Lower critical solution temperature (LCST)-mediated direct air capture of carbon dioxide with high capacity and low regeneration energy

Abstract

Direct air capture of carbon dioxide has been limited by the energy-intensive thermal regeneration of adsorbents imposed by the low CO₂ uptake capacity, high regeneration temperature and parasitic water evaporation. Herein, we demonstrated a class of thermo-responsive CO₂ adsorbents—PEI impregnated PNIPAM-grafted cellulose network (TRCNF/PEI)—achieving a high CO₂ uptake of 2.82 mmol g⁻¹ at 25 °C and 4.33 mmol g⁻¹ at 5 °C from air and low regeneration temperature below 65 °C. The thermo-responsive adsorbent features a lower critical solution temperature (LCST) phase transition that enables a swell-adsorption/collapse-desorption mechanism. At room temperature, TRCNF/PEI is at a hydrophilic swollen state, effectively exposing the PEI amine sites for efficient CO₂ capture. Above LCST, it transitions to a hydrophobic collapsed state. Water is expelled in the liquid phase, destabilizing bicarbonate intermediates and thereby driving rapid CO₂ desorption. This obviated the latent heat penalty of water, leading to an unprecedentedly low regeneration thermal energy demand of 4.36 GJ ton⁻¹_CO2 that can be satisfied solely by low-grade waste heat. Unlike zeolite-activated carbons and MOFs that are vulnerable to high humidity, where liquid water inhibits CO₂ uptake due to competitive adsorption, TRCNF/PEI leverages water to enhance adsorption and exhibits exceptional cyclic stability. Notably, the CO₂ uptake of TRCNF/PEI further increases at lower temperatures (4.33 mmol g⁻¹ at 5 °C), demonstrating its carbon capture potential in cold areas. The unique LCST phase transition presents a transformative strategy to couple DAC with vast low-grade industrial waste heat, significantly advancing the energy efficiency and durability of carbon removal technologies.

---

Broader context

Direct air capture (DAC) is increasingly recognized as an essential climate mitigation solution by actively removing the excessively emitted carbon dioxide. However, its large-scale deployment remains constrained by the high energy demand of adsorbent regeneration, limited uptake capacity, and parasitic water evaporation. Many existing porous adsorbents (e.g., zeolites and MOFs) also lose performance in the presence of liquid water due to competitive adsorption. Here, we present a thermo-responsive DAC scheme based on a lower critical solution temperature (LCST) phase transition, enabling a swell-adsorption/collapse-desorption mechanism that combines high CO₂ capture from air with low regeneration energy. The collaborative water–CO₂ adsorption/desorption along with the LCST swing transforms water from a challenge into privilege, thereby reducing thermal penalties. Enhanced CO₂ uptake at lower temperatures further supports applicability in cold areas. More broadly, this LCST-enabled concept offers insights into designing next-generation adsorbents to advance scalable carbon dioxide removal technologies and highlights a route to coupling DAC with omnipresent low-grade waste heat sources.

1. Introduction

An excess of thousands of giga metric tons of CO₂ (Gt_CO2) has been emitted into the atmosphere since industrialization.^1,2 With less than three decades to achieve carbon neutrality, in addition to rapid decarbonization, it is now imperative to actively remove ∼1 Gt CO₂ per year before 2050 to mitigate global temperature overshoot.^3–5 Direct air capture (DAC) of carbon dioxide (CO₂), which selectively removes CO₂ from the atmosphere, is deemed as a pivotal negative emission technology to achieve this goal, but it has been long impeded by the prohibitive land, energy, water and cost scaling limitations.^6,7 Conventional amine-grafted solid adsorbents (e.g., amine-grafted silicas or certain metal–organic frameworks) parallelly adsorb a significant amount of water (∼18 mmol_water g⁻¹) and require a high temperature steam at 80–120 °C to overcome the heat of adsorption of CO₂ and water, sensible heat, and parasitic heat losses.⁸ Furthermore, the co-adsorbed water in the traditional amine-appended adsorbents often leads to severe amine leaching and oxidative degradation during high-temperature steam scrubbing, thereby drastically limiting the adsorbent lifetime.^9,10 As a result, current DAC technologies consume 5–12 GJ of energy to capture 1 ton of CO₂ from air, 70% of which is in the form of heat,^11–13 translating into 10 quads of energy to achieve the aforementioned milestone of removing 1 Gt CO₂ per year by 2050 (U.S. DOE, 2025).^14–16 This is equivalent to roughly 12% of global electricity consumption and is near-impossible, placing the 2050 target under serious threat.¹⁷ To make matters worse, high regeneration temperature precludes the utilization of omnipresent low-grade waste heat sources (40–65 °C) such as from data centers,^18,19 power stations,²⁰ and so on.^21,22 Most DAC plants today are powered by grid electricity or natural gas,²³ leading to a high carbon intensity of ∼0.73–1.1 kg_CO2 emitted per unit kg of CO₂ captured, resulting in the global average carbon footprint of grid electricity to be ∼486 g_CO2 kWh⁻¹ and that of natural gas to be ∼450 g_CO2 kWh⁻¹.^17,24 To unlock the full potential of DAC, we must explore economically affordable and sustainable heat sources for thermal regeneration, such as low-grade waste heat from data centers and industrial processes at 40–65 °C, which remains inaccessible to most existing adsorbents.

Addressing these thermal and stability challenges necessitates a paradigm shift at the material level. Thermo-responsive polymers, notably poly(N-isopropylacrylamide) (PNIPAM), have attracted extensive attention due to their unique lower critical solution temperature (LCST) phase transition. Below the LCST (∼32 °C), PNIPAM is in a hydrophilic state and exhibits affinity to water; upon heating above the LCST, it undergoes a sharp phase transition into a hydrophobic state and rapidly collapses. Water is released in the liquid phase without evaporative latent heat penalty. This reversible, temperature-driven steric reconfiguration has been widely explored to facilitate drug delivery^25–27 and smart surface fabrication^28–32 and in tissue engineering^33–35 and point-source carbon capture.^36,37 Recent studies revealed collaborative adsorption between CO₂ and water, i.e., water molecules promote CO₂ uptake kinetics and capacity due to enhanced proton transfer and the formation of bicarbonate instead of ammonium carbamate compounds.⁸ It is thus possible to regulate the CO₂ uptake kinetics, capacity and desorption temperature by tuning the hydrophilicity of adsorbents, which warrants the development of thermo-responsive CO₂ adsorbents for DAC.

