Introduction

Lithium’s applications span across glass manufacturing, pharmaceuticals, and nuclear energy sectors, with its demand for lithium batteries significantly amplified by the burgeoning electric vehicle market1. The reserves of terrestrial and marine lithium resources are about 14 million tons and 230 billion tons respectively2,3,4. Currently, lithium is sourced through industrially processes from lithium ores; however, it supply about 34% of the global lithium demand with only 0.1% of the global lithium reserves2,5. Lithium extraction from ores involves processes such as calcination, acid leaching, and precipitation. These are energy-intensive and require large quantities of chemical reagents like sulfuric acid, limestone, and sodium carbonate, also leading to environmental pollution6,7,8. Despite seawater’s vast availability and lack of geographic constraints, the exceedingly low lithium ion concentration ( ~ 0.2 ppm) and extremely high sodium ion concentration (~10,500 ppm) make the purification of lithium ions very challenging3,9,10,11. Terrestrial salt lakes are difficult to develop widely due to geographical factors and other interfering ions, and the current extraction techniques from salt lakes, such as evaporation concentration and chemical precipitation, are not only time-consuming but also suffer from low efficiency and cause considerable environmental pollution12.

To address these issues, significant research has been focused on alternative extraction methods, including solvent extraction13, membrane separation14,15,16, ion sieve adsorption17, and electrochemical techniques18. However solvent extraction involves the use of hazardous organic solvents, membrane separation struggles with the separation of monovalent ions16, and ion sieve adsorption requires acids and bases leading to dissolution losses during extraction19. Recently, Chen et al. proposed a novel approach for lithium extraction using a physical method where hydrophilic cellulose ropes were employed to induce water evaporation, resulting in the selective ion separation in distinct spatial locations20. Despite this advancement, there remains an urgent need to develop efficient, environmentally friendly, and cost-effective lithium extraction technologies.

In this work, we developed a physical technology for the precise and selective separation of alkali or alkaline earth metal ions using spherical sorbents composed of polyacrylonitrile and chitosan, utilizing the distinct deliquescent humidity ranges for salts. The hydrophilic sorbents can absorb a large amount of liquid dissolving the specific salt by controlling the environmental humidity. Taking lithium extraction as representative, this physical method indicated extremely high capacity, along with high purity and fast ion recovery rate by controlling relative humidity (RH) at 40%. Additionally, apart from using water vapor, this method can also utilize organic solvent vapors (such as ethanol) for extracting LiCl and other deliquescent salts. We also calculated the costs associated with this method and found that it is comparable to current commercial lithium extraction methods, indicating a certain potential for practical application.

Results and discussion

Selective lithium extraction via humidity-controlled deliquescence

The process for physical lithium extraction by controlled humidity is illustrated in the Fig. 1a, using solid mixed salts obtained from ore through calcination and from salt lakes (rich in alkali metal ions and low in alkaline earth metal ions)/seawater through evaporation21,22,23,24. The mixed salts were then mixed with spherical sorbents and placed in an adsorption column. The spherical sorbents made of polyacrylonitrile and chitosan were synthesized using a phase transformation approach. By controlling the humidity of the air introduced into the adsorption column, specific salts were gradually absorbed by the sorbent spheres after deliquescence for a certain period. The other types of salt adhering to the sorbent could be recovered by sieving, and the adsorbent spheres were then placed in deionized water, thereby achieving the concentrated solution of target ion. By controlling different ranges of relative humidity, selective separation of specific ions can be achieved (Fig. 1b).

Fig. 1: Physical sorbent for extracting lithium from mixed salts using polyacrylonitrile-chitosan composite spheres by manipulating humidity.
Fig. 1: Physical sorbent for extracting lithium from mixed salts using polyacrylonitrile-chitosan composite spheres by manipulating humidity.The alternative text for this image may have been generated using AI.
Full size image

a Schematic illustration of the procedure for lithium extraction. b Separating a specific salt in the mixture by adjusting the range of humidity. c The state of various salts at different temperatures and vapor pressure of water. d Photos of the spheres as sorbent before and after lithium extraction.

For the detailed experiments, we selected to separate specific cations from a mixed salts of alkali or alkaline earth metal chlorides (LiCl, NaCl, KCl, MgCl2), because Li+, Na+, K+, Mg2+ are the primary cations in salt lakes and seawater25. In addition, we obtained solid mixed salts by mixing anhydrous salts. In Supplementary Note 1 (Supplementary Figs. 1–4, Supplementary Table 1–3), we discuss the differences between solid mixed salts obtained via anhydrous salt mixing and those obtained through solution evaporation. The results show that using anhydrous salt mixtures can, to some extent, simulate the solid mixed salts obtained from solution evaporation. The magnitude of a salt’s hydration energy significantly affects its ability to bind with water molecules. Due to the lithium ion’s smallest radius and highest point charge density, it has the greatest hydration capacity and is therefore most inclined to bind with water, followed by the order of Mg2+, Na+, and K+. When the mixed salts are exposed to different water vapor pressures, as the water vapor pressure increases, LiCl absorb water and deliquesce first, followed by magnesium chloride, then sodium chloride, and finally potassium chloride (Fig. 1c). At 25 °C and 1 atm, LiCl, MgCl2, NaCl, and KCl deliquesce at relative humidities (RH) of 11%, 33%, 75%, and 84%, respectively. The critical humidity for deliquescence of these salts varies little across different temperatures (0-50 °C) (Supplementary Note 2, Supplementary Fig. 5). Based on this principle, we can selectively deliquesce the desired salts by controlling the humidity of the environment in which the mixed salts are placed, thereby achieving selective separation of the target ions. For example, when RH is maintained within the range of 11-33% in the adsorption column, LiCl deliquesces first, while MgCl2, NaCl, and KCl do not (Supplementary Fig. 6). Since the concentration of the deliquesced LiCl solution is usually very high, typically above 7 M (depending on RH) (Supplementary Note 3), it can significantly inhibit the dissolution of other salts, resulting in a relatively pure LiCl solution. At this stage, the sorbent spheres could absorb the solution from the mixed salts. The other types of salts adhering to the sorbent spheres could be recovered by sieving. After absorbing the deliquescent solution, the sorbent spheres change color from white to light yellow, and the color can indicate the extent of solution adsorption by the sorbent (Fig. 1d). Finally, the sorbents are washed at a ratio of their pre-adsorption mass (g) to the volume of ultrapure water (mL) ranging from 1:10 to 1:2000 to obtain about 65.4 mg L-1 to 9.1 g L-1 target ions. In the experiment, to maximize the recovery of ions from the sorbents for a more accurate measurement of the ion adsorption capacity, a ratio of 1:2000 was used for recovery and testing (Supplementary Note 4, Supplementary Fig. 7). It can be observed from Supplementary Movie 1 that LiCl rapidly diffuses from the into the water. After the Li+ has been extracted from the solid salts, RH can be tailored within the range of 33−75% to further separate Mg2+, and this process can be continued to separate the subsequent salts (Fig. 1b).

