WorksPhysics and Astronomy
Enhancing Direct Air Capture through Potassium Carbonate Doping of Activated Carbons
Abstract
Direct air capture of carbon dioxide (CO) is one of the most promising strategies to mitigate rising atmospheric CO levels. Among various techniques, adsorption using porous materials is a viable method for extracting CO from air, even under humid conditions. However, identifying optimal adsorbent materials remains a significant challenge. Moreover, the performance of existing materials can be improved by doping with active species that boost gas capture, a relatively unexplored field. In this study, we perform atomistic simulations to investigate the adsorption, structural, and energetic properties of CO and water in realistic models of activated carbons. We first analyze the impact of explicitly considering surfaces containing functional groups, which aims to imitate the chemical environment of experimental samples. Additionally, we introduce potassium carbonate within the pores of the adsorbent to evaluate its effect on CO and water adsorption. Our results demonstrate that both functional groups and potassium carbonate enhance adsorption, primarily by shifting the adsorption onset pressures to lower values. Specifically, potassium carbonate clusters act as extra adsorption sites for CO and water, facilitating the nucleation of water molecules and promoting the formation of a hydrogen bond network within the activated carbon pores.
Introduction
The ever increasing levels of carbon dioxide in the atmosphere of our planet is one of the largest environmental concerns of our time.[1] is one of the major contributing greenhouse gasses, and its rise is strongly correlated to global warming and anthropogenic climate change.[2] This has led to, and is increasingly leading to, many adverse effects such as: increases in droughts, desertification, melting of icecaps and permafrost, rising of sea levels, increase in floods, increasing probabilities of extreme weather events, and destruction of ecosystems.[3, 4, 5] However, our dependency on fossil-fuels and other processes emitting as a byproduct are not feasible to be replaced easily. Emissions are expected to continue to increase due to growing demands, while alternatives cannot keep up or are economically infeasible.[5, 6, 7] Current trends show that atmospheric concentration is heading to an average of 420 ppm in 2024, which is up from about 280 ppm in pre-industrial times, and is expected to exceed 550 ppm by 2050.[2, 8] To be able to reduce our emissions to acceptable levels according to climate goals, ‘negative emissions’ from capturing and storing from the atmosphere are required on top of proposed reductions.[7, 9] Among various methods of carbon capture and storage technologies being worked on for this purpose, our focus is on Direct Air Capture (DAC).
A promising method for DAC is the use of solid regenerable sorbents made of alkali-metal carbonates. This material can capture by a reaction with in the presence of water at room temperature, creating potassium bicarbonate:
A complete reverse reaction can be achieved at temperatures of 150 °C, which is relatively low.[10] Although the actual process is slightly more complex due to the formation of hydrates with , the adsorbing and heating process to desorb forms the basis of carbon capture. This process can match the energy efficiency of current commercial liquid amine processes in flue gas treatment. Compared to them, solid is practically beneficial, is thermally stable, can capture at atmospheric concentrations, and is more environmentally friendly in use and disposal.[10, 11, 12, 13, 14] Other solid sorbents that utilize amine functionalities exist; however, the anticipated limited lifetime of these materials makes them less interesting.
The must be deposited on some porous support structure to increase the surface area of the reaction, to prevent attrition, and to improve regeneration.[4, 14] The choice of this support is very influential; the pore structure, pore size distribution (PSD), and surface groups are all important to determine the capture capacity and kinetics.[13] These properties are important to create a fast adsorption technology in which the sorbent can be saturated quickly, which is desired for large-scale industrial applications. Activated carbons have become a popular material type choice for this owing to the tunability of both these aspects. Moreover, carbons are cheap because they are created from organic waste material, are thermally stable, and have a high surface area—all of which are desirable properties.[12] Finding a perfect carbon is difficult for a variety of reasons. By far, one of the greatest challenges is that of excessive adsorption of water over in these doped carbons. Potassium carbonate is naturally hygroscopic, and surface functional groups additionally increase water adsorption. This is a major hindrance in the viability of a sorbent; Desorbing excess water requires significant energy, which is detrimental to the technology commercially and reduces the net amount of captured considering what is exhausted generating energy for the process. Lowering the ratio is therefore the goal of a sorbent and process combination, for which adsorption and its kinetics must be known to improve this.
The addition of on activated carbons has been proposed for capture and separation applications.[15, 16, 17, 18, 19] However, the physical and chemical properties of all these sorbents and the external conditions of temperature and relative humidity affect the adsorption, and its mechanisms are not well understood at the molecular scale. Molecular simulations allow for exploring these types of matter and are an increasingly more prevalent and valuable tool for screening and understanding carbon capture.[20, 21] Relevant to the question at hand, this includes research on improving the selectivity of over water (and other gases) in many classes of porous materials. [22, 23, 24] Doping, modifying the surfaces, or exchanging ions in the case of zeolites have seen good success with these techniques.[25, 26, 27, 28, 29] For using , research explaining its capture mechanisms is proceeding, but, to our best knowledge, none directly focus on its direct inclusion in carbons to study adsorption in realistic environments for carbon capture.[30, 31] As such, in this paper, we present two methods of including in carbon models and explore the effects on adsorption and its mechanisms of and water separately using Monte Carlo simulations. For carbon, a model of the purely microporous activated carbon CS1000a is used. This specific carbon was chosen due to the accuracy of the model and the availability of two versions, which allows for an analysis between the presence and absence of surface functionalized groups.[32]
Methodology
Monte Carlo simulation
We used classical molecular simulations to assess the adsorption properties of and water in the activated carbon CS1000a, dependent on the presence of surface functionalization and potassium carbonate. For this purpose, Monte Carlo simulations in the grand canonical ensemble (GCMC) were used to compute single adsorbate isotherms and heats of adsorption at various loadings.[33] Specifically, we employ configurational bias Monte Carlo using the RASPA software package.[34] Unless otherwise specified, all components have translation moves; and if they are also molecules, they have rotation moves. The adsorbates additionally have moves for their complete reinsertion and moves to either add or remove them from the system. Moves are always chosen with equal probability.