Herein, we report a class of thermo-responsive adsorbents for DAC that were synthesized from hyperbranched cellulose nanofibers (CNF) grafted with poly(N-isopropylacrylamide) (PNIPAM) and cross-linked with polyethyleneimine (PEI). This architecture leverages the unique LCST-mediated hydrophilic–hydrophobic transition to facilitate CO₂ uptake below LCST and rapid CO₂ desorption under mild heating. At ambient temperature (below LCST), the adsorbent is hydrophilic and swollen, exposing PEI sites to adsorb CO₂ and form carbonate groups, rendering a high CO₂ uptake of 2.81 mmol g⁻¹ from air at 25 °C. Upon heating to 65 °C (above LCST), it transitions to the hydrophobic state and release water in the liquid phase, rapidly desorbing CO₂ by destabilizing the bicarbonate species without the latent heat penalty of water. By harnessing the LCST hydrophilic–hydrophobic transition, our adsorbent exhibited a unprecedentedly high CO₂ uptake yet a substantially lowered thermal energy demand of 4.36 GJ ton⁻¹_CO2, exclusively relying on low-grade waste heat below 65 °C. Crucially, unlike conventional amine-grafted solid adsorbents, which are prone to amine leaching and oxidation in the presence of water, the crosslinked architecture of our adsorbent firmly anchors amine, guaranteeing both high water tolerance and exceptional durability over 500 cycles. This work therefore presents a robust and energy-efficient adsorbent material that paves the way for scalable and energetically viable direct air capture.

2. Results and discussion

2.1. Synthesis and characterization of the adsorbent

The design strategy for TRCNF/PEI is illustrated in Fig. 1. First, thermo-responsive monomer NIPAM was grafted on the cellulose nanofibers (CNFs) via Ce⁴⁺-initiated radical polymerization.^38,39 The thermo-responsive PNIPAM-grafted cellulose nanofibers (TRCNFs) was then in situ crosslinked with polyethyleneimine (PEI) via epichlorohydrin (ECH) into the TRCNF/PEI adsorbent. More details of the synthesis of the TRCNF/PEI adsorbent can be found in the Supporting information S1 section. Structural characterizations via SEM and EDS reveal the uniform morphology and elemental distribution, respectively (Fig. S2). Below the LCST, the cellulose network remains in a swollen, hydrophilic state that favors the CO₂ adsorption of amine sites due to enhanced proton transfer. When heated above the LCST, the NIPAM segment transitions to a hydrophobic state and undergoes a coil-to-globule contraction. Water molecules are repelled from the cellulose network in liquid phase (Fig. S3 and Video S1), which facilitates CO₂ desorption by destabilizing the bicarbonate group. Released liquid water could be easily separated and collected before evaporation under continuous hot gas flow (Fig. S4a and Video S2), with no detectable leaching of amine or polymer from the adsorbent (Fig. S4b and c). Therefore, the latent heat penalty of water could be avoided. This temperature-induced hydrophilic–hydrophobic switch of TRCNF/PEI is evidenced by the different water sorption performances at below LCST (142% at 25 °C, 95% RH) and above LCST (0.05% at 65 °C, 95% RH) (Fig. S5). The TRCNF and PEI mass ratio was engineered to maximize the CO₂ capture efficiency and exhibited the optimal performance at a TRCNF : PEI mass ratio of 1 : 22 (calculated based on their exact dry solid masses, equivalent to the molar ratio of 1 : 3.76) (Fig. S6). A lower PEI loading leads to the deficit of active amine sites for chemisorption, while excessive PEI loading forms dense crosslinked network that impedes CO₂ diffusion, which is evidenced by its lower equilibrium swelling ratio (480% compared to 600% for TRCNF/PEI (1 : 22), as shown in Fig. S7a.^40,41 The N₂ adsorption/desorption isotherm analysis shows that the BET specific surface areas of TRCNF/PEI (1 : 22) is larger than that of TRCNF/PEI (1 : 24) (Fig. S7b). Therefore, the TRCNF : PEI mass ratio of 1 : 22 was chosen for all subsequent experiments.

Fig. 1 Synthesis and working principle of the thermo-responsive TRCNF/PEI adsorbent for the energy-efficient direct air capture (DAC) of CO₂. The hyperbranched cellulose nanofibers (CNF) are grafted with temperature-responsive PNIPAM and crosslinked with PEI to form a robust three-dimensional network. Below the LCST (at 25 °C), the hydrophilic/swollen state facilitates CO₂ uptake; above the LCST (at 65 °C), the hydrophobic/collapsed state triggers rapid CO₂ desorption and water expulsion in the liquid phase. The low regeneration temperature and evaporation-free desorption warrant integration of DAC with low-grade waste heat to minimize thermal energy consumption.

The structure and chemical composition of the adsorbent were rigorously characterized by various spectroscopies. The solid-state ¹³C nuclear magnetic resonance (NMR) spectroscopy of pristine CNF revealed the typical carbon signals of cellulose I (C1–C6), with a distinct peak at δ = 179 ppm (Peak 7) corresponding to the carbonyl carbon of the carboxyl groups (–COOH) (Fig. 2a).⁴² After the successful grafting with temperature-responsive PNIPAM and crosslinking with PEI, a new near-baseline hump peak at 165 ppm appeared, corresponding to the carbonyl carbons (C=O) of amide and carboxyl groups (Fig. 2b). A series of intense peaks emerged in the high-field region (20–60 ppm), corresponding to the methylene carbons (–CH₂–CH₂–NH–) of the PEI backbone and the aliphatic carbons (isopropyl and main chain methylene) of the PNIPAM side chains. The successful grafting of PNIPAM was also validated by the ¹H NMR analysis (Fig. S8), which identified characteristic peaks at δ = 0.8 ppm (methyl protons of isopropyl) and δ = 4.2 ppm (methine proton of isopropyl). The ¹H NMR signals at δ = 3.2 ppm (methylene protons) validated the presence of PEI chains.

Fig. 2 Spectroscopic analysis and thermal stability of TRCNF/PEI. ¹³C solid-sate NMR spectra of (a) pristine CNF and (b) TRCNF/PEI, highlighting the characteristic signals of cellulose backbone, grafted PNIPAM and crosslinked PEI. Inset: molecular structure of the corresponding repeating unit. (c) FT-IR spectra of CNF and TRCNF/PEI. (d) Full scan XPS spectra of CNF and TRCNF/PEI. (e) High-resolution N 1s XPS spectra for TRCNF/PEI, along with peak deconvolution to quantify the chemical states of different amine groups. (f) Thermogravimetric analysis of CNF and TRCNF/PEI samples under nitrogen atmosphere. The samples were first heated at 90 °C for 60 min to ensure the complete removal of adsorbed moisture prior to the thermal stability assessment.

