adsorption and diffusion of water, methanol, and …...the adsorption and diffusion of water and...

Adsorption and Diffusion of Water, Methanol, and Ethanol in All-Silica DD3R:Experiments and Simulation

Jelan Kuhn,*,† Juan Manuel Castillo-Sanchez,† Jorge Gascon,‡ Sofia Calero,§

David Dubbeldam,⊥ Thijs J. H. Vlugt,† Freek Kapteijn,‡ and Joachim Gross*,†

Process & Energy Laboratory, Delft UniVersity of Technology, Leeghwaterstraat 44, 2628CA Delft, TheNetherlands, DelftChemTech, Delft UniVersity of Technology, Julianalaan 136, 2628BL Delft, The Netherlands,Department of Physical, Chemical, and Natural Systems, UniVersity Pablo de OlaVide, Ctra. Utrera km 1,41013 SeVilla, Spain, and Chemical and Biological Engineering Department, Northwestern UniVersity,2145 Sheridan Road, EVanston, Illinois 60208

ReceiVed: February 20, 2009; ReVised Manuscript ReceiVed: June 23, 2009

This study comprises both experimental measurement as well as molecular modeling of the adsorption ofwater, methanol, and ethanol on the hydrophobic all-silica decadodecasil 3R (DD3R) zeolite. The simulationdata are compared with permeation data of these components, measured using a DD3R membrane underpervaporation conditions. The pure-component isotherms are measured by vapor-phase adsorption and arecalculated by grand canonical Monte Carlo (GCMC) simulations. The simulations are conducted usingmolecular models from the literature, without any adjustments. The simulation results are in fair agreementwith the experimental adsorption data. Computed mixture adsorption isotherms show that the water loadingis significantly increased as compared to pure-component adsorption. The shape of the computed water isothermchanges from type II to type I when alcohol is present. The loading of both methanol and ethanol at lowfugacities is slightly enhanced by the presence of water. The self-diffusivites are calculated using MDsimulations. The self-diffusivity of water in DD3R is substantially larger than the diffusivites of both alcoholsbut is decreased in the presence of methanol or ethanol. The ethanol diffusivity is too low to be determinedby conventional MD simulations. The flux calculated from the MD and GCMC results overpredicts theexperimental values, but the predicted membrane selectivity corresponds well to the experimental data. Themechanism of separation in the dewatering of alcohols using a hydrophobic DD3R membrane is based onthe high diffusivity of water as compared to that of the alcohols, where especially ethanol suffers from diffusionlimitations.

Introduction

Small-pore zeolites have gained increasing interest in thefields of membrane technology and sensor applications. All-silica DD3R has been successfully applied for CO2/CH4

separation1-3 and dewatering of alcohols and other organicsolvents.4,5 Also, hydroxy sodalite (SOD) membranes have beensuccessfully synthesized and tested for water separation.6,7

Because of the pronounced size selectivity of these small-porezeolites, they are also well-suited for sensor applications.Dehydration of alcohols is already performed on an industrialscale using LTA-type zeolite membranes.8,9 To design andoptimize these types of membrane separation processes, under-standing and modeling the behavior of water and alcohols inzeolite membranes is of key importance.

The DDR zeolite topology consists of window-connectedcages, which makes this material a member of the group ofclathrasils.10 The crystal structure is built by corner-sharing SiO4

tetrahedra that are connected to pseudohexagonal layers of face-sharing pentagonal dodecahedra (512 cages). These layers are

stacked in an ABCABC sequence and are interconnected byadditional SiO4 tetrahedra that form six-membered rings betweenthe layers. Thus, two new types of cages arise, a smalldecahedron, 435661 cage, and a large 19-hedron, 435126183 cage(Figure 1). By connecting the 19-hedra cavities through a single8-ring with an aperture of 4.4 × 3.6 Å, a two-dimensionalchannel system is formed, which is accessible to small moleculesafter detemplation by thermal treatment at 773-973 K orozonication at 473 K.11 The small decahedron and pentagonal

* To whom correspondence should be addressed; E-mail: [email protected] (J.G.).