Equilibration and sampling proved challenging because of the complexity of adsorbate loading in the system. This is mainly due to water, which is known to form clusters at active sites.[35, 36] This occurs in simulation, which causes a wide variance in the cycles needed to equilibrate and properly average over cases with potentially slow and large fluctuations in the sizes of these clusters. Therefore, an approach was used where we monitored the loading status for each simulation after a short initial equilibration run of cycles. The cycles were run repeatedly in blocks in the range of – cycles, depending on the equilibration of the system in each operating condition. Subsequently, the gathered adsorption isotherms were fitted to a multisite version of the Sips model using the RUPTURA software for interpolation and interpretation.[37, 38] Further information on this, including fitting parameters for every isotherm, is included in the supplementary materials (S1).
For the implementation of the force fields, the two models of CS1000a with differing surface chemistry were implemented as they were by Madero-Castro et al.[32] The structures were originally obtained using a hybrid reverse Monte Carlo (HRMC) technique by Jain et al., based on the experimental structure of an activated carbon from pyrolyzing sucrose at 1000 °C.[39, 40] From this method, no oxygen-containing surface groups are present that are responsible for the partial hydrophilicity of the realistic activated carbon.[41] To this end, a version of the structure was created using the method by Peng et al. by outfitting it with carboxylic acid (-COOH) and alcohol groups (-OH) to closely match experimental O/C and H/C ratios of 0.0087 and 0.091, respectively.[39, 42, 43] This version with functional groups, henceforth referred to as CS1000a, is compared to the version with no surface functionalization, from now on referred to as CS1000a. In modeling, both structures are considered rigid frameworks and form cubic unit cells of systems with edges of 50 Å.
The water model used is the extended simple point charge (SPC/E) model by Berendsen et al.[44] It has been used in a previous study on CS1000a for its ability to predict the structure and thermodynamics of bulk liquid water at ambient temperature as well.[41] We used the force field for reported by García-Sánchez et al.,[45] and for potassium carbonate, we used a model by Jo and Banerjee, based on the CVFF force field.[46] It describes the as a carbonate polyatomic ion and two separate ions. This force field was chosen since it was created to be applied in the modeling of interactions with carbon nanotubes[46] We verified its implementation and use for more general applicability by computing radial distribution functions (RDF) at 300 K. We compared the RDFs created by the model with the RDF results of Ding et al., which proved agreeable[47] These RDFs and more information on this verification have been provided in the supplementary materials S2. For nonbonded interactions, all models use Lennard-Jones potentials to represent the van der Waals interactions and Coulombic potentials with partial charges centered on the atoms to represent the electrostatic interactions. Lennard-Jones cross-interaction parameters are determined with Lorentz-Berthelot mixing rules.[48] These interactions have a cutoff of 12 Å, and do not use any form of tail corrections. For longer-range electrostatic interactions, Ewald summations are used.[49] to assess behavior and stability.
Modelling CS1000a with potassium carbonate
For the studied mechanism of carbon capture, potassium carbonate is needed in the pores. As far as we are aware, there is no precedent in the literature on doping activated carbons with potassium carbonate in computational studies. Hence, we explore two different methods of adding KCO within the pores of ACs, and we explore the effects on adsorption and its mechanisms, both for CO and for water separately.
The easiest method to add KCO is by simply arbitrarily adding several molecules of it into the pores of the model. However, this method does not result in configurations that one would expect when considering experimental considerations. That is, during impregnation in porous structures, the potassium carbonate deposited on the surfaces can be left in the form of crystal clusters.[50, 51] As such, it is of interest to create structures with small clusters of KCO and compare the results with those when randomly adding molecules. Furthermore, considering the effects of the quantity of KCO is also insightful. To this end, starting from each structure, we prepared four new initial configurations: 5 randomly inserted KCO, 10 randomly inserted KCO, a single cluster made up of 5 KCO, and two clusters with 5 KCO. For reference, the mass percentages of KCO are summarized in Table I.
| CS1000a | CS1000a | |
| 5 KCO | 1.23% | 1.25% |
| 10 KCO | 2.43% | 2.46% |
Table I. Mass percentages of KCO in the simulated structures.
First, we generated a framework with randomly distributed KCO where the CO and K ions were first arbitrarily inserted in a 1:2 ratio. To create these molecules, a MC simulation was run, where both components were allowed to keep their regular set of MC moves, but K was additionally allowed to undergo random translations. This ensured a quick formation of molecules and an opportunity to get into equilibrium with its environment. Alternatively, we inserted the KCO ions forming small clusters. To create a realistic structure of small clusters, the AMS driver program from SCM was used.[52] First, a cluster of 5 molecules was selected from the lattice of a potassium carbonate crystal. This cluster was then optimized under vacuum conditions using the Machine Learning potential M3GNet.[53]
Subsequently, the optimized clusters were inserted into the pores and considered to be completely rigid. Using only non-bonded interactions as defined by the KCO model, they were relaxed with a 5000 cycle MC simulation. During this, the clusters were allowed to translate and rotate, enabling them to settle next to the walls. For the frameworks in which two clusters were inserted, the clusters were ensured to end up far apart from each other to minimize their interactions. Hereafter, the rigid bonds from the clusters were removed, switching definitions to the initially defined model of KCO. The local environment of the carbon and a change in the model have definite effects on the shape of the just decoupled clusters. Therefore, another round of equilibration is required for them to settle properly. For this, MC simulations of steps were performed using the normal set of MC moves.
Results and Discussion
Effects of surface chemistry
Before adding the effects of potassium carbonate, it is worthwhile to analyze the adsorption in pristine carbon models. This is to understand and review the mechanisms of water and CO adsorption by comparing their dependence on the surface functionality without added species. Figure 1 shows a comparison of the computed adsorption isotherms on CS1000a and CS1000a for CO (a) or water (b) at 300.15 K. Here, symbols indicate the data obtained from the MC simulations, while the lines are fits through these data created using RUPTURA.[38] For CO, both structures show the same behavior where adsorption increases continuously over the computed pressure range while never reaching saturation. Evidently, the loading in CS1000a is significantly greater than CS1000a, specially at carbon capture working conditions. Taking as reference 420 ppm atmospheric CO, at atmospheric pressure this equates to partial pressure conditions of 42.55 Pa.[8] Using the fitted functions for interpolation, CS1000a (2.937 × 10 mol/kg) shows about a two-order-of-magnitude increase in adsorption over CS1000a (5.251 × 10 mol/kg). For reference, this pressure has been marked in Figure 1a with a vertical line. However, the concentration of adsorbed CO under these conditions is still too low to be used in practical applications, and this emphasizes the need to combine activated carbons with sorbents such as potassium carbonate to enhance CO₂ capture.