The successful integration of organic amines and thermo-sensitive polymers onto the cellulose nanofibers was cross-validated by Fourier transform infrared spectroscopy (FT-IR, Vertex 70V + Hyperion II) (Fig. 2c). Pristine CNF exhibits a broad O–H stretching band at ∼3300 cm⁻¹ and characteristic COO⁻ IR peaks at 1600–1700 cm⁻¹.⁴³ With PNIPAM grafted (TRCNF), the fingerprint peaks appear at 1655 cm⁻¹ (amide I band, C=O stretching) and 1544 cm⁻¹ (amide II band, N–H bending) corresponding to the amide groups in PNIPAM (Fig. S9). Furthermore, the distinct doublet at 1387 cm⁻¹ and 1368 cm⁻¹ corresponds to the symmetric bending vibrations of the methyl groups in the N-isopropyl segment. Following the loading of PEI (TRCNF/PEI), prominent adsorption peaks emerged at 2930 and 2800 cm⁻¹, which are assigned to the C–H stretching vibrations of the methylene (–CH₂–) units in the PEI backbone. The intensity of these peaks is significantly higher compared to TRCNF, verifying the incorporation of PEI chains within the 3D intertwined network. Additionally, the broad band centered at ∼3300 cm⁻¹ widens, attributed to the overlapping of N–H stretching vibrations from PEI amines with the O–H bands of cellulose, suggesting the formation of a dense hydrogen-bonding network. Meanwhile, the strong characteristic peak of the cellulose skeleton (C–O–C stretching) at 1030–1060 cm⁻¹ was well-preserved, indicating that the multi-step modification did not compromise the structural integrity of the CNF framework.

The active amine sites for CO₂ chemisorption were quantified by X-ray photoelectron spectroscopy (XPS, ULVAC-PHI VersaProbe4). The PEI grafting was noted by the distinct N 1s emission peak at ∼400 eV, which was absent in pristine CNF (Fig. 2d).⁴⁴ Quantitative analysis indicates that the TRCNF/PEI surface is predominantly composed of C (61.6 at%), N (22.5 at%), and O (15.9 at%) (Fig. S10). The high nitrogen content confirms active amine sites in TRCNF/PEI. The high-resolution N 1s spectrum (Fig. 2e) was deconvoluted into four distinct peaks that belonged to amide groups (399.99 eV, 13.8%), tertiary amines (3°, 398.90 eV, 23.0%), primary amines (1°, 398.38 eV, 51.1%), and secondary amines (2°, 397.90 eV, 12.1%). The active amine sites, i.e., 1°, 2°, and 3°, together as the primary chemisorption sites for CO₂, account for 86.2% of the total nitrogen species. Based on these quantitative XPS data, the theoretical total amine density is estimated to be 14.82 mmol g⁻¹ and is dominated by primary amines (8.78 mmol g⁻¹), suggesting the potential of TRCNF/PEI for high-capacity CO₂ capture. We also analyzed the C 1s and O 1s spectra to corroborate the successful grafting of PNIPAM and the subsequent integration of the PEI network onto the cellulose matrix (Fig. S11).

The thermal stability of TRCNF/PEI was evaluated by thermogravimetric analysis (TGA, TA 5500) (Fig. 2f). The curve remains stable in the range of 100–280 °C, proving that the adsorbent was thermal stable at typical regeneration temperatures of <65 °C. A sharp mass loss occurred at 280 °C due to the pyrolysis of the PEI and PNIPAM organic chains. The mass loss above 280 °C was found substantially higher in TRCNF/PEI than in pristine CNF, which confirms the high loading of PEI and PNIPAM in the adsorbent.

2.2. Performance of LCST-mediated CO₂ adsorption–desorption

The CO₂ adsorption and desorption performance of the TRCNF/PEI adsorbent for 400 ppm CO₂ stabilized in 1 bar N₂ was evaluated by fixed-bed breakthrough experiments. The details of the experimental procedures can be found in the Supporting Information. In brief, 0.5 g of dry adsorbent was hydrated to achieve 600 wt% water content and was loaded into an adsorption column of 1 cm in diameter, with the simulated gas stream flowing through it at 25 °C. The CO₂ adsorption performance was obtained by monitoring the outlet CO₂ concentration using an online mass spectrometry (MS, INFICON), and the cumulative CO₂ uptake was calculated by the area between the inlet concentration baseline and outlet breakthrough curve.

The CO₂ uptake was found sensitive to the initial hydration state of the adsorbent (Fig. 3a). The dry adsorbent (0% water content) barely adsorbed CO₂ despite fast saturation at the breakthrough time of 0.05 h (Fig. 3b). In distinct contrast, the pre-hydrated adsorbent (initial water content of 600 wt%) exhibited an excellent adsorption capacity of 2.82 mmol g⁻¹. It showed a typical S-shaped adsorption curve, reaching 1.41 mmol g⁻¹ (50% saturation) at 14.2 h and 2.26 mmol g⁻¹ (80% saturation) at 23.1 h. The S-shaped breakthrough profile indicates potential diffusional resistances in overall CO₂ uptake kinetics. This distinct difference highlights a key feature of our material: unlike conventional amine-grafted solid adsorbents (e.g., amine-grafted metal–organic frameworks such as Mg₂(dobpdc)(3-4-3)) or certain zeolite-based materials) where water can hinder CO₂ capture, the presence of water facilitates CO₂ uptake in TRCNF/PEI. The presence of pre-loaded water promotes both CO₂ diffusion via adsorbent swelling and CO₂ chemisorption by favoring the formation of bicarbonate intermediates. It has been long acknowledged that under dry conditions, CO₂ binds with primary/secondary amines through a bimolecular mechanism to form carbamates (R–NHCOO⁻), which requires two amine groups to fix one CO₂, whereas in the presence of water molecules, which act as efficient proton-transfer mediators, the formation of bicarbonate intermediates (R–NH₃⁺ and HCO₃⁻) is kinetically and thermodynamically favored, enabling one amine group to independently capture one CO₂. This not only improves amine utilization efficiency but also activates otherwise inert tertiary amine sites. Prior studies have also reported the enhanced CO₂ uptake kinetics due to enhanced proton transfer by water molecules.⁸ In addition, owing to the hydrophilic nature of TRCNF/PEI below LCST, the adsorbent remains hydrated and swollen in humid environments, expanding the originally dense cellulose-based network into an open structure. CO₂ diffusion is facilitated in enlarged internal pores, and the sterically hindered amine groups are exposed to CO₂ molecules. Multiple identical samples were prepared to be pre-hydrated at different levels for parametric study. The water-facilitated CO₂ adsorption can be further validated by the progressive CO₂ uptake from 0.05 mmol g⁻¹ in dry state (0% water content) to 2.62 mmol g⁻¹ at 500% water content (Fig. S12). CO₂ concentration-dependent adsorption found that CO₂ adsorption in TRCNF/PEI was reaction dominated, validated by the high CO₂ uptake of ∼6.5 mmol g⁻¹ with a reduced breakthrough (<3 min) in 10% CO₂ stabilized in 1 bar N₂ (Fig. S13).

