as the support salt under N₂, 10% CO₂, and pure CO₂, respectively. 1 was scanned anodically at a rate of 50 mV s⁻¹. Source data for Fig. 4 are provided as a Source Data file.

Full size image

Amines with sp³-N are usually considered susceptible to oxidation. Therefore, the electrochemical stability windows of amine solutions were determined by linear sweep voltammetry under both N₂ and CO₂ (Supplementary Fig. 21). Primary amines possess greater oxidative stability than secondary amines in N₂ atmosphere, and CO₂ stabilizes amines from oxidative decomposition due to adduct formation, as indicated by the positively shifted oxidation potentials. As one of the most readily available amines, MEA displays a stability window from −2.72 to 0.9 V (vs. Fc^+/0) under CO₂, which is sufficient to cover the redox range of 1. Furthermore, we observed that monoamines exhibit decent solubility in their CO₂ adduct forms. Using MEA and THIQ as examples, they are liquid and fully miscible with pure DMSO or DMSO with 0.5 M LiTFSI. When saturated with CO₂, the amine solutions (amine:DMSO = 1:1 v/v) exhibit no obvious precipitation (Supplementary Fig. 20c). On the contrary, we found that the CO₂ adducts of diamines, like Pz and DODA, possess poor solubility in DMSO (Supplementary Fig. 20d). White precipitates quickly formed when purging 10% CO₂ into a solution of Pz or DODA (0.5 M) in DMSO. This is probably caused by the formation of zwitterionic species or hydrogen bonded polymers of anionic carbamates and cationic ammoniums⁴⁸. Therefore, primary monoamines are considered the best candidates for RAMAR.

Finally, we tested the kinematic viscosity of the RA-amine mixtures under practical CO₂ absorption conditions (Supplementary Table 4). For the case of 0.5 M LiTFSI, 0.5 M 1, 0.1 M 1-ox, 0.2 M HTFSI, and 0.5 M MEA in DMSO, the 10% CO₂ saturated mixture displays as a free-flow liquid with a kinematic viscosity of 6.76 mm² s⁻¹ at 25 °C, which is ~3 times that of pure DMSO and ~5 times that of aqueous MEA solution (0.5 M, saturated with 10% CO₂). We also determined the viscosity of highly concentrated amine solutions (5 M MEA) saturated with 10% CO₂. A DMSO solution containing 5 M MEA and 0.5 M LiTFSI and an aqueous 5 M MEA solution exhibit comparable kinematic viscosities of 4.28 and 2.44 mm² s⁻¹ at 25 °C, respectively. The decent viscosity of our RAMAR formulation ensures reasonable pumping costs and favourable mass transfer kinetics for CO₂ capture.

Determining the CO₂ absorption capacity in RAMAR

Although the reaction between CO₂ and amines in aqueous environments has been extensively studied⁴⁹, the amine-CO₂ adducts are less understood in nonaqueous media. A previous report revealed that amines, such as 2-ethoxyethylamine, can form both carbamic acid and carbamate adducts with CO₂ in DMSO⁵⁰. The former involves a one-amine-one-CO₂ formation process, while the latter requires two amine equivalents and one captured CO₂ to generate the carbamate-ammonium dyads (Fig. 5a). Therefore, it is of great importance to understand the amine adduct composition to determine the CO₂ absorption capacity in RAMAR.

a Regeneration of carbamic acid adduct via 1H⁺−1CO₂ pathway and carbamate adduct via 2H⁺−1CO₂ pathway. b ¹H NMR study of the amine-CO₂ reaction pathways in DMSO as a factor of amine concentration, supporting salt, and CO₂ partial pressure. Black, MEA (50 mM) under N₂; red, MEA (50 mM), HTFSI (50 mM) under N₂; blue, MEA (50 mM) saturated with 100% CO₂; green, MEA (50 mM) in DMSO with LiTFSI (500 mM) saturated with 100% CO₂; purple, MEA (500 mM) in DMSO with LiTFSI (500 mM) saturated with 100% CO₂; yellow, MEA (500 mM) in DMSO with LiTFSI (500 mM) saturated with 10% CO₂. c Illustration of the titration experiment setup. The solution was purged by a stream of 20% CO₂ (balanced with N₂) at a flow rate of 10 sccm. The CO₂ level at the exit of the septum-capped vial was monitored by an IR-based CO₂ sensor. For the CO₂ absorption experiment, various amounts of MEA were added to the vial containing either neat DMSO or DMSO with 0.5 M LiTFSI. For the CO₂ desorption experiment, a solution of MEA (500 mM) and LiTFSI (500 mM) in DMSO was saturated with 20% CO₂ and then added with various amounts of TfOH. For both experiments, the CO₂ level was allowed to return to 20% before the next addition. d CO₂ absorption capacity at various MEA concentrations. Grey, pure DMSO solvent; red, DMSO with 500 mM LiTFSI as the supporting salt. e CO₂ desorption capacity with various concentrations of TfOH added. Source data for Fig. 5 are provided as a Source Data file.