† Process & Energy Laboratory, Delft University of Technology.‡ DelftChemTech, Delft University of Technology.§ University Pablo de Olavide.⊥ Present address: University of Amsterdam, Van’t Hoff Institute for

Molecular Sciences, Nieuwe Achtergracht 166, 1018 WV Amsterdam, TheNetherlands.

Figure 1. The DDR topology unit cell (a) normal to and (b) down onthe [001]plane; (c) the window opening, and (d) the decahedron,pentagonal, and 19-hedra cages that build up the structure. Taken fromref 15.

J. Phys. Chem. C 2009, 113, 14290–1430114290

10.1021/jp901869d CCC: $40.75 2009 American Chemical SocietyPublished on Web 07/15/2009

cages are only accessible through four-ring and five-ringwindows; therefore, they are inaccessible for sorbed molecules.Three different materials with the DDR topology have beenreported in the literature, all-silica decadodecasil 3R (DD3R),10

sigma-1,12 and ZSM-58.13 For the synthesis of DD3R and sigma-1, 1-adamantanamine (1-ADA) is used as a structure-directingagent (SDA), while methyltropinium iodide (MTI) is used asthe SDA in the synthesis of ZSM-58. A DDR unit cell containssix 19-hedra cavities, each of which has a volume of ap-proximately 0.35 nm3.14

NGK insulators (Japan) have developed a high-silica DD3Rzeolite membrane (Si/Al ratio ) 980) consisting of a 2 µmzeolite layer deposited on an asymmetric R-alumina support.16,17

These membranes have been successfully tested in gas separa-tion experiments, for example, CO2/CH4,2,3,18 and in thedehydration of solvents and acids.4,5 Adsorption of several gasesand vapors on DD3R powder was measured. Zhu et al. reportedadsorption of several saturated and unsaturated hydrocarbonson DD3R.14,19 The DD3R zeolite exhibits selective adsorptionof 1,3-butadiene and trans-2-butene over 1-butene and cis-2-butene, revealing the size and shape selectivity of the DD3Rzeolite.14 Den Exter et al. investigated the adsorption of wateron DD3R20 and showed that the zeolite adsorbs water only toa very small extent, confirming its hydrophobic character.Nevertheless, the DD3R membranes have shown to removewater from alcohols at high selectivity under pervaporationconditions.4,5 Pervaporation is a mode of operation in membraneseparation processes (Figure 2) where a liquid mixture is fed toa membrane while a low pressure or gas purge is applied to thepermeate side. One or more components from the feed mixtureselectively permeate through the membrane and evaporate intothe vapor phase (hence, the word “pervaporation”). In an earlierstudy,5 we reported pure-component permeation fluxes of water,methanol, and ethanol as well as mixtures of these componentsat temperatures ranging from 344 to 398 K. The mixturemeasurements were conducted at liquid molar feed fractions ofwater of xw ) 0.1-0.5, leading to water/ethanol separationfactors of up to Rw,e ) 1500 and Rw,m ) 9 for water/methanol.

The behavior of water in hydrophobic zeolites is complex,and simulations on water adsorption and permeation throughsmall-pore hydrophobic zeolite membranes, such as DDR, arenot extensively studied. Molecular simulations of the adsorptionand diffusion of several compounds in the DDR structure arereported, but these involve mainly gaseous components, suchas CH4, CO2, and N2.21-23 There is a large number of simulationstudies on water in hydrophobic zeolites, although the systemsconsidered are nearly exclusively silicalite-1.24 One of the fewstudies focused on water in a hydrophobic small-pore zeolite isreported by Coudert et al., who studied the behavior of adsorbedwater in an all-silica LTA zeolite by means of the Car-Parrinellomolecular dynamics method (CPMD).25 However, so far, no

studies are reported on molecular modeling of water andalcohols in hydrophobic DDR-type zeolites.