Fig. 3 DAC performance of the TRCNF/PEI adsorbent. (a) CO₂ breakthrough curves of TRCNF/PEI with different water contents (600 wt% and 0%) under simulated air (400 ppm CO₂ in N₂) at 25 °C. (b) Corresponding cumulative CO₂ adsorption capacity derived from (a). (c) CO₂ desorption profile of the saturated sample under 1 bar pure N₂ flow at 65 °C. (d) Cyclic stability test showing CO₂ uptake over 15 adsorption (25 °C)-desorption (65 °C) cycles. Prior to each adsorption cycle, the adsorbent was rehydrated to its initial water content (600 wt%) to ensure a consistent starting state.

The critical advantage of TRCNF/PEI over conventional amine-grafted solid adsorbents lies in its low regeneration temperature and water evaporation-free CO₂ desorption. As illustrated in Fig. 3c, the adsorbent completely desorbs CO₂ under mild heating at 65 °C (above LCST, by purging 65 °C dry 1 bar N₂ stream) as opposed to the 80–120 °C hot steam routinely used in PEI-grafted adsorbents.^10,19 The amount of desorbed CO₂ was calculated by integrating the downstream CO₂ concentration over time. At 65 °C, 5 hours was needed to desorb all the adsorbed CO₂. The adsorbent underwent 15 adsorption–desorption cycles by the LCST temperature swing (Fig. 3d) and maintained a consistently high CO₂ uptake of 2.82 mmol g⁻¹ for all cycles. This exceptional cyclic stability highlights the inherent robustness of our crosslinked architecture, featuring the covalent immobilization of amine functionalities that effectively prevent PEI leaching even in the presence of water. This unique feature allows the material to harness water for enhanced CO₂ capture without compromising its long-term performance. Overall, the consistent adsorption–desorption performance under both simulated air (400 ppm CO₂) and concentrated CO₂ conditions underscores the functionality and durability of TRCNF/PEI, establishing a strong basis for practical direct-air-capture applications.

2.3. Mechanistic origin of LCST-mediated regeneration

To unravel the molecular mechanism of LCST phase transition-induced desorption, in situ diffuse reflectance infrared Fourier transform spectroscopy (DRIFTS, Thermo IS 50) and dynamic light scattering (DLS, Malvern Zetasizer Ultra) analyses were performed (see the Experimental section for details). For DRIFTS measurements, the dry adsorbent was placed in 400 ppm humid CO₂ (∼100% relative humidity, achieved by bubbling CO₂ through water) in 1 bar N₂ at 25 °C, and the evolving FTIR spectra were recorded during CO₂ adsorption. After saturation, the adsorbent was progressively heated up under a dry N₂ flow for CO₂ desorption, and the FTIR spectra were recorded at each temperature after equilibrium. In parallel, DLS measurements were performed on pre-hydrated TRCNF/PEI samples (600 wt% water content) to track the particle size evolution during the temperature-induced phase transition. During CO₂ adsorption at 25 °C, the signature peaks of CO₂ adsorption at 1715 cm⁻¹ (C=O stretching of bicarbonate) and water adsorption at 3640 cm⁻¹ (O–H stretching of the adsorbed water) were found to synchronously intensify over time (Fig. 4a and b).^45–49 The synergy between the two peaks manifests the formation of water-stabilized bicarbonate species. To provide robust evidence of the chemisorption nature of CO₂ uptake, we performed ¹³C solid-state NMR spectroscopy on our adsorbent. The new peaks at 165.2 ppm after CO₂ adsorption manifests the formation of water-stabilized bicarbonate species (Fig. S14). During desorption under dry N₂ purge (Fig. 4c and d), both peaks decayed concurrently and vanished eventually as temperature increased from 25 °C to 65 °C. To better elucidate the collaboration between LCST transition and CO₂ desorption, we correlated the temperature, normalized IR peak intensities and average particle diameter measured by DLS (the red line in Fig. 4d). The intensity of all IR peaks was normalized to that at 25 °C. Upon heating above the LCST (32 °C), the particle size of TRCNF/PEI gradually shrunk from ∼6 µm to ∼1 µm as temperature increased from 32 °C to 55 °C, indicating that the adsorbent transitioned to the hydrophobic collapsed state. As the cellulose-based network collapsed, water was fully expelled from the adsorbent, which was indicated by the vanishing O–H stretching peak of the adsorbed water (3640 cm⁻¹). The loss of water then triggers the destabilization of bicarbonate species, leading to rapid CO₂ desorption, evidenced by the vanishing of the C=O stretching peak of bicarbonate, following the O–H stretching peak of adsorbed water. The schematic in the lower panel of Fig. 4d illustrates this cascade: the hydrophobic collapse of cellulose network expels the solvated water, which in turn destabilizes the bicarbonate intermediates, ultimately forcing CO₂ desorption. This physical mechanism explains the negligible CO₂ affinity (0.05 mmol g⁻¹) at 65 °C and rationalizes the high regeneration efficiency achieved under mild heating (Fig. S15).

Fig. 4 Molecular mechanism of LCST-triggered CO₂ desorption. (a) Time-dependent in situ DRIFTS spectra during CO₂ adsorption at 25 °C and (b) corresponding normalized intensity evolution of peaks at 3640 cm⁻¹ (adsorbed water) and 1715 cm⁻¹ (bicarbonate) to that at t = 0. (c) Temperature-dependent in situ DRIFTS spectra during desorption and (d) correlation between temperature-dependent structural collapse and CO₂ desorption: average particle diameter measured by DLS (red line) and normalized IR peak intensities of adsorbed water (blue line) and bicarbonate (yellow line) as a function of regeneration temperature, with schematic of the LCST-induced desorption mechanism (lower row). (e) Radius of gyration (R_g) and (f) solvent-accessible surface area (SASA) of TRCNF/PEI over a 0–50 ns period at 25 °C and 65 °C. (g) Representative MD simulation snapshots of the TRCNF/PEI molecular chain at 65 °C, visualizing the conformational transition from a solvated coil (0 ns) to a collapsed globule (50 ns).