Figure 1. Comparisons of computed adsorption isotherms of CO₂ (a) and H₂O (b) in CS1000af (solid lines) and CS1000anf (dashed lines) at 300.15 K. Lines are fits through the data points obtained using RUPTURA.[38] The vertical line in (a) indicates the partial pressure of CO₂ equivalent to 420 ppm CO₂ (at 1 atm), its current approximate atmospheric concentration.[8] The vertical line in (b) represents the saturation pressure at the simulation temperature.
The results clearly indicate that the functionalized groups create more favorable adsorption conditions. For visualization of adsorption behavior, renderings were made using VMD software at different stages in the adsorption process.[54, 55] Figures 2a-c display representative snapshots of CO₂ adsorption in CS1000af at low, medium and high pressure values, respectively. There is relatively little adsorption compared to water (Figure 2d-Figure 2f), and the molecules appear to be randomly distributed. The effects of the functionalized groups are

Figure 2. Snapshots showcasing adsorption in CS1000af at 300.15 K under various partial pressure conditions. (a-c) Adsorption of CO at 316, 10000, and 316228 Pa respectively. (e-f) Adsorption of HO at 357, 1784, and 316228 Pa respectively. These conditions for water correspond to 10% relative humidity, 50% relative humidity, and saturation.
most clear when compared with the renders of CS1000anf, which have been provided in Figures S4a-c. At matching pressures, these carbons show considerably less CO₂ adsorption.
For water, the computed isotherms in Figure 1b differ considerably between CS1000af and CS1000anf. Their shapes and differences can be understood through the generally accepted water filling mechanism in microporous carbons. Starting at low pressure, water adsorbs at the functional groups due to their polarity. Subsequently, water starts adsorbing on the already adsorbed water through hydrogen bonding. This forms clusters that grow with increasing pressure, filling up the pores. After this, the loading increases continuously with increasing pressure.[35, 36, 41] These mechanisms are very clearly visualized in the renders of CS1000af going through Figure 2d-Figure 2f. A few molecules adsorbed at 10% relative humidity (RH), as described, grow large clusters concentrated at 50% RH. At the highest simulated RH conditions, the structure is fully saturated.
Without functional groups, the adsorbent is completely hydrophobic, so water has no place to form the initial hydrogen bonds to start nucleating. Due to the pore size, no nucleation can start either.[35, 36, 41] This results in no adsorption of water until well above the saturation pressure for water, the pressure of which has been indicated for the used simulation temperature in the Figure 1b with a vertical line. After some point, the pores practically instantly fill, and the loading becomes equal to that of CS1000af. This lack of adsorption into a jump is also visible in renders of the structure in Figures S4d-f. For carbon capture, the relevant range of conditions is up to 100% RH as this constitutes the conditions where water remains as vapour in the air. The adsorption of water below its vapour saturation pressure in CS1000anf is negligible, the amount of water captured by CS1000af is significant under the same conditions. An analogous study was presented by Billemont et al.[41] Though they use different operating conditions and a different molecular model for CO₂, our results and conclusions align with their work. In this context, this paper provides a more direct discussion with regard to DAC and will be used as a baseline for when K₂CO₃ is included.
Observations of an increase in adsorption strength due to functionalized groups are confirmed by the heat of adsorption curves depicted in Figure 3. Here, the horizontal dash-dotted lines represent the heat of vaporization of the respective components. At 300.15 K these are 17.0 kJ/mol for CO₂ and 43.9 kJ/mol for water.[56, 57] At low coverage, both CO₂ and water show heats of adsorption of a greater magnitude compared to CS1000anf. The trends for CS1000anf differ between the components. For CO₂, due to van der Waals interactions with the walls, slightly more favorable heats of adsorption are observed for adsorption. For water, due to the hydrophobicity of the surface of CS1000anf, it instead stays above -10 kJ/mol until a sudden jump toward the heat of vaporization at high loading. The other trends show a curve to wards their respective heats of vaporization. It is worth mentioning that the noise observed in the heats of adsorption of water at high loading is expected because of the difficulty of inserting additional molecules within the hydrogen bond network created by the adsorbed water molecules. However, the results clearly indicate that, on average, the heats of adsorption of water in the large pores of CS1000a converge to the vaporization enthalpy of water.[58]. For additional validation of this finding, a qualitative comparison can be made with the heats of adsorption of methanol in CS1000a, a polar molecule that behaves similarly to water. Madero et al. [32] compared the experimental and computed heats of adsorption of methanol in CS1000a, showing how similar to water the heat of adsorption converges to the enthalpy of vaporization of this molecule.

Figure 3. Comparisons of computed heats of adsorption of CO (a) and HO (b) in CS1000a (solid lines) and CS1000a (dashed lines) at 300.15 K. The horizontal dash-dotted lines represent the heats of vaporization of the respective components.
The effects of temperature were also investigated for the pristine carbon models at 288.15 and 313.15 K. For reference, results for the isotherms and heats of adsorption are available in Figure S2 and Figure S3 respectively in the supplementary materials. As expected, we observe that the isotherms shift properly when the temperature is modified, and the heats of adsorption show mostly overlapping trends, tending toward the heats of vaporization.

Figure S2. Comparisons of computed adsorption isotherms for differing temperatures (288.15, 300.15, and 313.15 K). Figures (a,b) present results for , (c,d) for (c,d), with (a,c) being in CS1000af, and (b,d) in CS1000anf. Lines are fits through the data points obtained using RUPTURA.[38] The black vertical line in the isotherms indicate the partial pressure equivalent to 420 ppm (at 1 atm), its current approximate atmospheric concentration.[8] The vertical lines in the isotherms represent the saturation pressures corresponding to their colors.

Figure S3. Comparisons of computed heats of adsorption for differing temperatures (288.15, 300.15, and 313.15 K). Figures (a,b) present results for , (c,d) for (c,d), with (a,c) being in , and (b,d) in . The horizontal dash-dotted lines represent the heats of vaporation of the respective components at 300.15 K. Heats of evaporation at the other temperatures do not change appreciably, and are therefore not shown.
Effects of potassium carbonate
After the analysis of the CO and water adsorption in the pristine adsorbents, we study the effect of doping the ACs with KCO species. We used two approaches to dope the adsorbents with KCO as described in the methodology. Figure 4 schematically recaps the insertion of random KCO ions and their insertion after forming small clusters. After the KCO ions are relaxed within the ACs pores, we computed the adsorption isotherms for CO and water 300.15 K. Figure 5 shows the results alongside comparisons with the pristine CS1000af carbon. As realistic carbons contain surface functionalization, the results for CS1000anf are shown in the supplementary materials Figure S5. They are still useful as they isolate the effects of the potassium carbonate, because there are no other active adsorption sites within the pores as in the case of the structures with functional groups. The computed water isotherms this time were limited to the 0–100% relative humidity range, covering the pressure range relevant to atmospheric conditions.

Figure 4. Schematic explanation of the two routes to create structures containing KCO. In the first case, the KCO units are randomly added to an empty structure. In the second case, previously optimized small clusters of 5 KCO were initially added to the pores. Then, the ions of KCO are relaxed within the structure to accommodate the clusters next to a wall correctly.

Figure 5. Comparisons of computed adsorption isotherms of CO (a-c) and HO (d-f) in CS1000af at 300.15 K for varying amounts of KCO. Figures (a,d) show isotherms in which the KCO was added randomly into the carbon, while (b,e) show isotherms in which the KCO was added in clusters of 5. Lines in these Figures are fits through the data points obtained using RUPTURA.[38] In Figures (c,f), isotherms using both methods of KCO inclusion are shown, scaled by the isotherm without KCO. The black vertical line in the CO isotherms indicates the partial pressure equivalent to 420 ppm (at 1 atm), its current approximate atmospheric concentration.[8]

Figure S5. Comparisons of computed adsorption isotherms of (a-c) and (d-f) in CS1000 at 300.15 K for varying amounts of . Figures (a,d) show isotherms in which the was added randomly into the carbon, while (b,e) show isotherms in which the was added in clusters of 5. Lines in these figures are fits through the data points obtained using RUPTURA.[38] In figures (c,f), isotherms using both methods of inclusion are shown, scaled by the isotherm without . The black vertical line in the (scaled) isotherms indicate the partial pressure equivalent to 420 ppm (at 1 atm), its current approximate atmospheric concentration.[8]
Figure 5a and Figure 5b show the CO adsorption isotherms with KCO randomly added or in cluster form, respectively, compared to the isotherm of the pristine carbon. Regardless of the insertion method, increasing the KCO content enhances adsorption. Quantitatively, increases in KCO lead to consistent increases in adsorption regardless of the presence of functional groups. This shows evidence that KCO is able to serve properly as an additional active sites for CO. The renderings of the adsorption behavior in Figure 6a-Figure 6d confirm this finding. These images visualize the CO adsorption in CS1000a with 10 KCO at various pressures. They indeed show significant quantities of CO congregating around the KCO surfaces. The influence of the functional groups is still relevant as clearly more CO is present compared to the counterparts without these in Figure S4-Figure S4.