Selection and preparation of adsorbent sphere materials

In this physical adsorption technology, the composition and structure of the sorbent are crucial for its lithium extraction performance. The materials for the sorbent of large-scale production should meet the following criteria: (1) good hydrophilicity and high mechanical properties; (2) preparation accessibility; (3) environmentally friendly; and (4) widely available in nature. The synthesis of different polymers and the preparation of corresponding sorbents are time-consuming. Therefore, machine learning serves as a powerful tool to facilitate material selection. The key performance indicator is water absorption capacity, which is closely related to the hydrophilicity of the material. As such, polymer information was extracted from existing databases to facilitate material screening and selection26, using the contact angle as the primary criterion for selecting hydrophilic polymers. Initially, we used the types of functional groups and their density on the polymer chains as features for input, with the contact angle as the label. We then conducted machine learning predictions for 64 different polymers using fully connected networks (Fig. 2a). Meanwhile, we performed parameter optimization for each model hyperparameter based on the root mean square error (RMSE) and selected the optimal parameter set accordingly (number of iterations = 10,000, learning rate = 0.0001, number of nodes = 500, activation function = ELU, loss function = MSELoss, and optimizer = RMSprop) (Supplementary Figs. 8–13, Supplementary Table. 4–9). The results showed that the neural network performed the best, with an R² of 0.9526 (Fig. 2b and source data). The predicted contact angles for PAN and CS were 53.78° and 46.96°, respectively, while the actual measured values were 56.72° and 48.28°, demonstrating good predictive accuracy (Fig. 2c and Supplementary Fig. 14a, b). In addition, we conducted both prediction and experimental measurements of the contact angles of several polymers (Supplementary Table 10). Based on the results, PAN (polyacrylonitrile) and CS (chitosan) were selected as the final materials for the sorbent spheres (Supplementary Note 5, Supplementary Fig. 15). Simultaneously, the PAN-CS composite adsorbent also exhibits good hydrophilicity, with a contact angle between the two, measured at 53.63° (Supplementary Fig. 14c). Based on these results, polyacrylonitrile and chitosan were ultimately selected as the materials for the adsorption sorbent.

Fig. 2: Select hydrophilic materials through neural network by machine learning.
Fig. 2: Select hydrophilic materials through neural network by machine learning.The alternative text for this image may have been generated using AI.
Full size image

a The contact angle of the polymer is predicted by the input of the functional group, surface energy, etc. b Training results for training sets and test sets. c The contact angle between several polymers was predicted.

Microstructural and component characterization of composite sorbent spheres

We synthesized sorbent spheres of polyacrylonitrile, polyacrylonitrile-chitosan, polyacrylonitrile-ethyl cellulose, and polyacrylonitrile-cellulose acetate through phase inversion. The sorbents have an average diameter of 3.7 ± 0.1 mm (Supplementary Fig. 16). From scanning electron microscope (SEM) images (Supplementary Fig. 17), it can be observed that the spheres have multilevel channels perpendicular to the sphere surface, with pore sizes gradually increasing from the surface to the interior. The pores on the sorbent surface are very small, with a diameter of 18.0 ± 9.8 nm (Supplementary Fig. 17a), and the diameter of channels ranges from 5.2 ± 1.2 μm near the surface to 58.4 ± 17.8 μm at the center of the sphere (Supplementary Fig. 17b, c). Additionally, smaller pores are present within the pore walls, with diameters of 291.1 ± 102.9 nm (Supplementary Fig. 17d). Brunauer-Emmett-Teller (BET) adsorption tests indicated that the pores in the composite sorbent have a wide distribution, ranging from 6 to 20 nm (Supplementary Fig. 18), which is consistent with the SEM results. These observations indicate that the internal structure of the sorbent is a three-dimensional network formed by pores of varying sizes interconnected throughout the sorbents.

The cross-section structures of the composite sorbents before and after lithium extraction were systematically characterized. For the composite sorbents before lithium extraction, the nitrogen (N) element derived from polyacrylonitrile and chitosan are uniformly distributed, while the oxygen (O) element exclusively from chitosan tends to aggregate, indicating that chitosan is embedded within the polyacrylonitrile in the sorbent sphere (Supplementary Fig. 19). This phenomenon occurs because, during the preparation of the composite sorbent, polyacrylonitrile dissolves in N,N-dimethylformamide (DMF), while chitosan does not. As a result, during the phase transition, chitosan becomes embedded within the composite sorbent. We tested the water absorption capacity of the composite sorbent by placing them in water and in a LiCl solution with a lithium ion concentration of 8.5 M, respectively, and leave for 10 days. The results showed that the composite sorbent could absorb 451% and 491% of their own weight in solution, respectively. This is because, after adsorption equilibrium is reached, the composite sorbent absorbed the same volume of liquid, but the LiCl solution has a higher density, resulting in a greater absorption mass for the LiCl solution compared to water (Supplementary Fig. 20).

Compared to the large depth-of-field (LDF) mode of the electron microscope and to achieve a more comprehensive observation of the element spatial distribution of sorbent spheres after Li extraction (extraction from LiCl/NaCl mixed salt of 1.03 wt.% LiCl, 40% RH, 28 °C and mass of sorbent: mass of mixing salt = 1:40), we used the same parameters to capture an 8\(\times\)8 cross-sectional SEM and EDS images and stitched them together (Fig. 3a–e and Supplementary Fig. 21, Supplementary Note 6). It can be observed that the internal pore structure of the composite sorbent remains unchanged after lithium extraction, still showing a radial pore arrangement. Energy disperse spectroscopy (EDS) analysis indicates that the signals of C, N, and O elements have significantly diminished (Fig. 3b, c), replaced by a nearly uniform distribution of Cl elements across the cross-sections of the sorbents (Fig. 3e). In contrast, Na elements are sparsely and isolatedly distributed on the section (Fig. 3d). Based on the law of charge conservation, we can indirectly infer that LiCl is extensively distributed within the interior of the sorbent sphere.