Demontis et al. demonstrated the complex temperature-dependent behavior of water in silicalite-1 by moleculardynamics (MD) simulations including full flexibility of boththe water molecules as well as the zeolite framework.26 Desbienset al. were able to reproduce experimental isotherms of waterin silicalite-1 using simple models in GCMC simulations,applying the Tip4P and MSPC/E potentials for water and a rigidzeolite framework.27 Similar interaction potentials have also beenused to study the adsorption of water and aromatics in thehydrophilic NaY zeolite.28 Puibasset et al. used the PN-TrAZwater model in GCMC simulations of water in silicalite-1;however, the SPC model describing the confined water had tobe adjusted to allow for a quantitative agreement with experi-mental data.29 Fleys et al. used the Compass force field toreproduce water adsorption isotherms on silicalite-1 and dealu-minated zeolite Y.30,31 The underprediction of the water loadingat low pressures was attributed to crystalline defects, resultingin silanol groups within the zeolite framework. Trzpit et al.studied the effect of local defects on water adsorption insilicalite-1, enabling qualitative reproduction of the experimen-tally observed adsorption behavior.32

The adsorption of methanol and ethanol is less comprehen-sively studied by computational methods. Usually, this adsorp-tion is studied in the presence of water. The works of Vigner-Maeder et al.33 and Pelmenschikov et al.34 are two pioneeringab initio studies of methanol adsorption in silicalite-1 wherethe adsorption energies, enthalpies of adsorption, influence ofdefects, and potential maps are calculated. Hinderer et al.35

proposed a Monte Carlo scheme together with a latticerepresentation of silicalite-1 to study the diffusion of methanol,including the possibility of chemical reactions in the system.Plant et al. studied water/methanol diffusion in a NaY zeolite,describing the temperature dependence, the hydrogen bonding,and the interaction with the cations.36 Gupta et al. studied water/ethanol adsorption in silicalite-1, indicating the preferential sitingof different molecules.37 Lu et al. simulated the adsorption andseparation performance of water and methanol mixtures in MFI,MOR, CFI, and DON zeolites.38

Several studies compared the zeolite membrane performancewith molecular simulations. Takaba et al. employed dualensemble Monte Carlo (DEMC) simulations to obtain theadsorption and diffusivities of water, ethanol, and binarymixtures in pervaporation across a silicalite-1 membrane.39 Jiaet al. studied pervaporation of different components and binarymixtures in silicalite-1, zeolite A, and chabazite membranes byMD simulations.40 Furukawa et al. used nonequilibrium mo-lecular dynamics (NEMD) simulations to study the adsorptionof ethanol/water mixtures in a hydrophilic NaA membrane,suggesting that the high selectivity for water is a cooperativeeffect of the adsorption selectivity and the high diffusivity ofwater within the zeolite.41 Yang et al. studied methanol/water42

and ethanol/water43 permeation across silicalite-1 membranesby combining GCMC and MD simulations in the NVTensemble. They were able to qualitatively predict the separationfactor and fluxes, while the discrepancies between simulationsand measurements were attributed to the existence of crystallinedefects in the experimental systems. Modeling an idealizedzeolite will always give a discrepancy with real systems sincezeolite crystals often consist of complex intergrown structureswith internal grain boundaries and defects containing silanolgroups and varying pore orientations.44

Figure 2. Schematic representation of a pervaporation process.

Water, Methanol, and Ethanol in All-Silica DD3R J. Phys. Chem. C, Vol. 113, No. 32, 2009 14291

This study is focused on measuring and simulating theadsorption of water, methanol, and ethanol on an all-silica DD3Rzeolite to elucidate the mechanism of water transport andseparation in DD3R zeolite membranes. To the best of ourknowledge, there is no force field that quantitatively describesthe adsorption and diffusion of water and alcohols in DDR-type zeolite. The development of a force field that accuratelydescribes the adsorption of water in hydrophobic zeolites issubject to severe difficulties, as adsorption is extremely sensitiveto the force field parameters, the atomic positions of theframework atoms, and the partial charges of the frameworkatoms.45 Therefore, the objective of the simulation workpresented in this article is to provide a qualitative explanationof the phenomena found experimentally using force fields takendirectly from the literature. The calculated mixture adsorptionisotherms are supplemented with self-diffusivities obtained byMD simulations and used to predict the component fluxes andselectivities in alcohol dehydration under pervaporation conditions.