Figure 6. Snapshots showcasing adsorption of CO (a-d) and HO (e-h) in CS1000af at 300.15 K with 10 KCO added in clusters (a,b,e,f) or randomly (c,d,g,h) under various partial pressure conditions. CO is shown at 10000 Pa in (a,c) and at 316228 Pa in (b,d). HO is shown at 357 Pa (10% relative humidity) in (e,g) and at 1784 Pa (50% relative humidity) in (f,h).

Figure S4. Snapshots showcasing adsorption in CS1000 at 300.15 K under various partial pressure conditions. (a-c) Adsorption of CO at 316, 10000, and 316228 Pa respectively. (e-f) Adsorption of HO at 357, 1784, and 316228 Pa respectively. These conditions for water correspond to 10% relative humidity, 50% relative humidity, and saturation.
To make a comparison of the effectiveness of KCO in increasing adsorption, Figure 5c scales all CO isotherms from the carbons with KCO with the CO isotherm of the pristine carbon. No matter the amount of KCO or its manner of insertion, the resultant shape is qualitatively similar. All trends show a peak around the Pa range, indicating that this is where KCO is most effective in increasing CO adsorption as a general trend. This is probably related to the covering of most of the surface of KCO with a monolayer of CO. After this, higher pressures are needed for continued filling, which becomes increasingly more pronounced as less CO is able to set near the surfaces. The atmospheric conditions of carbon indicated by the vertical line are within the broader peak of increased effectiveness, thus clearly benefiting the application.
Figure 5c additionally provides a direct comparison between the methods of inserting KCO randomly or in clusters. Although the differences in effectiveness are not major, a minimal but definite difference in trends can be found. This is most pronounced for 10 KCO, which below Pa shows CS1000a with clusters adsorbing a slightly higher amount of CO. Above this pressure, the structure with randomly added KCO adsorbs more. Considering that potassium carbonate has a larger surface area when it has been inserted randomly, it is understandable that more is able to adsorb. At low pressures, not enough CO is adsorbed for this extra surface to matter, and clustered KCO molecules instead act as stronger active centers for superior CO adsorption. The structure with only 5 KCO shows minimal effects of surface area differences, and either method is about as effective. Only at the Pa peak does the cluster insertion method show better adsorption, while at higher pressures the random insertion method only has slightly preferential adsorption.
Figures 5d and 5e show the water adsorption isotherms with KCO randomly added or in cluster form, respectively, compared to the isotherm of pristine carbon. At
lower RH, large quantitative increases in loading are found as more KCO are present. This also tends towards an earlier onset of saturation. The effects are even more pronounced for CS1000a (Figure S5,Figure S5), revealing large gaps in onset to saturation between 5 and 10 KCO. The KCO appears to act hygroscopically properly and serves as additional adsorption sites for water to form hydrogen bonds with. This is confirmed by the computed radial distribution functions (RDFs) between water and O as discussed below. This means that they serve as additional locations for the water filling mechanism to form clusters around, which is evident from the structures modeled in Figure 6 e-Figure 6h. These show snapshots of water adsorption in CS1000a with 10 KCO at 10% RH and 50% RH for both methods of KCO insertion. Large clusters of water are present, including around the locations of KCO, which grow to near saturation at 50% RH. The renderings of CS1000a are shown in Figure 6 e-Figure 6h, which unambiguously display the adsorption due to the added KCO.
The adsorption mechanism also leads to large differences in effectiveness between the insertion methods. This is best observed in Figure 5f, which displays all water isotherms with KCO, scaled with the water isotherm of the pristine carbon. The extra surface area created by random insertions of KCO is significant to the point where 5 randomly added KCO molecules have water adsorption similar to that of 2 clusters below 40% RH. After this, it even outperforms due to the earlier onset of filling. These differences are also visible in the renders; there is a tangible increase in adsorption in the structures with randomly added KCO (Figure 6g,Figure 6h) compared to their counterparts which used cluster insertion (Figure 6e,Figure 6f).
With Figure 5f also provides insight into the effective increases in adsorption due to the addition of KCO. Regardless of the amount or method of insertion, the trends are similar. At the lowest pressure conditions, a greater increase in the water uptake is observed, showing preferential adsorption. After this, the relative increase is stable up to the point at which the structure starts to fill rapidly to reach saturation. This filling is earlier than in the pristine carbon, and thus shows up as peaks whose locations and shape are dependent on the amount and method of insertion. In general, the random insertion method leads to a slightly higher water uptake because there are more active adsorption sites that trigger the nucleation of water within the pores. After the point of pristine carbon reaches saturation, all loadings are nearly the same, resulting in a flat line at value one.
Finally, we investigated the microscopic structure of KCO and surrounding adsorbates in the pores by means of computing radial distribution functions. To this aim, we focused on the structure of CS1000a with 10 KCO in cluster formation, which can serve as a good representation of a realistic carbon containing both surface groups and potassium carbonate in clusters [50, 51]. Following the previously obtained adsorption isotherms, additional runs of cycles were performed after 5000 cycles of equilibration at low and high loadings to obtain the RDFs between ions and adsorbates. Figure 7 shows the RDFs between the ions of the clusters at two different concentrations of adsorbed molecules. Interactions involving KCO–KCO in systems without adsorbates are also included as a baseline for comparison. Figure 7a shows the RDFs between the carbon atoms of KCO under the adsorption of CO. With increased pressure and loading, only slight changes occur in peak heights and in the position of the peak at 5.3 Å. RDFs between different ions come to the same conclusions, changing only slightly at higher loading, as exemplified by the RDF between K ions in Figure 7b. These small changes indicate the shape of clusters is not affected by CO. This indicates that KCO plays an important role on enhancing CO adsorption, while preserving its microstructure while confined within the ACs pores. Water predictably has a more significant influence on the clusters. This is evident from both Figure 7c and Figure 7d, which contain the RDFs of C–C and K–K, respectively under water adsorption. Both figures show that the peaks jump and the shape changes in the RDFs when comparing the baseline without adsorbates to that with 3.62 mol/kg HO. These changes are most pronounced for the C ions, which reveal a tendency of these ions to move further apart. Changes that occur with increasing loading to 9.12 mol/kg HO are noticeable, but not as extreme. Under current conditions, although there are changes, these results do not contain major shifts, indicating cluster destabilization and breakup.
Figure 7. Computed radial distribution functions (RDFs) with KCO in CS1000af containing 2 clusters of 5 KCO in the presence of CO (a,b) and water (c,d) at 300.15 K. The RDFs compare the effects of these adsorbates on the structure of the KCO cluster at different loading. RDFs are shown between C – C (a,c) and K – K (b,d).
Figure 8 shows the RDFs between the adsorbates and the cluster ions with increasing loading. The same trend is observed in all cases. With increasing pressure, the peaks shrink, but the locations of all peaks remain the same. This scaling effect is explained simply by the amount of adsorbates present in the system, which makes only a smaller proportion able to be near the clusters. We thus conclude that there are no significant changes in the distribution of atoms around the clusters in either case.