Full size image

We investigated the interaction between MEA and CO₂ in DMSO via a combination of NMR spectroscopy and titration experiments. ¹H NMR reveals that the carbamic acid (a’, 6.57, b’, 3.36, and c’, 2.98 ppm) pathway dominates in dilute MEA solution (50 mM) without supporting salts under 100% CO₂ due to the absence of ammonium species (b, 3.54 and c, 2.82 ppm) (Fig. 5b). To mimic RAMAR working conditions, LiTFSI (500 mM) was added as the supporting salt to examine its effect on CO₂ absorption pathways. As evidenced by the appearance of ammonium species in ¹H NMR, Li⁺ acts as a Lewis acid to stabilize the carbamate adduct, which lowers the CO₂ absorption capacity^40,50. Similarly, increasing the MEA concentration (from 50 mM to 500 mM) also induces the carbamate pathway via deprotonating the carbamic acid by free amines. Lastly, reducing the CO₂ partial pressure (from 100% to 10%) promotes the formation of carbamate-ammonium mixture due to decreased CO₂ availability in the solution. Under simulated RAMAR working conditions with a 10% CO₂ feed, the DMSO solution (with 500 mM MEA and 500 mM LiTFSI) exhibits a nearly 1:1 mole ratio of the carbamic acid and carbamate adduct by comparing the integration of the carbamic acid/carbamate peak (c’, 2.98 ppm) and ammonium peak (c, 2.82 pm), equivalent to 0.75 CO₂ absorbed per amino group.

Titration experiments (Fig. 5c) were conducted to quantitatively assess the CO₂ absorption capacity of MEA and the subsequent CO₂ release facilitated by triflic acid (TfOH). Under a constant influx of 20% CO₂ (balanced with N₂) at a flow rate of 10 standard cubic centimetres per minute (sccm), various amounts of MEA were injected into pure DMSO or DMSO containing 0.5 M LiTFSI (pre-saturated with 20% CO₂). The CO₂ concentration at the outlet was monitored using an IR-based CO₂ sensor to determine the net amount of CO₂ absorbed. Similarly, for CO₂ desorption, various amounts of TfOH were injected into a DMSO solution containing 0.5 M MEA and 0.5 M LiTFSI (pre-saturated with 20% CO₂), followed by monitoring of the outlet CO₂ level. As depicted in Supplementary Fig. 22, immediate changes in CO₂ readings were observed upon the addition of MEA or TfOH, indicating CO₂ capture and release, respectively. The total amounts of CO₂ captured/released per equivalent of MEA/TfOH are plotted against the final concentrations of MEA/TfOH in Fig. 5d, e.

Consistent with NMR, titration experiments suggest a lower CO₂ absorption capacity in DMSO containing LiTFSI compared to pure DMSO, alongside a notable reduction in CO₂ absorption capacity with increasing MEA concentration. Figure 5e reveals that the concentration of superacid has a negligible impact on CO₂ release capacity. Specifically, a DMSO solution with 0.5 M MEA and 0.5 M LiTFSI can release 0.6 equivalent of CO₂ for every equivalent of TfOH added. According to Fig. 3b, 1-ox’s conjugate acid exhibits a comparable pK_a to TfOH. Therefore, we posit that our RAMAR formulation (0.5 M MEA and 0.5 M LiTFSI in DMSO) has a theoretical capacity to separate 0.6 CO₂ per monoamine. That is, each electron input to RAs will generate one proton, which can liberate 0.6 CO₂. This theoretical value will be used subsequently to assess the actual performance of our RAMAR prototype.