Fig. 3: Micro morphology of sorbent spheres for the lithium extraction.
Fig. 3: Micro morphology of sorbent spheres for the lithium extraction.The alternative text for this image may have been generated using AI.
Full size image

a Stitched SEM image of cross-section of polyacrylonitrile-chitosan composite sphere after lithium extraction, with corresponding EDS mapping of (b) nitrogen, (c) oxygen, (d) sodium and (e) chloride. f 3D X-ray scanning image of the adsorbent spheres after lithium extraction. TOF-SIMS mapping images of (g) lithium and (h) Na elemental near the center of the sphere after lithium extraction (marked by a red box in Fig. 3a). XPS analysis of adsorption spheres after lithium extraction for (i) lithium, (j) sodium and (k) oxygen.

Additionally, similar with SEM image, the 3D X-ray scanning image (Fig. 3f and Supplementary Movie 2–4) also indicated that the pore size in the sorbent sphere decreases radially from the inside out. The 3D X-ray scanning image images also can reveal contrast variation due to the relative density differences. Since polyacrylonitrile (PAN) and the high-concentration LiCl solution have very similar densities ( ~ 1.18 vs. ~1.2 g cm-3), their contrast cannot be distinguished and it has the lowest density among all components. Therefore, the yellow regions represent areas where PAN and LiCl solution are distributed. Chitosan has a relatively higher density ( ~ 1.75 g cm-3), so the orange regions inside the adsorbent spheres likely correspond to the distribution of chitosan. NaCl has the highest density, and micron-sized NaCl particles cannot pass through the sphere’s surface pores (which are nano- or sub-micron in size). Thus, the red regions on the surface mainly represent NaCl. A large amount of LiCl envelops the composite sorbent, which diminishes the signals of C, N, and O elements. Meanwhile, a small amount of NaCl crystals is isolatedly distributed (Supplementary Figs. 22–24). The time-of-flight secondary ion mass spectrometry (TOF-SIMS) mapping images (Fig. 3g, h and Supplementary Figs. 25, 26) indicated a substantial distribution of lithium on the pore walls of the sorbent, with only a small amount of Na present. This confirms the abundant presence of lithium and indicates that the separation and purification process was effective. Similar results were observed in other locations as well (Supplementary Fig. 27). These results indicated the high Li selectivity in the physical adsorption technology by controlling RH.

To further confirm the presence of lithium in the adsorbent spheres after lithium extraction, we employed X-ray photoelectron spectroscopy (XPS). The XPS results revealed three distinct lithium signals both before and after etching. The peak at ~56.6 eV corresponds to anhydrous lithium chloride27, while the other two peaks are assigned to hydrated lithium chlorides28 (Fig. 3i). Similarly, the O element spectra showed the appearance of a new peak at lower binding energy after lithium extraction. Considering the uniform distribution of oxygen in the TOF-SIMS mapping (Supplementary Fig. 27), in contrast to the isolated distribution of oxygen observed in the SEM mapping before lithium extraction (Supplementary Fig. 19), we attribute this new O peak to water associated with hydrated lithium chloride (Fig. 3k). In comparison, since NaCl does not form hydrates, it only exhibits a single low-intensity peak (Fig. 3j). These results confirm the presence of lithium species in the adsorbent sorbents after lithium extraction.

Then, we evaporated the recovered solution to dryness, obtaining a solid powder (Supplementary Fig. 28a). X-ray diffraction (XRD) results further confirmed that the main component of the powder is LiCl (Supplementary Fig. 28b). SEM and EDS analyses revealed that the distribution of sodium was much less than that of chlorine, indicating that the primary component of the solid powder is LiCl (Supplementary Fig. 29). ICP-OES analysis showed that, among the cations in the solid powder, Li+ accounted for 93.28 wt.% and sodium ions for 6.72 wt.%, demonstrating that the LiCl is of very high purity.

Separation performance of humidity-controlled ion sieving

Next, we carried out the Li extraction experiments and optimized the parameters for this physical adsorption technology. We first investigated the effects of temperature, gas flow rate, and the ratio of adsorption spheres to mixed salts. Based on the results, we selected 28°C (Supplementary Note 7, Supplementary Fig. 30), a gas flow rate of 3 L min-1 (Supplementary Note 8, Supplementary Fig. 31), and a sphere-to-salt ratio of 1:40 for subsequent experiments (Supplementary Note 9, Supplementary Fig. 32 and source data).

Subsequently, we investigated the lithium extraction performance of the sorbents under various ambient RH (Fig. 4a and Supplementary Fig. 33a and source data). The results indicate that when the RH is between 25% and 40%, the adsorption capacity remains relatively stable at around 130 mg g-1. This stability is due to the fact that as the humidity increases, both the concentration and viscosity of the deliquesced LiCl slightly decrease, which is a trade-off relationship for adsorption capacity. At the same time, the decrease in solution concentration weakens the ability of the LiCl solution to inhibit the dissolution of impurity salts (such as NaCl), leading to a slight decrease in purity, which was 93.28 wt.%, 91.17 wt.%, and 89.11 wt.% at 25%, 30%, and 40% RH, respectively. However, as the humidity continued to increase to 70% RH, the adsorption capacity experienced a significant decline. At a 70% RH, the adsorption capacity dropped to 70.39 mg g-1, and the purity also decreased sharply to 52.17 wt.%. Therefore, considering both adsorption capacity and purity, 40% RH is optimal in this physical adsorption technology.

Fig. 4: Extracting performances of the sorbent spheres by controlling humidity.
Fig. 4: Extracting performances of the sorbent spheres by controlling humidity.The alternative text for this image may have been generated using AI.
Full size image

a The adsorption capacity and purity of lithium extraction under different relative humidity and (b) different lithium extraction time. c, Cycle performance of the sorbents at a relative humidity of 40%. d Practical applications for Li extraction including seawater, LiCl\NaCl mixing salt, LiCl\NaCl\KCl mixing salt, LiCl\NaCl\KCl mixing salt with ratio of Taijinai’er salt lake, LiCl\NaCl\KCl\MgCl2 mixing salt, LiCl\NaCl\KCl\MgCl2\CaCl2 mixing salt with ratio of Zabuye salt lake using the sorbent spheres by controlling humidity, and for Mg extraction. e Lithium extraction using ethanol vapor. f Comparison of Li adsorption capacity, rate in our work with different methods. g Comparison of energy consumption of lithium extraction by humidity-controlled and lithium extraction by electrochemistry methods. h Cost comparison between lithium extraction by humidity-controlled and evaporation ponds and metal-based adsorbent.