Figure 8. Computed radial distribution functions (RDFs) between and adsorbates in containing 2 clusters of 5 in the presence of (a,b) and water (d,e) at 300.15 K. For , RDFs are shown between (a) and (b). For water, RDFs are shown between (d) and (e). Snapshots of closeups are shown for (c) and water (f) in with 2 clusters to visualize the loading behavior clearly.
The RDFs between the adsorbates and KCO themselves shows a predominant first peak at distances lower than 4 Å that covers the first coordination shell of the added salt, even at very low adsorbate concentrations. This confirms the role of KCO ions being the most active adsorption sites for CO and water in doped ACs. For CO, the distribution is fairly homogeneous around the clusters. Here, RDFs show that distances between CO and O (Figure 8a,Figure 8b) are slightly smaller than between CO and K or C (Figure 7a-Figure 7d). The water RDFs show that HO acts as a hydrogen bond donor and O acts as a hydrogen bond acceptor. Peaks between HO and O in 8d,e clearly have a much larger amplitude at lower distances than those between HO and K or C in Figure S7-Figure S7. Further RDFs between the HO atoms themselves in Figure S8 show the hydrogen bonding between water molecules, which occurs when water clusters around active sites. Snapshots with close-ups of CO and water molecules adsorbing around the KCO clusters are shown in Figure 8c and Figure 8f to illustrate the observed behaviors of the RDFs. For visual clarity, these snapshots were made in CS1000anf, which isolates adsorption to the two clusters of KCO present.

Figure S7. Computed radial distribution functions (RDFs) between K₂CO₃ and adsorbates in CS1000a_f containing 2 clusters of 5 K₂CO₃ in the presence of CO₂ (a-d) and water (e-h) at 300.15 K. The RDFs compare the effects of increased pressure on the loading behavior. Under CO₂ adsorption, RDFs are shown between C_CO₂ − C_{K₂CO₃}^{4+} (a), C_CO₂ − K_{K₂CO₃}^{+} (b), O_CO₂ − C_{K₂CO₃}^{4+} (c), and O_CO₂ − K_{K₂CO₃}^{+} (d). Under water adsorption, RDFs are shown between H_H₂O − C_{K₂CO₃}^{4+} (e), H_H₂O − K_{K₂CO₃}^{+} (f), O_H₂O − C_{K₂CO₃}^{4+} (g), and O_H₂O − K_{K₂CO₃}^{+} (h).