Evaluating RAMAR performance in symmetric flow cells

A prominent challenge facing nonaqueous EMCC devices is the crossover of active species through the ion exchange membrane, stemming from the lack of membranes compatible with organic electrolytes^16,51,52, which compromises the cycling stability^23,24. Previous studies from others and us^21,23,24,28 utilize counter electrolytes such as ferrocene, leading to low electrochemical (Coulombic) efficiency (typically <90%) and incomplete regeneration of redox-active sorbents, largely due to counter electrolyte crossover. The use of counter electrolytes also gives rise to large overpotentials (energy input) between CO₂ capture and release steps. Symmetric flow cells, which utilize the same material in different redox states as both catholyte and anolyte, provide a plausible solution for mitigating crossover. Nevertheless, such a concept has yet to be demonstrated in nonaqueous EMCC due to the chemical susceptibility of redox-active CO₂ carriers in their reduced state. Consequently, prior to assembling symmetric cells, conventional redox-active CO₂ carriers must undergo electrochemical reduction against counter electrolytes in flow cells, a process that is time-consuming, complicates reactor design, introduces undesired impurities from counter electrolyte crossover, and increases the overall cost.

In contrast, our RA molecules are stable and isolatable in both reduced and oxidized forms. The synthetic feasibility of the oxidized counterparts of RAs streamlines the assembly of symmetric flow cells. Furthermore, we compared the crossover rate of 1 and ferrocene in LiTFSI DMSO solution across Nafion 212, a commonly used membrane in EMCC flow cells (Supplementary Fig. 23 and Supplementary Table 5)²³. Interestingly, the crossover rate of 1 is 20 times slower than ferrocene, a characteristic that is conducive to achieving high CO₂ separation efficiency in symmetric cells.

As a proof of concept, we built a bench-scale symmetric flow cell to evaluate the performance of RAMAR via cyclic CO₂ capture-release operations (Supplementary Fig. 24). In the working electrode tank (WE), the electrolyte (a mixture of RA in the reduced and oxidized forms) was first reduced and activated for CO₂ capture. In the counter electrode tank (CE), the electrolyte was first oxidized and deactivated for CO₂ release. The polarity of the applied potential was switched periodically.

We started performance evaluation using RA 1 and MEA (Fig. 6, 125 mM 1, 125 mM 1-ox, 500 mM MEA). In a typical cycle, under an influx of 10% CO₂ at a flow rate of 2 sccm, WE and CE were first reduced and oxidized at 10 mA (2 mA cm⁻²) for 90 min, respectively. The CO₂ readings at the gas outlet of WE and CE confirmed CO₂ capture and release. Subsequently, the CO₂ readings of WE and CE were allowed to return to the baseline by halting the current for 90 min. WE and CE were then oxidized and reduced at 10 mA for 90 minutes, respectively, followed by another 90 min of rest.

a CO₂ readings at the outlet of the WE tank (blue) and CE tank (red), respectively, over 21 repeating cycles for 126-h operation. The CO₂ sensor at the CE tank lost connection to the computer at ~40 h, resulting in a data loss of ~2 h during the 7th cycle of CO₂ capture. b, c CO₂ readings for selected capture-release cycles with the corresponding cumulative amount of CO₂ captured/released in WE (b) and CE (c). The yellow regions highlight the capture and release steps. For each separation cycle, both sorbent tanks were reduced/oxidized at 10 mA (2 mA cm⁻²) for 90 min followed by a 90 min rest. d, e Performance metrics of WE (d) and CE (e). WE and CE were composed of 10 ml DMSO solution containing 125 mM 1, 125 mM 1-ox, 250 mM HTFSI, 500 mM MEA, and 500 mM LiTFSI. Both tanks were stirred and continuously purged with 10% CO₂ (balanced with N₂) at a flow rate of 2 sccm. Source data for Fig. 6 are provided as a Source Data file.