Experimental Methods

Synthesis. To perform adsorption experiments, DD3R crystalswere prepared by the method described by Gascon et al.46 Allchemicals were purchased from Aldrich and used as received,without further preliminary purification. The synthesis mixturehad the molar ratio of 1-ADA/SiO2/ethylenediamine (EDA)/H2O ) 47:100:404:11240, and it was prepared as follows; 1.72 gof 1-ADA (97 wt %) was dissolved in 5.76 g of EDA (g99 wt%) and was placed in an ultrasonic bath for 1 h. Water (38 g)was then added rapidly. The mixture was placed in a shakingmachine for 1 h. After heating to 368 K for 1 h while stirring,the mixture was cooled down in ice. Aerosil 200 (1.41 g) wasthen slowly added while the solution was stirred vigorously.The mixture was again heated to 368 K in an open vessel whilestirring until the solution became clear (the solution changedfrom white to almost transparent). As-synthesized DD3R crystals(0.01 g) were added as seeds. The resulting solution was placedin a Teflon-lined autoclave and heated under hydrothermalconditions (autogenous pressure) to 433 K for 2 days underrotation at 60 rpm. The synthesized samples were calcined at873 K for 6 h using a heating and cooling rate of 60 K ·h-1.The calcination was carried out in static air (Carbolite CSF1100).

The synthesized DD3R crystals were homogeneous diamond-like crystals, as reported by den Exter et al.20 The size of thecrystals was 2.7 µm, measured by laser diffraction, and themicropore volume of 0.13 cm 3 ·g-1 was determined by nitrogenadsorption at 77 K using the t-plot method.11

Adsorption. The synthesized zeolite crystals were used forthe adsorption measurements of water, ethanol, and methanol.The single-component vapor adsorption (Quantachrome Au-tosorb-6B) experiments were conducted to determine theadsorption isotherms at different temperatures. Prior to thevapor-phase adsorption measurements, the samples were driedin vacuum at 573 K in order to remove moisture and othervolatile components (the weight of the sample after drying wasused in the calculations). The adsorption and desorptionisotherms were measured at 273, 303, and 333 K, which is themaximum temperature for vapor adsorption in this setup.

Analysis of Permeation Experiments. Pure-component andmixture permeation of water, ethanol, and methanol werepreviously reported.5 These permeation measurements wereconducted under pervaporation conditions, applying vacuumwith a permeate pressure of pperm ) 1 kPa. In this study, weuse the pure-component measurements as well as measurementsof the binary alcohol/water mixtures for comparison and

validation of the calculated adsorption data and for calculationof diffusivity data.

The Maxwell-Stefan (M-S) equations are generally usedto model mass transport in zeolites and zeolite membranes.47

They have been successfully applied to several systems.48-50

These equations balance the driving forces with the frictionexerted on molecules and are consistent with the theory ofirreversible thermodynamics.51,52 In the case of transport througha porous membrane, the membrane can be viewed as acomponent, M. Using the membrane as a frame of reference(choosing the membrane velocity, υM ) 0), the M-S equationsfor an n-component system can be written as53

where ∇µi,T denotes the chemical potential gradient at constanttemperature as the driving force for mass transport, Ni denotesthe molar flux of component i, ÐiM is the M-S diffusivity ofcomponent i in the microporous membrane, Ðij is the M-Scross-diffusivity that accounts for the interaction betweenpermeating components, and θi denotes the fractional loading,defined as the ratio of the loading qi over the saturation loadingqi

sat

In small-pore zeolites, the interaction between species withinthe zeolite can be neglected (i.e., Ðij ) 0).54 When only transportin the z-direction is considered, eq 1 reduces to

The chemical potential gradient can be converted to a gradientin loading using the thermodynamic correction factor, Γij, whichrelates the fractional loading θi to the fugacity fi

47

where

When the driving force is expressed as the gradient in loading,(dqi/dz) ) qi

sat(dθi/dz), the flux can be expressed as53

where in Fz denotes the density of the zeolite. We assume themembrane thickness to be sufficiently thin so that the dif-ferentials in loading can be replaced by differences over the

-Fz

θi∇µi,T

RT) ∑

j)1j*i

n (qjNi - qiNj)

qisatqj

satÐij

+Ni

qisatÐiM

(1)

θ )qi

qisat

(2)

-θi

RT

dµi,T

dz)