The LCST-induced adsorbent collapse was also validated by molecular dynamics (MD) simulations (see the supplementary information Section S11 for details). In brief, we tracked the dynamic trajectories of the hydrophilic domains (comprising the CNF and PEI backbones) and the hydrophobic isopropyl segments of the PNIPAM side chains at 25 °C and 65 °C. We used the radius of gyration (R_g) and solvent accessible surface area (SASA) to quantify conformational evolution (Fig. 4e and f). At 25 °C (below LCST), the PNIPAM segments of TRCNF/PEI formed stable hydrogen bonds with water molecules. Consequently, the chain maintained an extended, well-solvated conformation, characterized by a stable R_g (∼23 Å) and a relatively high SASA throughout the simulation. Snapshots in the supplementary information (Fig. S16) confirm that the cellulose-based network remains expanded with no significant aggregation at 25 °C. In contrast, at 65 °C (above LCST), intensified thermal vibrations first break the hydrogen bonds between water molecules and the PNIPAM amide groups, a process marked by a temporary kinetic expansion and an R_g spike within the initial 20 ns (Fig. 4e). Freed from these stabilizing hydrophilic interactions, the chain's behavior is then dominated by hydrophobic forces among the isopropyl segments. This shift triggers an LCST-driven collapse—a thermodynamically favorable aggregation of adsorbent that expelled bound water—ultimately resulting in a stable, compact state characterized by a lower R_g and a ∼14% reduction in SASA to ∼12 000 Ų (Fig. 4f). This thermally induced collapse is visually corroborated by the snapshots in Fig. 4g, where the cellulose-based network transitions from a loose coil (0 ns) to a compact globule structure (50 ns). These simulation results show similar trends to the DLS and in situ DRIFTS tests, together with macroscopic optical observations supporting the thermally induced hydrophobic coil-to-globule collapse (Fig. S17 and Video S3). While this significant volume shrinkage, crucial for adsorbent regeneration, could cause voids and airflow channeling in packed beds, it is not an insurmountable engineering challenge. Practical DAC reactor designs can effectively ensure macroscopic bed volume stability and thus fully leverage the material's advantages by integrating the active hydrogel into structured contactors such as coatings on honeycomb monoliths, open-cell foams, or hollow fiber surfaces. This approach ensures stable operation and efficient regeneration in practical DAC applications.^50–52

2.4. Levelized energy analysis and cyclic stability

The feasibility of TRCNF/PEI for practical DAC deployment is highlighted by its exceptional energy efficiency and durability. Our material achieves exceptional CO₂ uptake capacity for DAC (∼2.82 mmol g⁻¹) at a significantly reduced regeneration temperature (65 °C) compared to the state-of-the-art amine-based adsorbents (Fig. 5a, for detailed data see Table S1). To quantify the energetic advantage of TRCNF/PEI, we decoupled the regeneration energy components using simultaneous TGA-DSC analysis equipped with a DSC sensor (Fig. S18). The total regeneration energy (Q_total) was deconvoluted into three components (Fig. 5b): the sensible heat required to heat the sorbent itself (Q_sens), the enthalpy of CO₂ desorption (ΔH_CO2), and the heat associated with water released by the LCST effect (Q_H2O).^53–55 Q_H2O generally constitutes a major energy budget of 3–10 GJ ton_CO2⁻¹ to conventional DAC using MOFs/COFs due to the latent heat penalty for water evaporation.⁸ However, different from water desorption in vapor phase in conventional amine-grafted solid adsorbents (e.g., amine-grafted metal–organic frameworks), our adsorbents release water in the liquid phase in LCST transition and avoids the latent heat penalty (Fig. S19). Consequently, the energy required for water expulsion is reduced to 1.54 GJ ton⁻¹_CO2. It is important to note that energy estimations, including those in Table S2, generally exclude auxiliary components such as fans. In addition, the levelized sensible heat and heat of CO₂ desorption are reduced due to the high CO₂ uptake. The levelized energy of CO₂ capture for TRCNF/PEI reaches an unprecedently low level of 4.36 GJ ton⁻¹_CO2, which is 25–65% lower than commercial liquid MEA (10.7 GJ ton⁻¹_CO2)^54,56 or solid adsorbents like Mg₂(dobpdc)(3-4-3) (11.4 GJ ton⁻¹_CO2),⁸ and COF-999 (5.9 GJ ton⁻¹_CO2).⁵⁷ The levelized energy of TRCNF/PEI is close to the enthalpy of amine (2 GJ ton⁻¹_CO2) and is only 3.9 GJ ton⁻¹_CO2 higher than the thermodynamic limit of separating 400 ppm CO₂ from air (i.e., reaching 10.6% second-law efficiency). More importantly, the thermal energy for TRCNF/PEI can be sufficiently satisfied by omnipresent low-grade heat of temperatures <65 °C, while contemporary DAC pilot plants rely on high-grade energy like electricity and natural gas.

Fig. 5 Levelized energy analysis and cyclic stability of TRCNF/PEI. (a) Performance comparison with the state-of-the-art amine-based adsorbents regarding CO₂ capacity and regeneration temperature. (b) Deconvolution of levelized energy consumption into three components: sensible heat (green), CO₂ desorption (blue), and water desorption (red). The regeneration energy of TRCNF/PEI is compared with theoretical limits and that of the hydrophilic liquid MEA, hydrophilic Mg₂(dobpdc)(3-4-3), hydrophobic COF-999 and the Climeworks company. References provided in Table S2. (c) TGA-derived adsorption curves for TRCNF/PEI before (blue) and after 500 cycles (red). Each of the 500 intermediate cycles employed a 5-hour adsorption and 2-hour desorption step to accelerate cyclic aging, while the 0th cycle and 501th cycle (shown) used a 25-hour adsorption step to reach equilibrium and enable accurate capacity comparison. Inset: Detailed weight change profiles during 470–500 cycles under simulated DAC conditions (adsorption: 25 °C, 0.04% CO₂, ∼100% RH; desorption: 80 °C, 100% N₂, 0% RH).