Full size image

Our initial test exhibited consistent CO₂ capture-release behaviours over repeated cycling (Fig. 6a–c). 1 demonstrated fully reversible operation with an electrochemical efficiency of 100% (Fig. 6d, e). Furthermore, the system exhibited CO₂ capacity utilization efficiency close to unity for both WE and CE, which is defined as the actual amount of CO₂ separated relative to the theoretical value (See detailed calculations in Supplementary Note 2). The high reversibility of the CO₂ capture-release process is evidenced by the ~100% release/capture efficiency. Overall, the system reveals an average work of separation of 62.3 ± 3.8 kJ mol⁻¹ CO₂ over 21 cycles (Supplementary Fig. 25a), nearly half of the numbers previously reported for nonaqueous EMCC systems. A summary of benchmark EMCC performance is shown in Supplementary Table 6.

We also evaluated the EMCC performance of RA 3 (Supplementary Fig. 26). Over 41 cycles (205 h), the symmetric cell exhibited an average work of separation of 90.6 ± 3.8 kJ mol⁻¹ CO₂, an average CO₂ capacity utilization of ~85%, a release/capture efficiency of ~100%, and an electrochemical efficiency of ~100%.

The promising results from our initial tests motivated us to push the limit of this system further. Leveraging the low cost and synthetic feasibility of 1 and 1-ox, we upscaled the system by raising the sorbent concentration (0.5 M 1, 0.05 M 1-ox, 0.5 M MEA). Using the same symmetric flow cell configuration, we tested the EMCC performance at various current densities (1–10 mA cm⁻², Supplementary Fig. 27). By elevating the current density, the system exhibited an increasing rate of CO₂ capture and release, as evidenced by the sharper and deeper changes in CO₂ readings (Supplementary Fig. 27a). The three metrics of performance, including CO₂ capacity utilization efficiency, release/capture efficiency, and electrochemical efficiency, are summarized in Supplementary Fig. 27b. The release/capture and electrochemical efficiencies remained constant with increasing current, and the CO₂ capacity utilization displayed a moderate decrease to ~75% as the current was raised to 10 mA cm⁻². The work of separation gradually increased from 66 ± 0.8 to 177.6 ± 6.4 kJ mol⁻¹ CO₂ from 1 to 10 mA cm⁻² (Supplementary Fig. 27c, d).

Subsequently, we investigated the effect of water on the performance of 1 by adding 10% (v/v) of water to the DMSO electrolyte. As shown in Supplementary Fig. 28, the RAMAR flow cell maintained stable cycling over 16 cycles (112 h) with performance comparable to the water-free conditions. The system exhibited an average CO₂ capacity utilization efficiency of 98.8 ± 0.9%, an electrochemical efficiency of 100%, and an average work of separation of 57.1 ± 1.0 kJ mol⁻¹ CO₂. The slight decrease in energetics compared to the dry conditions (Fig. 6) may be attributed to the reduced impedance of the electrolyte in the presence of water.

Further, we examined the RAMAR performance of 1 under two aerobic conditions (10% CO₂ with 2% O₂ and 21% O₂) for 73 cycles (365 h) and 83 cycles (415 h), respectively (Supplementary Fig. 29 and Fig. 7). Remarkably, 1 maintained its chemical integrity without side products in both WE and CE, confirmed by the crude ¹H NMR after long-term cycling (Supplementary Figs. 30 and 31). At a current of 12 mA (2.4 mA cm⁻²), the former condition (2% O₂) exhibited an average CO₂ capacity utilization of ~90%, a release/capture efficiency of ~100%, and a work of separation of 84.5 ± 9.6 kJ mol⁻¹ CO₂. The latter condition (21% O₂) revealed an average CO₂ capacity utilization of ~83%, a release/capture efficiency of ~100%, and a work of separation of 84.6 ± 30 kJ mol⁻¹ CO₂. These values align with the results obtained in our current step experiment (2 mA cm⁻²) under O₂-free conditions, indicating that O₂ has a minimal impact on RAMAR chemistry (Supplementary Fig. 27c).

a CO₂ readings at the outlet of the WE tank (blue) and CE tank (red), respectively, over 83 repeating cycles for 415-h operation. b, c CO₂ readings for selected capture-release cycles with the corresponding cumulative amount of CO₂ captured/released in WE (b) and CE (c). For each separation cycle, both sorbent tanks were reduced/oxidized at 12 mA (2.4 mA cm⁻²) for 60 min followed by a 90 min rest. d, e Performance metrics of WE (d) and CE (e). WE and CE were composed of 10 ml DMSO solution with 0.5 M 1, 0.05 M 1-ox, 0.1 M HTFSI, 0.5 M MEA, and 0.5 M LiTFSI. Both tanks were stirred and continuously purged with 10% CO₂ + 21% O₂ (balanced with N₂) at a flow rate of 2.5 sccm. Source data for Fig. 7 are provided as a Source Data file.