Ni

FzqisatÐiM

(3)

θi

RT

dµi,T

dz) Γij

dθi

dz(4)

Γij ≡θi

fi

∂fi

∂θj)

qjsat

qisat

qi

fi

∂fi

∂qji, j ) 1, n (5)

Ni ) -FzÐiM[Γii∇qi + Γij

qisat

qjsat

∇qj] i * j (6)

14292 J. Phys. Chem. C, Vol. 113, No. 32, 2009 Kuhn et al.

membrane. By taking an average value of Γij (i, j ) 1, 2, ..., n),eq 6 simplifies to

where δ denotes the membrane thickness and ∆qi ) qiperm -

qifeed. The value of Γij is approximated by averaging the value

obtained at the permeate and the feed side conditions

The temperature dependence of the self-diffusivity at constantloading is generally described by an Arrhenius type of equation,using an energy of activation, EA.

In pervaporation processes, the liquid feed side fugacity, fifeed,

can be calculated using55

where xifeed is the liquid molar fraction and γi denotes the activity

coefficient in the mixture, p is the system pressure, and Vm,il is

the partial molar volume of component i in the liquid feed. TheAntoine equation is used to calculate the vapor pressure, pi

sat,as a function of temperature.56 The fugacity coefficient φi

sat andthe Poynting factor (exponent term in eq 10) can be assumedto equal unity at pervaporation conditions. In this study, theactivity coefficients were calculated using the NRTL Gibbsexcess energy model.

To calculate the adsorption from a given liquid feed, thefugacities were calculated, and these were used as input forthe GCMC calculations. The composition can be expressed asthe ratio of the component fugacity over the sum of thefugacities of all species in the mixture

As the permeate side is operated at low pressures, it is wellapproximated by the ideal gas limit

and therefore, the fugacity ratio at the permeate side equals thevapor molar fraction, yi

perm.In experimental studies, the permeation is often described

by the permeance, Πi, defined as the ratio of the flux and thedifference in fugacity over the membrane.57 The temperaturedependency of the permeance is then characterized by anapparent energy of activation, Eapp, which can be seen as the

sum of the activation energy of diffusion and the enthalpy ofadsorption.58

Simulation Methods. Adsorption isotherms in confinedsystems are conveniently computed using GCMC simulations,in which the temperature, the volume, and the chemical potentialof the system are prescribed. The imposed chemical potentialis directly related to the fugacity, which is used as input for thesimulations. At high densities, the probability of successfullyinserting a molecule in the system is very low. Therefore, theinsertion and deletion of molecules in and from the system wereperformed using the configurational bias Monte Carlo (CBMC)technique59 in combination with other MC moves, namely,regrowth, rotation, and translation. In the case of ethanol/watermixtures, identity change moves were also used to increase theperformance of the simulation.60 More details on this simulationtechnique can be found elsewhere.61,62 The saturation loadingswere calculated from a single adsorption simulation at a highfugacity. The enthalpies of adsorption were computed as theisosteric heats of adsorption at zero loading using MC simula-tions in the NVT ensemble.63

There are several diffusion coefficients that can be used todescribe the motion of molecules within a zeolite. Generally,the Fick, or transport diffusivity, Di

T is the most intuitivedefinition of a diffusion coefficient, relating a molar flux to agradient in concentration.64 The actual thermodynamic drivingforce for transport is, however, the gradient in chemicalpotential.65 Using the framework of irreversible thermodynamics,the collective, corrected, or M-S diffusivity, Di, can be defined,relating the thermodynamic driving force to a molar flux.58 Theself-diffusivity, Di

s, describes the motion of individual particlesand can be related to the mean square displacement of theindividual molecules during equilibrium MD simulations by theEinstein relation.66 The M-S and the transport diffusivity canbe obtained from NEMD but also from equilibrium MDsimulations by considering the collective motion of molecules.66

In this study, water is modeled using the Tip5pEw potential.67

The alcohols are described with the TraPPE force field,68 whichwas also applied by Jakobtorweihen et al. for modeling ethene,propene, and trans-2-butene in DD3R.23 No hydrolysis isexpected during the adsorption of these molecules in zeolitesunder the conditions applied.36,69 Intramolecular potentials areincluded to describe the flexibility of alcohols, while the watermolecules are kept rigid. The bond lengths are fixed for allmolecules. Bond bending potentials are considered for methanoland ethanol, and a torsion potential is used for ethanol.68