The materials were stress-tested at the desorption temperature of 80 °C, which was much higher than the operational temperature. Therefore, the predicted lifetime represents a conservative estimation. TRCNF/PEI was stress-tested under accelerated swing between 25 °C, 0.04% CO₂, ∼100% RH for 5 h for adsorption and 80 °C, 100% N₂, 0% RH for 2 h for desorption (Fig. 5c inset and Fig. S22). After 500 cycles, the TRCNF/PEI material retained high capacity with negligible degradation (Fig. 5c). The working capacity stabilized at ∼1.93 mmol g⁻¹, retaining 96.5% of its initial cycling capacity (2.00 mmol g⁻¹). Surface chemical stability was confirmed by the minimal change (≤1.2%) in atomic percentages of O, N, and C (Fig. S23). Beyond dry regeneration, the material's adaptability to realistic DAC conditions was thoroughly investigated using TGA, including the presence of humidity and varying regeneration temperatures. Desorption kinetics under humid N₂ carrying gas flow (∼100% RH) at 80 °C revealed 100% complete regeneration within 70 minutes (Fig. S24). Furthermore, complete CO₂ desorption could be achieved in approximately 30 minutes at 65 °C and significantly faster at higher temperatures, requiring only about 10 minutes at 85 °C (Fig. S25a). While a pure CO₂ atmosphere slightly extended the full regeneration time to ∼50 minutes (Fig. S25b, compared to ∼19 minutes in N₂), this remains practically manageable, especially with the application of vacuum or elevated temperatures. This versatility means that the regeneration temperature can be adjusted from 60–85 °C to efficiently match available low-grade waste heat, guaranteeing both complete CO₂ removal and >95% purity. Additional cyclic TGA tests at 75 and 85 °C (Fig. S26) confirmed excellent stability at elevated temperatures over 10 cycles, showing minimal changes in CO₂ uptake kinetics and capacity. Finally, long-term cycling under humid simulated air (0.04% CO₂, 21% O₂ in N₂, ∼100% RH) demonstrated remarkable oxidative stability, with the adsorbent preserving 90.6% of its original CO₂ uptake and minimal amine loss after 53 cycles (Fig. S27), further confirming the robustness of TRCNF/PEI for practical DAC operation. This stability highlights a critical advantage over conventional amine-appended solid adsorbents, where moisture often induces severe amine leaching and degradation. The elongated lifetime of our adsorbent is attributed to the much milder regeneration condition than status quo using high temperature steam for aggressive CO₂ extraction. Notably, TRCNF/PEI exhibits promising DAC performance at sub-ambient temperatures, achieving a CO₂ uptake of 4.33 mmol g⁻¹ at 5 °C (Fig. S28), and its synthesis is readily scalable using commercially available precursors (∼10 USD kg⁻¹ laboratory grade, significantly lower at scale), coupled with its straightforward synthesis process, positioning TRCNF/PEI as a highly promising, robust, and economically practical candidate for large-scale DAC deployment.

3. Conclusion

In summary, we developed a class of thermo-responsive CO₂ adsorbent for DAC with unique LCST-mediated hydrophilic–hydrophobic transition to address the critical energy and stability bottlenecks in DAC. The TRCNF/PEI adsorbent exhibits exceptional CO₂ capture capability (2.82 mmol g⁻¹ at 25 °C, which increases to 4.33 mmol g⁻¹ at 5 °C) and rapid regeneration at a mild temperature of 65 °C. The enhanced performance at lower temperatures, driven by stronger hydration and increased amine accessibility, underscores the material's potential for DAC operation in cold areas. Multidimensional spectroscopic analyses (in situ DRIFTS and DLS) reveal that the regeneration mechanism is driven by the thermally induced hydrophobic collapse of the PNIPAM network. This structural contraction expels water molecules essential for stabilizing bicarbonate species, effectively shifting the chemical equilibrium towards CO₂ release. It obviates the sensible heat penalty for high-temperature regeneration, latent heat penalty of water evaporation, and aggressively out-of-equilibrium temperature-pressure swing in conventional amine-grafted solid adsorbents, leading to unprecedently low levelized energy of 4.36 GJ ton⁻¹_CO2 and exceptional material durability. The crosslinked architecture of our adsorbent anchors amine and avoid leaching in the presence of water, which has been a challenge to conventional amine-appended solid adsorbents, leading to long-term stability over 500 cycles. This work provides a generic paradigm for designing energy-efficient and scalable adsorbent materials to overcome the long-existing adsorbent performance barrier impeding the rapid scale-up of DAC.

4. Experimental section

Details of the synthesis, characterization, and molecular dynamics simulation are explained in the supplementary information.

4.1. Cyclic adsorption–desorption stability test using dynamic breakthrough measurements

The cyclic adsorption–desorption performance was evaluated on the BSD-MAB breakthrough analyzer equipped with a mass spectrometer (MS, INFICON) for the real-time monitoring of CO₂ concentrations. Dry TRCNF/PEI (0.5 g) was hydrated to achieve a water content of 600 wt% and was packed into a glass column (inner diameter = 1 cm). Each cycle comprised (1) adsorption at 25 °C under humidified 400 ppm CO₂/N₂ flow (50 mL min⁻¹) until breakthrough equilibrium and (2) desorption at 65 °C under a dry N₂ flow (50 mL min⁻¹) until complete CO₂ release. The temperature was controlled by the column wall heater, and the system was cooled to 25 °C between cycles. Before each subsequent breakthrough cycle, water was manually replenished to restore the adsorbent to its target hydrated state (600 wt% water content). A total of 15 consecutive cycles were performed. The amount of adsorbed CO₂ was determined by integrating the area between the inlet concentration baseline and outlet breakthrough curve. The cyclic stability was assessed by comparing the CO₂ uptake capacity across all cycles.

4.2. Accelerated aging of adsorbents using thermogravimetric analysis

The cyclic adsorption and desorption behavior of TRCNF/PEI were evaluated using a Mettler Toledo TGA/DSC 3+ analyzer. A customized gas mixing system was employed to generate humidified streams by bubbling dry gases through a water saturator. Before the cycling test, approximately 5 mg of the sample was activated at 100 °C under a dry N₂ flow (50 mL min⁻¹) for 20 min to remove pre-adsorbed moisture and impurities, followed by cooling to 25 °C. The long-term stability test consisted of 500 consecutive adsorption–desorption cycles. Each cycle comprised two steps: (1) adsorption: the sample was exposed to a humidified CO₂ stream (400 ppm CO₂ in N₂, ∼100% RH) at 25 °C for 5 h and (2) desorption: the temperature was ramped to 80 °C under a dry N₂ flow (50 mL min⁻¹) and maintained for 2 h to ensure complete regeneration. The humid adsorption gas was generated by passing the feed gas through a water bubbler, such that the adsorbent was rehydrated in situ during each adsorption step, and no manual water addition was required between cycles. The 0th cycle and 501th cycles (shown) used a 25-hour adsorption step to reach equilibrium and enable accurate capacity comparison. The working capacity for each cycle was calculated based on the mass change between the adsorption and desorption endpoints. The elevated regeneration temperature of 80 °C was deliberately selected as an accelerated aging condition to stress-test the adsorbent under significantly higher-than-normal temperature. The lifetime estimated here thus represents a conservative scenario. To further evaluate oxidative stability, an additional TGA cycling test was performed under humidified simulated air containing 21% O₂. In this experiment, each cycle consisted of adsorption at 25 °C for 5 h under 0.04% CO₂ and 21% O₂ in N₂ at ∼100% RH by flowing the feed gas through a water bubbler, followed by desorption at 80 °C for 2 h under dry N₂ (50 mL min⁻¹). A total of 53 consecutive cycles were conducted.