Full size image

Lastly, we conducted a deep CO₂ removal test on 1 using 25 cm² flow fields, which are five times larger than those used in prior experiments (Fig. 8). Despite the absence of gas contactors facilitating CO₂ absorption, our prototype cell achieved a single pass removal of > 85 % under a 10% CO₂ feed, with an average work of separation of 150.6 ± 17.6 kJ mol⁻¹ CO₂ over 20 cycles (200 h). 1 maintained high purity after cycling, confirmed by crude ¹H NMR (Supplementary Fig. 32). Both WE and CE exhibited faster CO₂ release than capture (evidenced by the sharper gas desorption profile). This phenomenon is attributed to insufficient mass transport in our prototype absorber, which leads to a CO₂ concentration gradient and variations in the amine-CO₂ addition/elimination pathways during capture/release (Fig. 5a). Specifically, the 2H⁺−1CO₂ pathway dominates under extremely low CO₂ concentration during capture, reducing the absorption capacity and kinetics. Conversely, the 1H⁺−1CO₂ pathway prevails under high CO₂ concentration during release, facilitating a more effective desorption process. In practical operation, this challenge can be addressed by enhancing CO₂ mass transport via device and process optimization.

a CO₂ readings at the outlet of the WE tank (blue) and CE tank (red), respectively, over 20 repeating cycles for 200-h operation. b, c CO₂ readings for selected capture-release cycles with the corresponding cumulative amount of CO₂ captured/released in WE (b) and CE (c). For each separation cycle, both sorbent tanks were reduced/oxidized at 60 mA (2.4 mA cm⁻²) for 60 min followed by a 240 min rest. d, e Performance metrics of WE (d) and CE (e). WE and CE were composed of 15 ml DMSO solution with 0.5 M 1, 0.125 M 1-ox, 0.25 M HTFSI, 0.5 M MEA, and 0.5 M LiTFSI. Both tanks were stirred and continuously purged with 10% CO₂ (balanced with N₂) at a flow rate of 1.2 sccm. Flow fields of 25 cm² were used in this experiment. Source data for Fig. 8 are provided as a Source Data file.

Full size image

In conclusion, by harnessing the synergy between RAs and classic amines, our RAMAR mechanism presents an effective strategy to address the practical limitations of many current EMCC absorbents, including limited stability under aerobic conditions, susceptibility to parasitic reactions during electrochemical cycling, high synthetic cost, and incompatibility with ion-exchange membranes. The Wurster-type RAs exhibit a wide electrochemical pK_a swing with a range of > 20 units in organic media, allowing them to effectively and reversibly regenerate amines and potentially other basic absorbents for carbon capture. Through a combination of experimental and DFT studies, we provide fundamental insights into the PCET pathways of RAs, the molecular design principles for finetuning RA properties, and the underlying mechanisms of CO₂–amine interactions in nonaqueous systems. The synthetic versatility of RAs, coupled with their exceptional (electro)chemical stability, enable the development of a symmetric flow-based carbon capture prototype with superior stability (>80 cycles and 400 h) under aerobic conditions (21% O₂). Overall, this work facilitates the process engineering of sustainable and reliable electrochemical amine carbon capture systems for complex gas environments at the molecular level.

Materials

Unless otherwise stated, all the chemicals were purchased from Sigma Aldrich in ACS reagent grade and used without further purification. N¹,N⁴-Diphenylbenzene−1,4-diamine (1) (CAS No.: 74-31-7, purity: 98%) and N¹,N⁴-di(naphthalen-2-yl)benzene−1,4-diamine (2) (CAS No.: 93-46-9, purity: 98%) were purchased from Ambeed. N¹,N⁴-Di-sec-butylbenzene−1,4-diamine (3) (CAS No.: 101-96-2, purity: >98%) was purchased from TCI. DMSO (anhydrous, ≥99.9%) was purchased from Sigma Aldrich for cyclic voltammetry experiments.