Subsequently, we investigated the effect of adsorption time on adsorption capacity and purity at 40% RH using 5:200 wt. ratio of mixed salt to sorbent spheres (Fig. 4b and Supplementary Fig. 33b and source data). The results show that the adsorption capacity increased with time, reaching its peak at 133.60 mg g-1 after 3 h with a fast lithium extraction rate of 44.53 mg g-1 h-1. Afterward, the adsorption capacity slightly decreased and stabilized at around 121 mg g-1 from the 5th hour onward. Similarly, purity also decreased slightly over time, dropping from an initial 92.32 wt.% to 85.45 wt.%. This decline in purity is due to the sorbent’s diminished ability to inhibit the dissolution of NaCl, which occurs alongside the continuous reduction in the concentration of the deliquescent solution over time. We fitted the adsorption data using pseudo-first-order (PFO) and pseudo-second-order (PSO) kinetic models (Fig. 4b and source data). The results show that the PFO kinetic model fits the data better, indicating that the adsorption rate is primarily influenced by the number of available adsorption sites, suggesting that the process is dominated by physical adsorption.

Additionally, to determine the adsorption behavior of the sorbent in the solution and to identify whether the rate-determining step in the deliquescence adsorption process is the liquefaction of the salt or the adsorption by the sorbent. We immersed the composite sorbent in a LiCl solution with a Li+ concentration of 8.5 M (corresponding to the deliquescence concentration at 40% RH) (Supplementary Fig. 34). The adsorption capacity of the composite sorbent exhibited an “S”-shaped curve. During the initial ~50 s, the LiCl solution uptake increased continuously, corresponding to the process of the sorbent being wetted by the LiCl solution. From 50−150 s, the solution uptake increased significantly was observed, which corresponds to the filling of the large radial pores inside the composite sorbent. From approximately 150–600 s, there was a slow increase in adsorption, corresponding to the filling of the small pores within the sorbent. After 600 s, the adsorption process reached equilibrium. It is noteworthy that the time required for the sorbent to reach equilibrium (10 min) is significantly shorter than the time required for salt deliquescence (3–4 h) (Supplementary Fig. 35). Therefore, the rate-determining step in the entire humidity-controlled lithium extraction process is the deliquescence LiCl. Therefore, increasing the adsorption temperature, enhancing the gas flow rate, and raising the humidity (which may reduce product purity) can all accelerate the deliquescence of LiCl.

Next, we also investigated the adsorption behavior of the sorbent spheres under different concentrations (viscosities). The viscosity of LiCl aqueous solutions at various concentrations (5, 10, 20, 40, 60, 80, 100, 200, 300, 360 (8.5 M), 400, and 500 g L-1) was tested at 28°C, along with the corresponding adsorption performance of the adsorbent spheres after 3 h (Supplementary Fig. 36a). At 28°C, the viscosity of pure water is 0.836 mPa·s. As shown in Supplementary Fig. 36a, the viscosity of LiCl solutions increases gradually with concentration at lower levels ( < 100 g L-1), but exhibits an exponential increase once the concentration exceeds 100 g L-1. Therefore, at lower concentrations ( < 360 g L-1), the relatively low viscosity of the solution facilitates adsorption within the adsorption spheres, allowing them to reach saturation within 3 h. However, as the concentration (and thus viscosity) increases further, the reduced diffusion rate hinders the spheres from achieving adsorption saturation within the same period (Supplementary Fig. 36b). Moreover, while highly concentrated LiCl solutions are advantageous for suppressing the dissolution of impurity ions, their high viscosity imposes limitations on the adsorption process (Supplementary Fig. 36c). At the same time, the results show that the adsorption process fits the Freundlich model very well, indicating that the adsorption process involves multilayer adsorption (Supplementary Fig. 37).

We conducted fourier transform infrared spectroscopy (FT-IR) tests on the raw materials used for synthesizing the composite sorbent—polyacrylonitrile, chitosan—as well as on the composite sorbent before lithium extraction, water-immersed composite sorbent, and the composite sorbent after lithium extraction (Supplementary Fig. 38). The results showed that none of the peaks shifted after lithium extraction, indicating that there was no chemical interaction between the lithium ions and the composite sorbent. Combined with the Freundlich model results that suggest multilayer adsorption, this indicates that the adsorption process is physical adsorption. Simultaneously, we conducted a cycling performance test of the sorbent using a mixed salt of LiCl (with a lithium ion content of 1.03 wt.%) and NaCl at 40% RH (Fig. 4c). The results showed that after 100 cycles, the adsorption capacity retained a high capacity of 104.70 mg g-1, indicating good cycling performance (source data). It can be observed that the appearance of the sorbent shows almost no change before and after the cycling process. (Supplementary Fig. 39a, b). The stable cycling performance is due to its physical process rather than chemical reaction.

Furthermore, we conducted lithium extraction tests under more practical conditions (The ionic composition of the simulated salt lake was set to match the ion ratios found in the untreated raw brine from the original salt lake.), with the testing environment set to 40% RH for 3 h (Fig. 4d and source data). Firstly, we attempted to separate Li from a Li, Na, and K salt mixture, for a LiCl\NaCl\KCl mixed salt with a Li+ content of 0.88 wt.%, the Li purity increased to 78.64 wt.%. Next, in a simulated mixed salt resembling the composition of Taijinai’er salt lake, containing Li, Na, and K, the lithium ion purity increased from 0.50 wt.% to 73.73 wt.%, and the Li/Na selectivity of 2186.68, Li/K selectivity of 52.60. It should be noted that since the deliquescence humidity of MgCl2 is 33%, it was necessary to lower the humidity to suppress the deliquescence of MgCl2 for extracting Li from the mixed salts with Mg using this methods. Hence, when the humidity was reduced to 20% and lithium extraction was conducted for 10 h, the lithium purity could be increased from 6.52 wt.% to 81.25 wt.% for the LiCl\NaCl\KCl\MgCl2 (Li+(6.52 wt.%), Na+(31.35 wt.%), K+(41.79 wt.%), Mg2+(20.34 wt.%)) mixed salts, and the Li/Na selectivity of 319.69, Li/K selectivity of 33.78, Li/Mg selectivity of 58.06. In a simulated Zabuye Salt Lake containing Li+(0.54 wt.%), Na+(80.88 wt.%), K+(18.42 wt.%), Mg2+(0.03 wt.%), and Ca2+(0.12 wt.%), the Li+ purity increased from 0.54 wt % to 74.02 wt.% after lithium extraction and the recovery rate was 95.41%. The adsorption rate was 28.17 mg g-1 h-1, with an adsorption capacity of 84.51 mg g-1. The Li/Na selectivity of 4662.08, Li/K selectivity of 137.42, Li/Mg selectivity of 1.52, and Li/Ca selectivity of 6.95 were achieved. We also prepared a simulated seawater mixed salt with a composition of LiCl and NaCl (0.0015 wt.% Li+ and 99.9985 wt.% Na+), based on the proportions found in seawater. After lithium extraction using this physical adsorption technology, the lithium ion purity increased from 0.0015 wt.% to 2.53 wt.%, a 1687-fold increase, with Li/Na selectivity of 1733.85.