Figure S8. Computed radial distribution functions (RDFs) between H₂O atoms in CS1000a_f containing 2 clusters of 5 K₂CO₃ at 300.15 K. These show hydrogen bonding between water molecules, evidencing the growth of water clusters through this mechanism.
Conclusion
In this work, we investigated the adsorption properties of activated carbon sorbents doped with potassium carbonate (KCO) for direct air capture (DAC) applications. This process captures CO through a reversible reaction with water and KCO, releasing the stored compounds at elevated temperatures. The amounts of adsorbed CO and water in the carbon pores under specific operating conditions are critical to the performance of this reaction. To gain insights at the molecular level, we employed Monte Carlo simulations to study CO and water adsorption in a representative activated carbon, CS1000a.
To assess the effect of pore surface chemistry, we considered two models of CS1000a: one with and one without surface functional groups. The presence of functional groups enhances the adsorption of both CO and water. While CO uptake increases slightly across the entire pressure range—without altering the overall shape of the isotherm—the impact on water adsorption is more pronounced. Functional groups serve as nucleation sites for water, facilitating pore filling at pressures below water’s saturation pressure.
We also explored two methods of incorporating KCO into the pores: random insertion of CO and K ions, and the addition of small KCO clusters. These approaches resulted in subtle differences in adsorption mechanisms. Random insertion creates more adsorption sites throughout the pore volume, while clusters act as strong interaction centers, particularly at low pressures. Higher KCO loadings enhance the adsorption of both adsorbates. Radial distribution functions (RDFs) and structural analyses revealed that CO tends to form monolayers on KCO surfaces. In the carbon model with functional groups, CO uptake increases by a factor of 3–5× compared to the pristine carbon under DAC conditions (420 ppm CO at atmospheric pressure).
Water interactions with KCO are even stronger than those of CO, due to water’s polar nature. Water forms extended hydrogen-bonded clusters around adsorption sites. As a result, the additional surface area from randomly inserted KCO significantly promotes water adsorption, which leads to earlier pore filling. Water nucleation through hydrogen bonding is the primary mechanism of adsorption, though interactions with KCO also slightly alter its structural arrangement, as indicated by the RDFs. Depending on the KCO insertion method and concentration, water uptake increases by a factor of 1.5–3× until reaching saturation.
These findings clarify the molecular mechanisms governing CO and water adsorption in activated carbons with large pores, emphasizing the roles of surface functional groups and embedded KCO. Further work is needed to investigate the chemical reactions involving CO, water, and KCO. In addition, activated carbons with different pore structures and surface chemistries should be examined to optimize CO capture under ambient conditions and limit excessive water uptake that may block the pores. Achieving the right balance between CO and water adsorption is crucial for enhancing the reactive capture process and improving current direct air capture technologies.
Supporting Information
The supporting information contains the necessary information to recreate all fits and the verification of the potassium carbonate model. Additional figures included are computed isotherms and heats of adsorption at different temperatures, isotherms for with , snapshots for , more RDFs between and adsorbates in with clusters, and RDFs between atoms in with clusters.
Competing interests
The authors declare no competing interests.
Fit information
For interpolation of the isotherm data, the isotherms are fitted using the RUPTURA software.[38] With it, we fit to a multi-site version of the Sips model.[37] The equation for this isotherm is given by
where is the number of sites for the fit. For site , is the saturation loading, is a heterogeneity factor, and is an equilibrium constant. The fitting parameters obtained for all isotherms in the study are provided in Table S2.
| Adsorbate | Framework | T [K] | KCO | State | |||||||||
| CO | CS1000a | 288.15 | 0 | - | 268.828 | 1.38801 | 1.22793 | 84.8083 | 8.6869 | 2.43502 | - | - | - |
| CO | CS1000a | 300.15 | 0 | - | 13596100000.0 | 2.84265 | 1.46273 | 0.03125 | 0.250001 | 2.87783 | - | - | - |
| CO | CS1000a | 313.15 | 0 | - | 0.0179896 | 520320.0 | 0.369437 | 12855500000.0 | 4.57488 | 1.43076 | - | - | - |
| CO | CS1000a | 288.15 | 0 | - | 1.7031 | 1.01808 | 1.00196 | 2.61121 | 4.64167 | 0.404941 | - | - | - |
| CO | CS1000a | 300.15 | 0 | - | 1.40375 | 8.81056 | 0.999719 | 2.67864 | 3.42886 | 0.480607 | - | - | - |
| CO | CS1000a | 313.15 | 0 | - | 0.946109 | 4.47819 | 0.320378 | 2.20404 | 3.9838 | 0.99978 | - | - | - |
| HO | CS1000a | 288.15 | 0 | - | 9.99289 | 0.000956674 | 0.374432 | 31.2495 | 0.00102457 | 0.0445515 | 19.9972 | 9.57408 | 2.93968 |
| HO | CS1000a | 300.15 | 0 | - | 1.38188 | 8.39528 | 32.3697 | 35.3644 | 0.000523475 | 0.0351599 | 17.2498 | 0.00011119 | 1.32998 |
| HO | CS1000a | 313.15 | 0 | - | 5.56278 | 5.18107 | 9.78818 | 33.4967 | 0.000245162 | 0.0421983 | 16.0021 | 8.42108 | 0.999371 |
| HO | CS1000a | 288.15 | 0 | - | 1.77099 | 7.47617 | 0.292032 | 47.2963 | 0.000136922 | 0.0726278 | 4.44898 | 1.36593 | 0.604888 |
| HO | CS1000a | 300.15 | 0 | - | 3.9837 | 3.22697 | 0.77107 | 44.5314 | 0.000136447 | 0.0699203 | 6.49791 | 3.268 | 0.577798 |
| HO | CS1000a | 313.15 | 0 | - | 46.2306 | 3.88966 | 0.0541569 | 4.81713 | 1.06071 | 0.377755 | - | - | - |
| CO | CS1000a | 300.15 | 5 | random | 9.21091 | 1.55896 | 0.95973 | 0.00706898 | 4.36171 | 0.497889 | 1.3026 | 4.66398 | 2.09134 |
| CO | CS1000a | 300.15 | 5 | clusters | 0.0147388 | 461549.0 | 0.227748 | 0.185653 | 0.0071673 | 1.29103 | 132849.0 | 3.06468 | 1.53769 |
| CO | CS1000a | 300.15 | 10 | random | 898.99 | 2.80833 | 1.72671 | 0.492482 | 0.00103027 | 1.64134 | 0.00994535 | 15717500.0 | 0.582258 |
| CO | CS1000a | 300.15 | 10 | clusters | 0.054664 | 0.0109155 | 0.823911 | 3.33096 | 3.35559 | 2.56939 | 5.37919 | 2.71963 | 0.730417 |
| CO | CS1000a | 300.15 | 5 | random | 1.19784 | 3.54875 | 1.49932 | 2.87602 | 5.09154 | 0.52059 | 0.0294278 | 0.0327476 | 0.26237 |
| CO | CS1000a | 300.15 | 5 | clusters | 1.76909 | 7.86441 | 1.49079 | 2.34509 | 4.7797 | 0.441332 | 0.104001 | 0.0165566 | 0.983537 |
| CO | CS1000a | 300.15 | 10 | random | 0.101647 | 0.0330879 | 0.299067 | 1.03646 | 0.000164145 | 1.57597 | 4.76986 | 3.56738 | 0.715113 |
| CO | CS1000a | 300.15 | 10 | clusters | 3.95357 | 3.80472 | 0.62092 | 0.245232 | 0.00403509 | 1.18782 | 1.11226 | 2.32895 | 1.95332 |
| HO | CS1000a | 300.15 | 5 | random | 10.0559 | 0.000728463 | 0.288344 | 26.9863 | 0.000616398 | 0.0815502 | 17.9595 | 2.3479 | 4.6894 |
| HO | CS1000a | 300.15 | 5 | clusters | 3.205 | 0.081493 | 2.16001 | 10.8897 | 0.000636007 | 0.409244 | 28.5783 | 0.000577098 | 0.0384296 |
| HO | CS1000a | 300.15 | 10 | random | 24.844 | 0.000637759 | 0.0843337 | 138.659 | 1.27962 | 5.60692 | 10.0312 | 0.00079509 | 0.358133 |
| HO | CS1000a | 300.15 | 10 | clusters | 18.0159 | 0.000543396 | 0.345302 | 20.2467 | 0.00059274 | 0.0382297 | 9.4785 | 0.000540827 | 3.68507 |
| HO | CS1000a | 300.15 | 5 | random | 27.1104 | 0.000319319 | 0.0456701 | 13.6949 | 0.00039854 | 0.143852 | 7.9201 | 2.8328 | 4.73668 |
| HO | CS1000a | 300.15 | 5 | clusters | 16333.2 | 4.73563 | 0.461951 | 36.2897 | 0.000303782 | 0.0311494 | 6.60539 | 1.97649 | 5.14148 |
| HO | CS1000a | 300.15 | 10 | random | 10.5485 | 6.81628 | 5.46445 | 7.58773 | 0.000472452 | 0.38534 | 30.1085 | 0.0000494304 | 0.0387433 |
| HO | CS1000a | 300.15 | 10 | clusters | 2.14465 | 0.000484455 | 0.26035 | 6.55638 | 6.40015 | 5.26029 | 36.1044 | 0.000336004 | 0.0439293 |
Table S2. Multi-site Langmuir Sips fitting parameters for pure component adsorption of carbon dioxide and water in CS1000a.
As may be noted, there are places where the fits do not have the appropriate data points available in order to fully predict the correct behavior of the isotherms. This is especially pertinent for the water isotherms, where various data points are unavailable due to the lengthy equilibration process. This leaves many possible fits, primarily for the pressures range where the carbon fills quickly. However, all fits were carefully created such that the unequilibrated points lay slightly beneath the lines shown.
RDF potassium carbonate verification
We computed RDFs of bulk KCO at 300 K with MC using the model in the work of Jo and Banerjee.[46] We compared these with the solid state results from Ding et al. who simulated RDFs with MD with using a different model.[47] First, before computing the RDFs, a simulation box was created with initial sides of 60 Å containing 192 CO and 384 K. The system was relaxed with an NPT simulation at 300 K and atmospheric pressure for cycles. Then the RDFs were measured under NVT conditions for cycles after 5000 equilibration cycles under the now fixed volume conditions.
The resultant RDFs are shown together with those traced from Ding et al. in figure S1[47]. Overall, comparable peaks are observed. Differences observed occur mostly with our results being shifted right as if the sizes of the ions are larger. In several places, the large peaks also show smaller peaks/perturbations not present in the reference. We note in the two Figures S1c and S1f containing the RDFs between C – O and O – O respectively, the first peaks are completely absent. This is merely because RASPA does not compute the RDF within a defined molecular unit, so in this case within the polyatomic ion CO the first strong peaks are not taken into account. To also not, due to the height of the C – O peak in the original work of Ding et al., the fidelity of the data is low and hence not as accurate. Disregarding the matters based on these points, we conclude the behavior is acceptable.