4.3. In situ DRIFTS measurements

In situ diffuse reflectance infrared Fourier transform spectroscopy (DRIFTS) measurements were performed using a Thermo IS50 spectrometer equipped with an in situ reaction cell over a wavenumber range of 400–4000 cm⁻¹. Prior to the adsorption, the sample was pretreated in an N₂ flow of 50 mL min⁻¹ at 100 °C for 30 min (heating rate: 10 °C min⁻¹) to remove the adsorbed impurities. After the sample cooled naturally to 25 °C and stabilized for 15 min, a background spectrum was collected. Adsorption was initiated by switching the feed gas to a humidified CO₂ stream (400 ppm CO₂ balanced in N₂, flow rate: 50 mL min⁻¹). Subsequently, the regeneration mechanism was investigated by switching the atmosphere back to dry N₂ (50 mL min⁻¹). A stepwise temperature-programmed desorption method was employed, where the sample was heated from 25 °C to 65 °C in 5 °C increments (heating rate: 10 °C min⁻¹). At each target temperature, the sample was equilibrated for 3 min before spectrum acquisition to monitor the evolution of surface species during thermal regeneration.

4.4. Dynamic light scattering (DLS) measurements

The hydrodynamic diameter and size distribution of TRCNF/PEI were determined using a Zetasizer Ultra instrument (Malvern Panalytical, UK). For analysis, an aqueous dispersion of the microgels with a final concentration of 0.006 wt% was prepared. Prior to dilution and measurement, the sample was homogenized in deionized water by magnetic stirring for 10 min. Measurements were performed at 25 °C after a 300 s equilibration period. The multi-angle dynamic light scattering mode was utilized, incorporating data from three detection angles (173°, 90°, and 13°) to ensure angle-independent results. The average hydrodynamic diameter was derived from the autocorrelation function using the cumulants analysis method.

Author contributions

J. Ren synthesized the materials, designed and conducted the experiments, analyzed data, and drafted the manuscript. P. Liu performed spectroscopies. Y. Qin and X. Yuan assisted with the thermogravimetric analysis (TGA). X. Cao and T. Zhou helped with the dynamic breakthrough experiments. J. Zeng conceived the idea, supervised the project, and provided funding.

Conflicts of interest

The authors declare no conflict of interest.

Data availability

The data supporting this article have been included in the main text and the supplementary information (SI). Supplementary information is available. See DOI: https://doi.org/10.1039/d6ee01563h.

Acknowledgements

This work is supported by the National Natural Science Foundation of China (No. 52505525), the HKUST (Guangzhou) startup grant, the Guangzhou Municipal Science and Technology Bureau (No. 2025A03J3951 and No. 2025A04J4329), and the Guangdong Provincial Project (2023QN10L094). We acknowledge the support from Brilliant Energy Science and Technology Lab (BEST Lab) at the Hong Kong University of Science and Technology (Guangzhou).