At the same time, we performed magnesium extraction from a crude salt (crude salt is dried from sea water, and the content of Mg2+ is 6.58 wt.%) obtained by evaporating seawater, with the humidity controlled at 40% and an extraction time of 3 h. The results showed that the magnesium ion purity increased from an initial 6.58 wt.% to 62.84 wt.%. When extracting Mg2+ from a mixed salt with a composition similar to that of Taijinai’er Salt Lake (Na+: 70.12 wt.%, K+: 5.55 wt.%, Mg2+: 24.08 wt.%, Ca2+:0.25 wt.%), the Mg purity increased from 24.08 wt.% to 90.21 wt.%, and the Mg/Na selectivity of 54.71, Mg/K selectivity of 310.48, Mg/Ca selectivity of 0.39, demonstrating a good purification effect. Since the relative humidity is maintained at 40%, both MgCl2 and CaCl2 undergo deliquescence and are absorbed by the sorbent. As a result, Ca2+ ions are enriched alongside Mg2+ ions. Given that CaCl2 has a lower relative equilibrium humidity (ERH = 30%) compared to MgCl2 (ERH = 33%), CaCl2 deliquesces more readily and is therefore more effectively enriched than MgCl2. This leads to an Mg/Ca selectivity of <1. These results indicate that by controlling the humidity within a specific range, we can selectively purify salts that have deliquescence points within that humidity range.

Additionally, this deliquescence purification can occur not only in water vapor but also in the vapors of other solvents such as ethanol (Fig. 4e and Supplementary movie 5 and source data). Therefore, we conducted lithium extraction tests on LiNa and LiNaK mixed salts using ethanol vapor. The results showed that the lithium ion purity increased from 1.03 wt.% and 0.88 wt.% to 94.88 wt.% and 92.24 wt.%, respectively. This is a better result compared to the purification under 40% relative humidity (89.11 wt.% and 78.64 wt.%, respectively). The improved separation is due to the fact that impurity salts (NaCl and KCl) are less soluble in ethanol, leading to more effective separation. Additionally, the adsorption capacities for lithium extraction using ethanol were 62.74 mg g-1 and 49.52 mg g-1, respectively, which are lower than the adsorption capacity for lithium extraction using water vapor (133.60 mg g-1). This is because the solubility of LiCl in ethanol (5.73 mol kg-1 at 20 °C) is significantly lower than its solubility in water (19.82 mol kg-1 g at 20 °C). And, the adsorption rate was 20.91 mg g-1 h-1 using ethanol vapor.

Finally, we compared our results with the most recent literatures on lithium extraction methods (Fig. 4f and Supplementary Table 11–14). In this work, within a mixed LiCl-NaCl salt of 1.03 wt.% LiCl, RH = 40%, an extremely fast lithium extraction rate of 83.64 mg g-1 h-1 was achieved with an adsorption time of 1 h and a high adsorption capacity of 83.64 mg g-1. The optimal adsorption time of 3 h resulted in an adsorption rate of 44.53 mg g-1 h-1 and an adsorption capacity of 133.60 mg g-1 and capable of adsorbing LiCl solutions with concentrations as high as 8.5 M. For the currently chemical ion sieves methods, such as aluminum-based, manganese-based, and titanium-based adsorbents, the average adsorption rates and capacities are 2.16 mg g-1 h-1 and 6.32 mg g-1, 1.40 mg g-1 h-1 and 34.11 mg g-1, and 1.91 mg g-1 h-1 and 34.36 mg g-1, respectively. In comparison, the deliquescence adsorption method shows a significant improvement by 4–62 times. Moreover, compared to electrochemical adsorption methods, which has a high adsorption capacity (averaging ~22.31 mg g-1) and fast kinetics (averaging ~10.59 mg g-1 h-1), our technology also demonstrates significant advantages. Moreover, compared to other methods, the chemicals used in this humidity-controlled lithium extraction technology using water vapor are eco-friendly, energy-efficient, and cost-effective. In the laboratory, the energy consumption for lithium extraction comes from gas output, and it is estimated that the energy consumption is 0.43 kWh kg-1 Li, making it an energy-efficient method29,30,31,32,33,34,35,36,37,38,39,40,41,42,43 (Fig. 4g). The cost for existing lithium extraction methods is over 6500 USD ton-1 of lithium carbonate (LCE)36, while the humidity-controlled lithium extraction technique only costs 4954 USD (Fig. 4h, Supplementary Note 10, Supplementary Table 15).