Self-diffusivities were calculated from the mean squaredisplacement (MSD) of the molecules by the Einstein equation51

and using transition-state theory (TST). The MSD was obtainedby means of molecular dynamics (MD) simulations in thecanonical ensemble.59 The equations of movement were inte-grated using a Verlet algorithm with a time step of 0.5 fs. Thetemperature in the system was controlled with a Nose-Hooverchain thermostat (chain length of 3). In order to compute thediffusivity of water and methanol, we have performed MDsimulations of at least 10 ns. The square root of the mean squaredisplacement was at least 60 Å, and we verified that this issufficiently large to correctly compute the diffusivity. Thetransport diffusivities were calculated using the Onsager for-mulation, relating the diffusivity to the collective displacementof all molecules in the simulation box.66 Due to the slowdiffusion of the relatively large ethanol molecule within the DDRstructure, some of the simulated conditions are in the limit ofapplicability of the MSD method. Therefore, TST59 was usedto obtain a more accurate value of the ethanol diffusivity.

Ni ) -FzÐiM

δ [⟨Γii⟩∆qi + ⟨Γij⟩qi

sat

qjsat

∆qj] i * j (7)

⟨Γij⟩ )Γij

feed + Γijperm

2i, j ) 1, 2, ..., n (8)

Dis ) Di,0

s exp(-EA,i

RT ) (9)

fifeed ) xi

feedγipisat

φisat exp[Vm,i

l (p - pisat)

RT ] (10)

Yi )fi

∑j)1

n

fj

(11)

fiperm ) pi

perm ) yipermpperm (12)


The values of Γij can be directly obtained from the GCMCresults. In this study, we used a method inspired by the approachof Chen et al.70 to calculate Γij.

The crystallographic positions of the zeolite atoms wereobtained from the X-ray diffraction (XRD) characterization dataof the structure.71 The simulations were performed using 1 × 2× 1 unit cells with periodic boundary conditions, so that thesize of the simulation box was more than twice the Lennard-Jones cutoff radius. To reduce the calculation time, the simula-tions were performed using a rigid framework.72 Since the43.56.61 and the 512 cages of the DDR structure only containfive-membered ring windows, they are inaccessible for sorbedspecies. To ensure that no molecules were adsorbed in thesecages during the simulations, they were blocked.23

The models of Beerdsen et al.73,74 and Calero et al.75,76 wereused to assign partial charges to the zeolite atoms. TheCoulombic interactions were calculated using the Ewald sum-mation with a relative precision of 10-6.77 The non-Coulombicinteractions between nonbonded atoms or atom groups weremodeled using 12-6 Lennard-Jones potentials, truncated andshifted to 0 at the cutoff of 12 Å.

The van der Waals interactions between the adsorbates andthe framework are dominated by the dispersive forces betweenthe atom groups of the sorbed molecules and the oxygen atomsof the zeolite.78,79 For this reason, they are modeled through aneffective potential that only takes the oxygen atoms of the zeolitestructure into account. One set of potential parameters that hasbeen used in previous studies on water adsorption in zeoliteshas been implemented,80 together with Lorentz-Berthelotmixing rules in order to account for the rest of the interactions.All parameters for nonbonded interactions used in this workare listed in Table 1.

Experimental Results

Single-Component Adsorption. The experimental adsorptionand desorption isotherms and calculated adsorption isothermsfor water, methanol, and ethanol are shown in Figures 3-5,respectively. No significant hysteresis effects were observed inthe desorption measurements.