Figure S1. Computed radial distribution functions (RDFs) between elements of at 300 K. The solid red lines are RDFs computed with Monte Carlo simulations using the model in the work of Jo and Banerjee [46]. The dashed blue lines are results taken directly from Ding et al., who performed Molecular Dynamics simulations using a different model. Shown here are RDFs for (a) $C - C$, (b) $C - K$, (c) $C - O$, (d) $K - K$, (e) $O - K$, and (f) $O - O$. [47]
Figures

Figure S6. Snapshots showcasing adsorption of CO (a-d) and HO (e-h) in CS1000anf at 300.15 K with 10 KCO added in clusters (a,b,e,f) or randomly (c,d,g,h) under various partial pressure conditions. CO is shown at 10000 Pa in (a,c) and at 316228 Pa in (b,d). HO is shown at 357 Pa (10% relative humidity) in (e,g) and at 1784 Pa (50% relative humidity) in (f,h).
References
- [1]N. M. Dowell, P. S. Fennell, N. Shah, and G. C. Maitland, The role of co2 capture and utilization in mitigating climate change, Nature Climate Change 7, 243 (2017).DOI
- [2]M. R. Raupach, G. Marland, P. Ciais, C. L. Quéré, J. G. Canadell, G. Klepper, and C. B. Field, Global and regional drivers of accelerating co2 emissions, Proceedings of the National Academy of Sciences 104, 10288 (2007).pubmed.ncbi.nlm.nih.gov/17519334
- [3]K. M. K. Yu, I. Curcic, J. Gabriel, and S. C. E. Tsang, Recent advances in co2 capture and utilization, Chem-SusChem 1, 893 (2008).DOI
- [4]G. Yang, H. Luo, T. Ohba, and H. Kanoh, Co 2 capture by carbon aerogel–potassium carbonate nanocomposites, International Journal of Chemical Engineering 2016, 1 (2016).
- [5]J. gang Lu, M. dong Cheng, Y. Ji, and H. Zhang, Membrane-based co2 absorption into blended amine solutions, Journal of Fuel Chemistry and Technology 37, 740 (2009).DOI
- [6]J. Conti, P. Holtberg, J. Diefenderfer, A. LaRose, J. T. Turnure, and L. Westfall, International energy outlook 2016 with projections to 2040, Tech. Rep. (USDOE Energy Information Administration (EIA), Washington, DC (United States), 2016).DOI
- [7]E. S. Sanz-Pérez, C. R. Murdock, S. A. Didas, and C. W. Jones, Direct capture of co2 from ambient air, Chemical Reviews 116, 11840 (2016).pubmed.ncbi.nlm.nih.gov/27560307
- [8]X. Lan, P. Tans, and K. W. Thoning, Trends in globally-averaged co2 determined from noaa global monitoring laboratory measurements., version 2023-12.
- [9]IPCC, AR6, Climate change 2013: the physical science basis, Contribution of working group I to the fifth assessment report of the intergovernmental panel on climate change 1535 (2013).
- [10]N. N. A. H. Meis, A. M. Frey, J. H. Bitter, and K. P. de Jong, Carbon nanofiber-supported k2co3 as an efficient low-temperature regenerable co 2 sorbent for post-combustion capture, Industrial & Engineering Chemistry Research 52, 12812 (2013).
- [11]C. Zhao, X. Chen, and C. Zhao, Carbonation and active-component-distribution behaviors of several potassium-based sorbents, Industrial & Engineering Chemistry Research 50, 4464 (2011).DOI
- [12]Q. Wang, J. Luo, Z. Zhong, and A. Borgna, Co2 capture by solid adsorbents and their applications: current status and new trends, Energy Environ. Sci. 4, 42 (2011).DOI
- [13]J. Wang, L. Huang, R. Yang, Z. Zhang, J. Wu, Y. Gao, Q. Wang, D. O’Hare, and Z. Zhong, Recent advances in solid sorbents for co 2 capture and new development trends, Energy Environ. Sci. 7, 3478 (2014).DOI
- [14]N. Masoud, G. Bordanaba-Florit, T. van Haasterecht, and J. H. Bitter, Effect of support surface properties on co 2 capture from air by carbon-supported potassium carbonate, Industrial & Engineering Chemistry Research 60, 13749 (2021).DOI
- [15]N. Querejeta, F. Rubiera, and C. Pevida, Enhanced capacity to co2 sorption in humid conditions with a k-doped biocarbon, Journal of Energy Chemistry 34, 208 (2019).DOI
- [16]A. Ali and R. Idris, Utilization of low-cost activated carbon from rapid synthesis of microwaved sorbents for wc nanoparticles preparation, Advanced Materials Letters 8, 70 (2017).DOI
- [17]S. Kundu, T. Khandaker, M. A.-A. M. Anik, M. K. Hasan, P. K. Dhar, S. K. Dutta, M. A. Latif, and M. S. Hossain, A comprehensive review of enhanced co2 capture using activated carbon derived from biomass feed-stock, RSC Adv. 14, 29693 (2024).
- [18]M. M. Korah, K. Culp, K. S. Lackner, and M. D. Green, Activated carbon fiber felt composites for the direct air capture of carbon dioxide, ChemSusChem 18, e202401188 (2025).DOI
- [19]Y. Guo, J. Sun, R. Wang, W. Li, C. Zhao, C. Li, and J. Zhang, Recent advances in potassium-based adsorbents for co2 capture and separation: a review, Carbon Capture Science & Technology 1, 100011 (2021).DOI
- [20]T. E. Akinola, P. L. Bonilla Prado, and M. Wang, Experimental studies, molecular simulation and process modelling/simulation of adsorption-based post-combustion carbon capture for power plants: A state-of-the-art review, Applied Energy 317, 119156 (2022).DOI
- [21]B. Wadi, A. Golmakanl, T. N.Borhani, V. Manovic, and S. A. Nabavi, Molecular simulation techniques as applied to silica and carbon-based adsorbents for carbon capture, Energies 16, 5013 (2023).DOI
- [22]Y. Liu and J. Wilcox, Molecular simulation studies of co2 adsorption by carbon model compounds for carbon capture and sequestration applications, Environmental Science & Technology 47, 95–101 (2012).
- [23]E. Di Biase and L. Sarkisov, Molecular simulation of multi-component adsorption processes related to carbon capture in a high surface area, disordered activated carbon, Carbon 94, 27–40 (2015).DOI
- [24]L. Jiang, R. Xie, W. Shi, E. Wu, B. Li, and X. Zhang, Water effect on adsorption carbon capture in metal-organic framework: A molecular simulation, Carbon Capture Science & Technology 4, 100061 (2022).DOI
- [25]R. Babarao, S. Dai, and D.-e. Jiang, Nitrogen-doped mesoporous carbon for carbon capture – a molecular simulation study, The Journal of Physical Chemistry C 116, 7106–7110 (2012).DOI
- [26]A. Abdelrasoul, H. Zhang, C.-H. Cheng, and H. Doan, Applications of molecular simulations for separation and adsorption in zeolites, Microporous and Mesoporous Materials 242, 294–348 (2017).
- [27]F. Akbari Beni and M. Niknam Shahrak, Alkali metals-promoted capacity of zif-8 and zif-90 for carbon capturing: A molecular simulation study, Polyhedron 178, 114338 (2020).