Mechanism and outlook of humidity-controlled ion sieving

Last, we systematically investigated the mechanism behind deliquescence separation and purification, using the LiCl-NaCl binary salt system as an example (Fig. 5a and source data). From the phase diagram, we can observe that the boundaries between the blue region and the NaCl saturation region (orange area) and between the blue region and the LiCl saturation region (the area indicated by the arrow) represent the solubility curves of the salt. Additionally, because LiCl has a very high solubility, the region where hydrated LiCl and the liquid phase coexist (the area indicated by the arrow) is very small. Typically, most phase diagrams focus on displaying the transitions between solid and liquid phases at atmospheric pressure (or under a fixed pressure). Furthermore, it is essential to represent the phase diagram in terms of the water vapor partial pressure (or relative humidity) to accurately depict the relevant phase behavior (Fig. 5b). In practice, measuring the water vapor partial pressure of a system is quite challenging and requires sophisticated instruments, especially for a complete phase diagram. Therefore, we used an ideal theoretical model based on Raoult’s law, combined with empirical formulas for the saturation vapor pressures of LiCl and NaCl solutions, to qualitatively calculate the three-dimensional phase diagram of LiCl, NaCl, water, and relative humidity (Supplementary Note 11). From Fig. 5B, we can observe that when the salt concentration is zero, the water vapor partial pressure corresponds to the saturation vapor pressure of water, which aligns with 100% relative humidity. In a LiCl solution, as the concentration of LiCl increases, the relative humidity decreases continuously, stabilizing at around 11% once the solution reaches saturation. After this point, adding more solid LiCl does not change the composition of the solution, and therefore, the relative humidity remains constant. The same phenomenon occurs in NaCl solution, where the relative humidity stabilizes at 75% for saturated NaCl solution. Along the solubility line, there is a “ridge” in the relative humidity, meaning that adding LiCl to a saturated NaCl solution, or adding NaCl to a saturated LiCl solution, causes the vapor pressure to first increase and then decrease. Taking the addition of LiCl to a saturated NaCl solution as an example: according to the phase diagram, when LiCl is added to a saturated NaCl solution, a small amount of LiCl can precipitate more NaCl. This means that the reduction of NaCl decreases its inhibitory effect on water evaporation, leading to an increase in the vapor pressure of the solution. Since the added LiCl is present in a small quantity, its inhibitory effect on water evaporation cannot compensate for the impact caused by the reduction of NaCl. As a result, the vapor pressure/RH of the solution increases. In the region where both LiCl and NaCl are saturated, the vapor pressure remains around 11%, because the solution composition no longer changes with the addition of any solid, hence the relative humidity stays constant.

Fig. 5: Mechanism on lithium extraction by humidity control.
Fig. 5: Mechanism on lithium extraction by humidity control.The alternative text for this image may have been generated using AI.
Full size image

a Phase diagram of LiCl, NaCl and water. b NaCl, LiCl and ambient relative humidity contour maps46. c Phase diagram iso-humidity line and zoomed-in image of the region with LiCl mass fraction at range of 0-5% and NaCl mass fraction 90−100%. d The relationship between the purity of lithium and humidity44. e Equilibrium relative humidity of common salt-saturated solutions. f Solubility of common salts in common solvents.

Additionally, by projecting the three-dimensional phase diagram onto the z-axis, we can obtain a phase diagram of LiCl, NaCl, and water that includes information about humidity. This projection provides a clearer understanding of how relative humidity interacts with the solubility of these salts in water (Fig. 5c). For our experimental system, the initial composition was a mixed salt of 2.4 wt.% LiCl (Li+: 1.03 wt.%) and 97.6 wt.% NaCl, which corresponds to the gray point on the phase diagram. When water vapor with 40% RH is introduced, the point representing the system on the phase diagram moves along a direction passing through the origin (indicated by the blue dashed line) until it reaches the equilibrium point on the 40% relative humidity contour (the blue point). Since the composite sorbent absorb liquid, the liquid corresponds to the point on the solubility curve where the equilibrium RH is 40%—represented by the pink point. From the phase diagram, it can be observed that at this stage, the solution is composed of a very concentrated LiCl solution with only a small amount of NaCl. We then compared the lithium extraction purity obtained at different relative humidities with the theoretically calculated purity (Fig. 5d). The overall trend shows good agreement, supporting the validity of our theoretical model.

Humidity-controlled ion separation has broad application potential. In addition to lithium and magnesium salts, different salts have different equilibrium relative humidities44 (Fig. 5e). Similar to the lithium ion separation experiments described earlier, the equilibrium relative humidity of a salt is often closely related to its deliquescence humidity. A lower equilibrium relative humidity indicates that the salt binds more readily with water, leading to higher solubility. This trend can be observed in Fig. 5e and Supplementary Fig. 40. In our previous experiments, LiCl exhibited the lowest deliquescence humidity and the highest solubility (1.97 mol/100 mL water). Therefore, when the humidity was controlled at 40%, the dissolution of NaCl (0.61 mol/100 mL water) and KCl (0.46 mol/100 mL water) was still significantly suppressed. Theoretically, this method could be applied to separate other salts with similar properties, such as ZnBr2 and CaBr2, KCl and CH3COOK, LiCl and CoCl2, and LiCl and SrCl2. The deliquescence humidity of FeCl3 is below 11%, while that of AlCl3 is around 40%. We mixed LiCl–FeCl3–AlCl3 with NaCl (Li+ = 0.98 wt.%, Na+ = 87.38 wt.%, Fe3+ = 7.86 wt.%, and Al3+ = 0.038 wt.%), maintained the humidity at 16% for 3 h, and then measured the composition of the liquid phase. The results showed that, after separation, the composition of the solution was Li+ = 15.33 wt.%, Na+ = 0.26 wt.%, Fe3+ = 81.72 wt.%, and Al3+ = 2.69 wt.% (Supplementary Fig. 41). Therefore, substances with low deliquescence humidity and those with high deliquescence humidity can be separated from each other.

Furthermore, in addition to water vapor, controlling the vapor pressure of organic solvents can also be used for the screening of different cations (Fig. 5f). By selecting the appropriate solvent type and vapor pressure, highly selective separation of target ions can be achieved.

Discussion

The humidity-controlled ion separation method efficiently extracts lithium and magnesium from solid mixed salts derived from ore or brine, using water vapor through physical processes. We systematically studied the effects of humidity, adsorption time, and solvent vapor types on the lithium and magnesium extraction performances. This method indicated an extremely high lithium capacity of 133.60 mg g-1, a significant improvement by 4−62 times compared to conventional lithium sieves. Meanwhile, the lithium extraction rate surpassed almost all existing methods. Additionally, a good stability was achieved, showing 78.37% capacity retention for 100 cycles. Importantly, costs (4954 USD ton-1 of lithium carbonate) and energy consumption (0.43 kWh kg-1 Li) have also been significantly reduced. Building on this, we hypothesize that this method could also be used to extract other elements such as cesium, cobalt, strontium, calcium, and zinc. Moreover, considering the Moon’s environment, where the lunar soil samples returned by China’s Chang’e-5 mission have confirmed the presence of water45, but the environment does not meet the high water demands of traditional mining methods, using water vapor could significantly conserve water, making lunar mining more feasible. Overall, this work provides a fast, low-cost, and environmentally sustainable approach to purify a wide range of cations.