Synthesis of N ¹ ,N ⁴-diphenylcyclohexa-2,5-diene−1,4-diimine (1-ox)

To a stirred solution of FeCl₃ • 6H₂O (16.2 g, 60 mmol) in water (500 ml) was added a solution of 1 (5.2 g, 20 mmol) in acetone (100 ml) dropwisely. The mixture was stirred for 5 h at room temperature and the precipitate was filtered, washed with copious amounts of water and cooled diethyl ether. The orange powder was recrystallized in toluene and dried in vacuo at 60 °C to give a brown coloured crystal (3.48 g, 67%). ¹H NMR (400 MHz, DMSO) δ 7.41 (dt, J = 13.3, 7.8 Hz, 4H), 7.24–7.15 (m, 2H), 7.13 (d, J = 2.1 Hz, 1H), 7.04 (dd, J = 10.2, 2.2 Hz, 1H), 6.87 (ddd, J = 12.5, 10.1, 5.4 Hz, 5H), 6.78–6.75 (m, 1H). ¹³C NMR (101 MHz, DMSO) δ 157.83, 157.78, 149.69, 137.64, 136.32, 129.07, 129.04, 125.27, 125.05, 125.00, 124.31, 120.35, 120.21. The NMR agrees well with previous literatures^52,53.

Synthesis of N ¹ ,N ⁴-di-sec-butylcyclohexa-2,5-diene−1,4-diimine (3-ox)

To a stirred solution of 3 (1.8 ml, 7.68 mmol) in acetone (100 ml) was added silver oxide (2.66 g, 5.76 mmol) in portions. The mixture was stirred at room temperature for 1 h and filtered through a Celite cake. The filtrate was dried in vacuo to give a dark red oil, which solidified in ambient condition after a few days to afford a red crystalline solid (1.44 g, 85%). ¹H NMR (400 MHz, DMSO) δ 7.27–6.99 (m, 2H), 6.84–6.58 (m, 2H), 3.85 (dd, J = 12.5, 6.2 Hz, 2H), 1.64–1.40 (m, 4H), 1.09 (d, J = 6.1 Hz, 6H), 0.77 (t, J = 7.4 Hz, 6H). ¹³C NMR (101 MHz, DMSO) δ 156.96, 156.73, 137.59, 135.24, 121.99, 120.69, 56.65, 56.57, 30.78, 30.75, 21.97, 21.84, 10.77.

Synthesis of diethyl 2,5-bis(phenylamino)terephthalate (4)

To a mixture of diethyl 2,5-dioxocyclohexane−1,4-dicarboxylate (512.5 mg, 2 mmol) and p-toluenesulfonic acid (100 mg, 0.58 mmol) in toluene (20 ml) was added aniline (0.548 ml, 6 mmol) under argon. The mixture was refluxed for 1 day using a Dean-Stark trap before being allowed to cool to room temperature. The organic phase was washed with water for three times, dried over Na₂SO₄, and evaporated to dryness to give an orange powder. The crude product was recrystallized in dichloromethane/hexane to afford an orange crystal. The orange product was dissolved in tetrahydrofuran (50 ml). The solution was added with trifluoroacetic acid (0.05 ml) and heated at 50 °C for 8 h in the air. The tetrahydrofuran solvent was removed in vacuo and the crude product was dissolved in ethyl acetate. The organic layer was washed with sat. NaHCO₃ (aq) for three times and brine and dried over Na₂SO₄. The solvent was evaporated to dryness to give a red powder (678 mg, 84%). ¹H NMR (400 MHz, DMSO) δ 8.56 (s, 2H), 7.81 (s, 2H), 7.30 (t, J = 7.7 Hz, 4H), 7.10 (d, J = 7.6 Hz, 4H), 6.95 (t, J = 7.2 Hz, 2H), 4.24 (dd, J = 14.0, 7.0 Hz, 4H), 1.21 (t, J = 7.1 Hz, 6H).