- [28]C. Zhang, Y. Zhang, T. Su, B. Yuan, X. Zhang, L. Wang, Y. Tian, and J. Wang, Molecular simulation on carbon dioxide capture performance for carbons doped with various elements, Energy Storage and Saving 2, 435–441 (2023).DOI
- [29]M. Sadeghi, F. Esmaeilzadeh, D. Mowla, and A. Zandifar, Improving co2 capture in utsa-16(zn) via alkali and alkaline earth metal introduction: Gcmc and md simulations study, Separation and Purification Technology 338, 126534 (2024).
- [30]H. Liu, Q. Qin, R. Zhang, L. Ling, and B. Wang, Insights into the mechanism of the capture of co2 by k2co3 sorbent: a dft study, Physical Chemistry Chemical Physics 19, 24357–24368 (2017).DOI
- [31]T. Cai, M. Wang, X. Chen, Y. Wu, J. Ma, and W. Zhou, Operando modeling and measurements: Powerful tools for revealing the mechanism of alkali carbonate-based sorbents for co2 capture in real conditions, Frontiers in Energy 17, 380–389 (2023).DOI
- [32]R. M. Madero-Castro, J. M. Vicent-Luna, X. Peng, and S. Calero, Adsorption of linear alcohols in amorphous activated carbons: Implications for energy storage applications, ACS Sustainable Chemistry & Engineering 10, 6509 (2022).DOI
- [33]D. Dubbeldam, A. Torres-Knoop, and K. S. Walton, On the inner workings of monte carlo codes, Molecular Simulation 39, 1253 (2013).DOI
- [34]D. Dubbeldam, S. Calero, D. E. Ellis, and R. Q. Snurr, RASPA: molecular simulation software for adsorption and diffusion in flexible nanoporous materials, Molecular Simulation 42, 81 (2016).DOI
- [35]J. K. Brennan, T. J. Bandosz, K. T. Thomson, and K. E. Gubbins, Water in porous carbons, Colloids and Surfaces A: Physicochemical and Engineering Aspects 187–188, 539 (2001).DOI
- [36]J. Alcañiz-Monge, A. Linares-Solano, and B. Rand, Water adsorption on activated carbons: Study of water adsorption in micro- and mesopores, The Journal of Physical Chemistry B 105, 7998 (2001).
- [37]R. Sips, On the structure of a catalyst surface, The Journal of Chemical Physics 16, 490 (1948).DOI
- [38]S. Sharma, S. R. G. Balestra, R. Baur, U. Agarwal, E. Zuidema, M. S. Rigutto, S. Calero, T. J. H. Vlugt, and D. Dubbeldam, RUPTURA: simulation code for breakthrough, ideal adsorption solution computations, and fitting of isotherm models, Molecular Simulation 49, 893 (2023).DOI
- [39]S. K. Jain, J.-M. Pellenq, J. P. Pikunic, and K. E. Gubbins, Molecular modeling of porous carbons using the hybrid reverse monte carlo method, Langmuir 22, 9942 (2006).pubmed.ncbi.nlm.nih.gov/17106983
- [40]X. Peng and S. K. Jain, Understanding the influence of pore heterogeneity on water adsorption in realistic molecular models of activated carbons, The Journal of Physical Chemistry C 122, 28702 (2018).DOI
- [41]P. Billemont, B. Coasne, and G. D. Weireld, Adsorption of carbon dioxide, methane, and their mixtures in porous carbons: Effect of surface chemistry, water content, and pore disorder, Langmuir 29, 3328 (2013).DOI
- [42]X. Peng, S. K. Jain, and J. K. Singh, Adsorption and separation of N2 /CH4 /CO2 /SO2 gases in disordered carbons obtained using hybrid reverse monte carlo simulations, The Journal of Physical Chemistry C 121, 13457 (2017).
- [43]X. Peng, J. M. Vicent-Luna, and Q. Jin, Separation of CF4 /N2 , C2F6 /N2 , and SF6 /N2 mixtures in amorphous activated carbons using molecular simulations, ACS Applied Materials & Interfaces 12, 20044 (2020).
- [44]H. J. C. Berendsen, J. R. Grigera, and T. P. Straatsma, The missing term in effective pair potentials, The Journal of Physical Chemistry 91, 6269 (1987).DOI
- [45]A. García-Sánchez, C. O. Ania, J. B. Parra, D. Dubbel-dam, T. J. H. Vlugt, R. Krishna, and S. Calero, Transferable force field for carbon dioxide adsorption in zeolites, The Journal of Physical Chemistry C 113, 8814 (2009).
- [46]B. Jo and D. Banerjee, Effect of solvent on specific heat capacity enhancement of binary molten salt-based carbon nanotube nanomaterials for thermal energy storage, International Journal of Thermal Sciences 98, 219 (2015).DOI
- [47]J. Ding, L. Du, G. Pan, J. Lu, X. Wei, J. Li, W. Wang, and J. Yan, Molecular dynamics simulations of the local structures and thermodynamic properties on molten alkali carbonate k2co3, Applied Energy 220, 536 (2018).DOI
- [48]D. Boda and D. Henderson, The effects of deviations from lorentz–berthelot rules on the properties of a simple mixture, Molecular Physics 106, 2367 (2008).
- [49]P. P. Ewald, Die berechnung optischer und elektrostatischer gitterpotentiale, Annalen der Physik 369, 253 (1921).DOI
- [50]T. Cai, X. Chen, J. Zhong, Y. Wu, J. Ma, D. Liu, and C. Liang, Understanding the morphology of supported na2co3/γ-alooh solid sorbent and its co2 sorption performance, Chemical Engineering Journal 395, 124139 (2020).DOI
- [51]N. Masoud, V. Clement, T. van Haasterecht, M. Führer, J. P. Hofmann, and J. H. Bitter, Shedding light on solid sorbents: Evaluation of supported potassium carbonate particle size and its effect on co 2 capture from air, Industrial & Engineering Chemistry Research 61, 14211 (2022).DOI
- [52]R. Rüger, M. Franchini, T. Trnka, A. Yakovlev, E. van Lenthe, P. Philipsen, T. van Vuren, and T. S. B. Klumpers, AMS 2023.1, SCM (2023).
- [53]C. Chen and S. P. Ong, A universal graph deep learning interatomic potential for the periodic table, Nature Computational Science 2, 718 (2022).DOI
- [54]W. Humphrey, A. Dalke, and K. Schulten, VMD – Visual Molecular Dynamics, Journal of Molecular Graphics 14, 33 (1996).pubmed.ncbi.nlm.nih.gov/8744570
- [55]J. Stone, An Efficient Library for Parallel Ray Tracing and Animation, Master’s thesis, Computer Science Department, University of Missouri-Rolla (1998).
- [56]R. M. Stephenson and S. Malanowski, Handbook of the Thermodynamics of Organic Compounds (Springer Netherlands, 1987).DOI
- [57]The Engineering ToolBox (2010), Water - Heat of Vaporization vs. Temperature, [online] Available at: https://www.engineeringtoolbox.com/water-properties-d_1573.html [Accessed 10-04-2024].
- [58]X. Peng, J. M. Vicent-Luna, S. K. Jain, Q. Jin, and J. K. Singh, Computational study of the effect of functional groups on water adsorption in mesoporous carbons: Implications for gas adsorption, ACS Applied Nano Materials 2, 7103 (2019).DOI