Methods

Preparation of polyacrylonitrile-chitosan composite sorbents

Mix polyacrylonitrile (PAN), chitosan (CS), and N,N-dimethylformamide (DMF) in a ratio of 2 g: 1 g: 20 mL and stir thoroughly for 6 h to ensure that PAN is fully dissolved in DMF, forming a suspension containing CS. Then, use a dropper to slowly drop the suspension into 200 mL of ultrapure water to form sorbent, maintaining a height of 13 cm between the dropper outlet and the water surface. Subsequently, wash the sorbents three times with 400 mL of ultrapure water and twice with 200 mL of ethanol. Finally, dry the spheres under vacuum at room temperature to obtain dried sorbents.

Humidity-controlled lithium extraction device

The photo of the lithium extraction setup is presented in Supplementary Fig. 42a. The adsorption column consists of a cylindrical body and a fritted glass (G0, pore size ~120–160 μm), with a total length of ~20 cm and an inner diameter of about 3.5 cm (Supplementary Fig. 42b). When 20 g of LiCl-NaCl mixed salt and 0.5 g of adsorbent spheres are added, the packing height reaches ~2.4 cm (Supplementary Fig. 42c).

Filling amount of solid mixed salts and sorbent spheres

The total mass of the mixed salts and adsorbent is ~20.5 g (comprising 20 g of salts and 0.5 g of adsorbent). The volume of 20 g of mixed salt is ~19 mL, as shown in Supplementary Fig. 43a. After mixing 20 g of mixed salt with 0.5 g of polyacrylonitrile-chitosan adsorbent spheres using a vortex mixer, the total volume increases to ~22 mL (Supplementary Fig. 43b). As observed in Supplementary Fig. 43c, the adsorbent spheres are in close contact with the mixed salt following the mixing process, as highlighted by the red dashed circles. Therefore, we can roughly estimate that the volume of 0.5 g of adsorbent spheres is ~3 mL. In addition, this stacking method creates relatively large voids within the packed salts in the adsorption column, which allows the airflow to pass through smoothly (Supplementary Note 12, Supplementary Fig. 44). Meanwhile, it also enables the airflow to pass through all regions of the solid mixed salts (Supplementary Note 13, Supplementary Fig. 45).

Humidity control

Humidity control of water

Pass air through different concentrations of salt solutions to achieve the desired humidity (Supplementary Note 14). For example, to obtain air with a RH of 40%, start by adding 500 mL of saturated CH3COOK solution into a 1 L gas-washing bottle. Then, pass air through the bottle at a rate of 3 L min-1 and use a hygrometer to measure the humidity at the outlet. Initially, the outlet humidity will be <40% (since the equilibrium relative humidity (ERH) of saturated potassium acetate solution is around 26% at 25 °C). Next, slowly add ultrapure water into the gas-washing bottle while continuously monitoring the outlet humidity. Keep the system running for 1 h, ensuring the humidity remains stable. As water is added to the bottle, the concentration of CH3COOK decreases, which increases the ERH. Continue this process until the outlet RH stabilizes at 40%, at which point the specific humidity control is achieved.

Humidity control of ethanol

Dry air was passed through a 1 L gas-washing bottle containing 500 mL of anhydrous ethanol at a flow rate of 3 L -1min, generating air saturated with ethanol vapor.

Gas pressure in the adsorption column

We made a simple modification to the setup to enable a rough measurement of the internal gas pressure within the column, as shown in Supplementary Fig. 46a. Taking a gas flow rate of 3 L min-1 as an example (Supplementary Fig. 46b), the observed height difference in the U-tube was ~2.5 cm (Supplementary Fig. 46c, due to the camera angle, there may be some distortion in the image.), corresponding to a pressure of around 0.245 kPa, or ~0.0024 atm. This relatively low pressure indicates that the gas can pass through the adsorption column with minimal flow resistance. It also suggests that the internal pressure within the column is slightly above atmospheric pressure, about 1.0024 atm, but very close to atmospheric pressure.

Lithium extraction procedure

Mix the sorbent with the mixed salt in a mass ratio of 1:20, and place the mixture into a chromatography column (used as an adsorption column) that is 30 cm long with an outer diameter of 50 mm and equipped with a sintered glass filter. Use a glass rod to thoroughly stir and mix the sorbents with the mixed salt. Then, introduce humid air of specific humidity into the adsorption column (flow speed is 3 L min-1) to ensure sufficient deliquescence and adsorption of LiCl, temperature maintained at 28 °C. Stir the mixture with a glass rod for 1 min every hour to promote more thorough adsorption by the sorbents. After the adsorption process is complete, use an 18-mesh screen to separate the sorbents from the impurity salts, and purge the sorbents with nitrogen gas (The aim is to remove the impurities salt that stick to the surface). Finally, add the sorbents to 1000 mL of ultrapure water and let them stabilize for 12 h to recover the LiCl.

Viscosity measurement (using Ubbelohde viscometer)

The viscosity of LiCl solutions with different concentrations was measured at 28°C using an Ubbelohde viscometer. The LiCl solution was introduced into the viscometer, and the atmospheric outlet was sealed. A syringe was used to draw the solution above the upper bulb of the capillary section. After that, the outlet was opened to the atmosphere, and the time required for the liquid meniscus to pass between the two timing marks was recorded. Each measurement was repeated three times and the average value was taken. The viscosity was calculated by multiplying the average flow time by the viscometer’s calibration constant.

Solution ion concentration analysis

In the experiments, the concentration of all ions in the solution was measured using an Inductively Coupled Plasma Optical Emission Spectrometer (ICP-OES). Specifically, the concentrations of Li⁺, Na⁺, K⁺, Mg²⁺, and Ca²⁺ in all samples were measured using a Thermo Scientific ICP-OES iCAP 7400. To prevent dilution-related errors, multiple samples with varying dilution ratios were tested for each measurement.