References

1. J. Meckling and E. Biber, Nat. Commun., 2021, 12, 2051.
2. E. S. Sanz-Perez, C. R. Murdock, S. A. Didas and C. W. Jones, Chem. Rev., 2016, 116, 11840–11876.
3. S. M. Smith, O. Geden, G. Nemet, M. Gidden, W. F. Lamb, C. Powis, R. Bellamy, M. Callaghan, A. Cowie, E. Cox, S. Fuss, T. Gasser, G. Grassi, J. Greene, S. Lück, A. Mohan, F. Müller-Hansen, G. Peters, Y. Pratama, T. Repke, K. Riahi, F. Schenuit, J. Steinhauser, J. Strefler, J. M. Valenzuela and J. C. Minx, The State of Carbon Dioxide Removal, The University of Manchester Research, 1st edn, 2023.
4. R. Socolow, M. Desmond, R. Aines, J. Blackstock, O. Bolland, T. Kaarsberg, N. Lewis, M. Mazzotti, A. Pfeffer, K. Sawyer, J. Siirola, B. Smit and J. Wilcox, Direct Air Capture of CO₂ with Chemicals: A Technology Assessment for the APS Panel on Public Affairs, 2011.
5. International Energy Agency, Direct Air Capture A key technology for net zero, 2022.
6. J. Artz, T. E. Muller, K. Thenert, J. Kleinekorte, R. Meys, A. Sternberg, A. Bardow and W. Leitner, Chem. Rev., 2018, 118, 434–504.
7. H. Azarabadi and K. S. Lackner, Appl. Energy, 2019, 250, 959–975.
8. M. N. Dods, S. C. Weston and J. R. Long, Adv. Mater., 2022, 34, e2204277.
9. J. A. Wurzbacher, C. Gebald, S. Brunner and A. Steinfeld, Chem. Eng. J., 2016, 283, 1329–1338.
10. D. W. Keith, G. Holmes, D. Angelo and K. Heidel, Joule, 2018, 2, 1573–1594.
11. T. H. Kwan and Q. Yao, Int. Commun. Heat Mass Transfer, 2022, 139, 106459.
12. K. Z. House, A. C. Baclig, M. Ranjan, E. A. van Nierop, J. Wilcox and H. J. Herzog, Proc. Natl. Acad. Sci. U. S. A., 2011, 108, 20428–20433.
13. H. Bouaboula, J. Chaouki, Y. Belmabkhout and A. Zaabout, Chem. Eng. J., 2024, 484, 149411.
14. M. Ozkan, S. P. Nayak, A. D. Ruiz and W. Jiang, iScience, 2022, 25, 103990.
15. E. Wang, R. Navik, Y. Miao, Q. Gao, D. Izikowitz, L. Chen and J. Li, Cell Rep. Phys. Sci., 2024, 5, 101791.
16. U.S. Department of Energy, Strategy for the Carbon Negative Shot™, U.S. Department of Energy, Washington, D.C., 2025, https://www.energy.gov/fecm/strategy-carbon-negative-shot.
17. International Energy Agency, World Energy Outlook 2024, 2024.
18. F. Micali, M. Milanese, G. Colangelo and A. de Risi, Appl. Therm. Eng., 2025, 275, 126889.
19. C. D. Díaz-Marín and Z. J. Berquist, Energy Environ. Sci., 2025, 18, 8403–8413.
20. J. Dietsch, FEED Study for Climeworks Direct Air Capture at a California Geothermal Facility with Long-Term Storage, Illinois Sustainable Technology Center, 2022.
21. T. S. Nguyen and C. T. Yavuz, Joule, 2022, 6, 2452–2454.
22. Y. Hu, R. Gani, K. Sundmacher and T. Zhou, Chem. Eng. Sci., 2024, 298, 120423.
23. Y. Hu, X. Li and T. Zhou, Curr. Opin. Chem. Eng., 2025, 50, 101197.
24. Lifecycle carbon intensity of electricity generation, Data source: Ember ( 2026) – a OurWorldinData.org/energy. https://ourworldindata.org/grapher/carbon-intensity-electricity.
25. B. Y. Li, T. Y. Lin, Y. J. Lai, T. H. Chiu and Y. C. Yeh, Small, 2025, 21, e2407420.
26. Z. Li, N. J. Van Zee, F. S. Bates and T. P. Lodge, ACS Nano, 2019, 13, 1232–1243.
27. J. Chen, Z. Wang, S. Li, Q. Xie, C. Ma and G. Zhang, Adv. Funct. Mater., 2025, e25696,  DOI:10.1002/adfm.202525696.
28. E. Loccufier, M. G. Abebe, J. Geltmeyer, T. Meirman, O. I. K. Altan, J. F. R. Van Guyse, E. Khousakoun, H. Gidik, E. Mohsenzadeh, D. Lahem, R. Hoogenboom, B. Maes and K. De Clerck, Adv. Mater., 2025, e17161,  DOI:10.1002/adma.202517161.
29. W. Liu, Z. Wang, J. A. Serna, R. Debastiani, J. E. U. Gomez, L. Lu, W. Yang, Z. Dong and P. A. Levkin, Adv. Funct. Mater., 2024, 34, 2403794.
30. Y. Ji, X. Lin, H. Zhang, Y. Wu, J. Li and Q. He, Angew. Chem., Int. Ed., 2019, 58, 4184–4188.
31. D. Ji, J. Liu, J. Zhao, M. Li, Y. Rho, H. Shin, T. H. Han and J. Bae, Nat. Commun., 2024, 15, 3925.
32. Y. Shi, C. Ma, L. Peng and G. Yu, Adv. Funct. Mater., 2015, 25, 1219–1225.
33. Z. Ming, J. Zhang, W. Li, S. Wang, Y. Zhang, Z. Lu, T. Zhang, Z. Zhou, Y. Xia, Y. Zhang, X. Zhou and J. Xiong, Adv. Mater., 2025, 37, e2501809.
34. B. Arsuffi, G. Siqueira, G. Nyström, S. Titotto, T. Magrini and C. Daraio, Adv. Funct. Mater., 2024, 34, 2409864.
35. T. Buchecker, P. Schmid, I. Grillo, S. Prevost, M. Drechsler, O. Diat, A. Pfitzner and P. Bauduin, J. Am. Chem. Soc., 2019, 141, 6890–6899.
36. M. Yue, Y. Hoshino, Y. Ohshiro, K. Imamura and Y. Miura, Angew. Chem., Int. Ed., 2014, 53, 2654–2657.
37. Y. Hoshino, K. Imamura, M. Yue, G. Inoue and Y. Miura, J. Am. Chem. Soc., 2012, 134, 18177.
38. Y. Liu, H. Xu, L. Zhou and J. Zhang, Green Chem., 2023, 25, 3027–3033.
39. Y. Lu, Z. Li, Y. Xie, Y. Liu, J. Zhang and L. Zong, ACS Mater. Lett., 2024, 6, 976–984.
40. C. W. H. Rajawasam, O. J. Dodo, M. A. S. N. Weerasinghe, I. O. Raji, S. V. Wanasinghe, D. Konkolewicz and N. De Alwis Watuthanthrige, Polym. Chem., 2024, 15, 219.
41. P. J. Flory and J. Rehner, Jr., J. Chem. Phys., 1943, 11, 512–520.
42. Y. Liang, H. Zhu, L. Wang, H. He and S. Wang, Carbohydr. Polym., 2020, 249, 116876.
43. X. Shi, W. Lu, Y. Xue, H. Zhou, F. Xue, H. He, S. Wang and S. Wang, Chem. Eng. J., 2021, 424, 130362.
44. L. Zhan, W. Lu, Q. Xiang, Z. Chen, W. Luo, C. Xu, R. Huang, S. Wang and H. He, J. Colloid Interface Sci., 2025, 687, 261–270.
45. A. Al Mohtar and M. L. Pinto, J. Phys. Chem. C, 2022, 126, 5159–5166.
46. K. Coenen, F. Gallucci, B. Mezari, E. Hensen and M. Annaland, J. CO₂ Util., 2018, 24, 228.
47. J. R. Hoffman, A. E. Baumann and C. M. Stafford, Chem. Eng. J., 2025, 507, 160347.
48. J. M. Kolle, M. Fayaz and A. Sayari, Chem. Rev., 2021, 121, 7280–7345.
49. W. Lu, R. Chen, L. Zhan, Q. Xiang, R. Huang, L. Wang, S. Wang and H. He, Nano-Micro Lett., 2025, 18, 80.
50. J. Wu, Y. Chen, Y. Xu, S. Chen, H. Lv, Z. Gan, X. Zhu, R. Wang, C. Wang and T. Ge, Matter, 2024, 7, 123–139.
51. J. Chen, Z. Wang, S. Li, Q. Xie, C. Ma and G. Zhang, Adv. Funct. Mater., 2026, 36, e25696.
52. S. Wu, Z. Yang, L. Chen, L. Cao, Y. Qiu, H. Ma, Q. Guan, Y. Cai, Q. Fu and S. Yu, ACS Nano, 2026, 20, 6864–6874.
53. R. Wang, H. Zhao, C. Qi, X. Yang, S. Zhang, M. Li and L. Wang, Energy, 2022, 260, 125045.
54. J. Fang, R. Wang, Z. Chen, H. Zhou, L. Wang, Q. Li, S. Zhang and L. Wang, Sep. Purif. Technol., 2025, 359, 130475.
55. R. Wang, C. Qi, Z. Jian, H. Zhao, P. Zhang, S. An, Q. Li and L. Wang, Fuel, 2024, 359, 130382.
56. A. Kiani, K. Jiang and P. Feron, Front. Energy Res., 2020, 8, 92.
57. Z. Zhou, T. Ma, H. Zhang, S. Chheda, H. Li, K. Wang, S. Ehrling, R. Giovine, C. Li, A. H. Alawadhi, M. M. Abduljawad, M. O. Alawad, L. Gagliardi, J. Sauer and O. M. Yaghi, Nature, 2024, 635, 96–101.

---