Electrochemical measurements

Electrochemical measurements were performed with a BioLogic VSP potentiostat from BioLogic Science Instruments. Cyclic voltammetry (CV) was measured using a standard three-electrode cell with a glassy carbon electrode (3 mm diameter) as the working electrode, a platinum wire as the counter electrode, and a silver wire as the quasi-reference electrode, with ferrocene as the internal reference. In a standard CV test, redox-active molecules (5 mM) were dissolved in anhydrous DMSO with 100 mM NBu₄PF₆ as the supporting electrolyte. CV were typically recorded at a scanning rate of 50 mV s^–1. The electrochemical data were gathered and analysed by EC-lab V11.50.

Bulk electrochemical oxidation and reduction of 1

Constant current bulk electrolysis of 1 was carried out using a setup consisting of a three-neck round-bottom flask (RBF), with the three necks for the working, counter, and reference electrodes, respectively. Carbon felt (CT GF030; Fuel Cell Store) was used as the working electrode and a silver wire was used as the reference electrode. The counter electrode was separated from the solution 1 with a fritted electrode chamber (MR−1196; Bioanalytical Systems). LiClO₄ (0.5 M) in DMSO-d₆ was used as the electrolyte. A piece of carbon felt (1 cm²) coated with lithium-titanate-oxide (LTO) (64 mg) was used as the counter electrode, which was immersed in the electrolyte (1 ml). In a standard condition, 1 (26 mg, 0.1 mmol) was stirred and oxidized in the presence of monoethanolamine (12.1 μl, 0.2 mmol) at a constant current of 1.5 mA in DMSO-d₆ (4 ml, 0.5 M LiClO₄) under continuous 10% CO₂ bubbling for 3.6 h. After the oxidation was complete, 0.8 ml of the electrolyte in the working electrode chamber was withdrawn for crude NMR analysis without further purification. Subsequently, the solution left in the RBF was stirred and reduced at a constant current of −1.5 mA until the voltage was below −1.6 V under continuous 10% CO₂ bubbling. Similarly, another 0.8 ml of the electrolyte in the working electrode chamber was withdrawn for crude NMR analysis without further purification.

Treatment of Nafion membrane²³

The Nafion membranes (Nafion 212, 2.5 × 3 cm², 50 μm in thickness) were heated in boiling 3% H₂O₂ (aq) for 1 h. The membranes were then heated in 0.25 M sulfuric acid (aq) for 1 h and cleaned in deionized water for 30 min (twice) at 100 °C. The membranes were heated in 0.25 M NaOH (aq) for 1 h and cleaned in deionized water for 30 min (twice) at 100 °C. The membranes were dried under vacuum at 80 °C for 2 d and stored in DMSO.

EMCC symmetric flow cell

In a standard setup, two scintillation vials (20 ml) with septum caps, serving as the sorbent tank of WE and CE, were continuously purged with CO₂ feed gas (balanced with N₂) at a controlled flow rate using Alicat mass flow controllers. Simulated flue gas conditions were mimicked using a 10% CO₂ and 2% O₂ mix, balanced by argon. Ambient oxygen conditions were mimicked using a 10% CO₂ and 21% O₂ mix, balanced by N₂. CO₂ levels exiting the WE and CE tanks were continuously monitored using an infrared-based CO₂ sensor (SprintIR-W 100%), with data recorded via Labview 2021. The flow cell incorporated two graphite plates with 5 or 25 cm² interdigitated flow fields, pressing against three pieces of carbon paper electrodes (Sigracet 28 AA) on each side of the graphite plates. WE and CE electrolytes were circulated at a flow rate of 10- or 40-ml min⁻¹ (for 5 or 25 cm² flow fields, respectively) through a commercial flow cell (Scribner) by a two-channel peristaltic pump (Masterflex). A treated Nafion 212 membrane, flanked by Celgard 3501, was placed between the carbon electrodes. The cell was sealed by Kalrez fluoropolymer elastomer gaskets (0.51 mm thick). CO₂ capture/release was conducted in constant current mode. The cycling protocols of each experiment are provided in the corresponding figure captions. The temperature of the flow cell was kept at 25 °C using a pair of cartridge heaters connecting to a Digi-Sense TC6000 temperature controller.