Adsorption capacity, lithium recovery, and selectivity

After lithium extraction, to ensure the Li+ are fully leached out, the sorbents are placed in 1000 mL of ultrapure water and left to stand for 12 h. Subsequently, different dilutions of the sample are measured using ICP-OES to avoid errors caused by dilution. The lithium adsorption capacity is calculated as the total mass of Li+ leached into the solution divided by the weight of the sorbent (Eq. (1)), as follows:

$${{{\rm{Adsorption}}}}\; {{{\rm{capacity}}}}({{{\rm{mg}}}}\;{{{{\rm{g}}}}}^{-1})=\frac{{{{\rm{Total}}}}\; {{{\rm{mass}}}}\; {{{\rm{of}}}}\; {{{\rm{Li}}}}\; {{{\rm{in}}}}\;{{{\rm{re}}}}{{{\rm{covery}}}}\; {{{\rm{solution}}}\;}\left({{{\rm{mg}}}}\right)}{{{{\rm{Total}}}}\; {{{\rm{mass}}}}\; {{{\rm{of}}}}\; {{{\rm{sorbent}}}}\left({{{\rm{g}}}}\right)}$$
(1)

Since the desorption rate of lithium from the adsorbent reaches as high as 99.76% (Supplementary Note 4, Supplementary Fig. 7), the adsorption capacity calculated using this method is accurate.

The recovery is determined by dividing the total mass of Li in the recovery solution by the total mass of Li in the solid mixed salt before lithium extraction (Eq. (2)).

$$\begin{array}{c}{{{\rm{Lithium}}}}\; {{{\rm{recovery}}}}(\%)=\frac{{{{\rm{Total}}}}\; {{{\rm{mass}}}}\; {{{\rm{of}}}}\; {{{\rm{Li}}}}\; {{{\rm{in}}}}\; {{{\rm{recovery}}}}\; {{{\rm{solution}}}}}{{{{\rm{Total}}}}\; {{{\rm{mass}}}}\; {{{\rm{of}}}}\; {{{\rm{Li}}}}\; {{{\rm{in}}}}\; {{{\rm{mixing}}}}\; {{{\rm{salt}}}}\; {{{\rm{before}}}}\; {{{\rm{extraction}}}}}\end{array}$$
(2)

Selectivity is calculated by dividing the percentage of Li/M in the post-extraction solution by the percentage of Li/M in the solid salt (Eq. (3)), where M represents Li, Na, K, Mg, and Ca.

$$\begin{array}{c}{{{\rm{Selectivity}}}}=\frac{\frac{{{{\rm{Li}}}}}{{{{\rm{M}}}}}{{{\rm{wt.}}}}\%\; {{{\rm{in}}}}\; {{{\rm{solution}}}}}{\frac{{{{\rm{Li}}}}}{{{{\rm{M}}}}}{{{\rm{wt.}}}}\%\; {{{\rm{in}}}}\; {{{\rm{mixing}}}}\; {{{\rm{salt}}}}}\end{array}$$
(3)

The sorbent utilization is calculated by dividing the amount of lithium absorbed by the sorbent in the solid salt by the absorption capacity of the sorbent in LiCl deliquescent solution at the corresponding humidity (Eq. (4)).

$$\begin{array}{c}{{{{\rm{Sorbent}}}}\; {{{\rm{utilization}}}}\;}\left(\%\right)=\frac{{{{\rm{adsorption}}}}\; {{{\rm{capacity}}}}\; {{{\rm{in}}}}\; {{{\rm{mixed}}}}\; {{{\rm{salt}}}}\; \left({{{\rm{mg}}}}{{{\cdot{\rm{g}}}}}^{-1}\right)}{{{{\rm{adsorption}}}}\; {{{\rm{capacity}}}}\; {{{\rm{in}}}}\; {{{\rm{LiCl}}}}\; {{{\rm{solution}}}}\; \left({{{\rm{mg}}}}{{{\cdot{\rm{g}}}}}^{-1}\right)}\end{array}$$
(4)

Isothermal adsorption test

The experimental procedure and conditions for the adsorption isotherm measurement are as follows: sorbent spheres with an initial mass of m0 were added to a LiCl solution with a known initial concentration (Ce) at 28 °C and left for 48 h to ensure adsorption equilibrium. The same batch of sorbent was sequentially used in solutions of different concentrations. After equilibrium was reached, the sorbent spheres were removed, and the surface solution was gently wiped off using kimwipes. The mass after adsorption (m1) was recorded. The difference m1-m0 represents the mass of solution adsorbed.

We measured the densities of LiCl solutions at different concentrations. Since the process is based on physical adsorption, the solution concentration remains constant at Ce during adsorption. The equilibrium adsorption capacity (qe) was calculated as follows:

$$\begin{array}{c}{{{{\rm{q}}}}}_{{{{\rm{e}}}}}=\frac{\left(\frac{{{{{\rm{m}}}}}_{1}-{{{{\rm{m}}}}}_{0}}{{{{\rm{\rho }}}}}\times {{{{\rm{C}}}}}_{{{{\rm{e}}}}}\right)}{{{{{\rm{m}}}}}_{0}}\end{array}$$
(5)

where ρ is the density of the LiCl solution. The adsorption isotherm curve was then constructed based on the relationship between qe and Ce.

3D structural characterization

After lithium extraction from the solid-state salt, the internal structure of the adsorbent spheres was characterized in three dimensions using a ZEISS Xradia 515 Versa high-resolution X-ray micro-computed tomography (micro-CT) system. A central cross-section of the sphere was selected for detailed analysis, as shown in Fig. 3f. The two-dimensional tomographic images were reconstructed into three-dimensional volumes using Avizo software, either automatically or manually, to facilitate data integration, visualization, and animation. The reconstructed 3D structure of the adsorbent spheres is presented in Supplementary Movies 2-4.

Materials characterizations

X-ray diffraction (XRD) measurements were performed using a Bruker D8 Advance diffractometer equipped with a Cu Kα radiation source. Scanning electron microscopy (SEM) and energy-dispersive X-ray spectroscopy (EDS) analyses were carried out using a JEOL JSM-7800F microscope. X-ray photoelectron spectroscopy (XPS) analysis was performed using a Thermo Fisher Scientific ESCALAB 250Xi system. Time-of-Flight Secondary Ion Mass Spectrometry (TOF-SIMS) was conducted using a ION-TOF GmbH TOF SIMS 5. Fourier transform infrared (FT-IR) spectroscopy was conducted using a Bruker V70 spectrometer. Brunauer–Emmett–Teller (BET) surface area measurements were carried out with a Quantachrome Instruments Autosorb-iQ-MP-AG. Contact angle measurements were performed using an instrument provided by Solon Information Technology Co., Ltd. (Shanghai, China).