Adsorption data for water was previously published by denExter et al.20 Our measurements at T ) 333 and 323 Kcorrespond well with the previous data, whereas at 303 K, wefind a somewhat lower adsorption loading. The form of the wateradsorption isotherms (type II) is consistent with simple BETmodels valid for free surfaces or fine capillaries with noconstraint imposed on the adsorption from the opposite wall.However, it is expected to resemble type IV adsorption whenreaching saturation.81 The calculated water loading, however,

underpredicts the experimental values (Figure 3). The calculatedwater isotherms resemble type III adsorption behavior at lowfugacities, which is typical for water adsorption on hydrophobicsurfaces.82

The alcohol isotherms show a type I adsorption behavior,which is consistent with Langmuir adsorption (Figures 4 and

TABLE 1: Force Field Parameters Used in the Simulationsa

atom ε/kB/K σ/Å q/e ref.

Si (zeolite) 2.05 75O (zeolite) 93.53 3.0 -1.025 80O (water) 89.516 3.097 67H (water) 0.241 67M (water) -0.241 67CH3 (methanol) 98.0 3.75 0.265 68CH3 (ethanol) 98.0 3.75 68CH2 (ethanol) 46.0 3.95 0.265 68O (alcohol) 93.0 3.02 -0.7 68H (alcohol) 0.435 68

a The water model has two off-center charges that are labeled Min the table. The name “alcohol” refers to both methanol andethanol molecules. Figure 3. Water adsorption isotherms at 273 (this study, experimental

(b) and simulation (O)), 303 (this study, experimental (9) andsimulation (0), and den Exter et al.20 (×)), and 333 K (this study,experimental (2) and simulation (∆), and den Exter et al.20 (+)). Thesolid lines and the dashed lines are a guide to the eye for theexperimental and calculated isotherms, respectively.

Figure 4. Methanol adsorption isotherms at 273 (experimental (b)and simulation (O)), 303 (experimental (9) and simulation (0)), and333 K (experimental (2) and simulation (∆)). The bold lines indicatethe Langmuir fit through the experimental data, and the thin solid linesand the dashed lines are a guide to the eye for the experimental andcalculated isotherms, respectively.

Figure 5. Ethanol adsorption isotherms at 273 (experimental (b) andsimulation (O)), 303 (experimental (9) and simulation (0)), and 333K (experimental (2) and simulation (∆)). The bold lines indicate theLangmuir fit through the experimental data, and the thin solid linesand the dashed lines are a guide to the eye for the experimental andcalculated isotherms, respectively.


5). The fitted Langmuir parameters are given in Table 2. Thecalculated methanol and ethanol isotherms show a similar shapeand order of magnitude as the experimental data. However, atlow methanol fugacities, the loading is underpredicted, whileat higher fugacities, the calculated values are larger than themeasured data, suggesting that the force field parameters arenot well optimized for these types of systems.

The saturation loading is estimated by calculating the loadingat a fugacity of 1 GPa and 303 K. This yields a saturationloading of 13.9, 4.2, and 3.8 mol ·kg-1 for water, methanol, andethanol, respectively (Figure 6), corresponding to 33, 10, and 9molecules per 19-hedra cage, respectively. The calculated wateradsorption isotherm at 303 K resembles type V behavior. Itshould be noted that the water loading still showed a slightincrease when increasing the fugacity from 100 MPa to 1 GPa,but the increase was within the accuracy of the simulations.

Mixture Adsorption. Mixture adsorption isotherms havebeen computed at 303 and 360 K at a constant ratio of the waterfugacity, fw, divided by the sum of the water and the alcoholfugacity, fa, of Υw ) 0.1, 0.2, and 0.5. At low fugacities, thisequals the vapor molar fraction, yw. Figures 7 and 8 show themixture adsorption isotherms, together with the single-component isotherms at 303 K for methanol/water and ethanol/water mixtures, respectively. The loading of each componentis given as a function of the component fugacity. The isothermsshow that, at low fugacities, the loading is increased comparedto pure-component adsorption for all components. This effectis more pronounced for water adsorption than that for thealcohols. The shape of the water isotherm changes from typeV to type I, resembling the shape of the alcohol isotherms. Formethanol/water mixtures, the water loading decreases with thewater fugacity ratio, while for the ethanol/water mixures, thewater loading is insensitive to the water feed fraction between0.1 and 0.5. The alcohol loading at low fugacity shows a smallincrease with the water fugacity ratio, although the shape ofthe isotherm remains the same as that for single-componentadsorption. Although the ideal adsorbed solution theory (IAST)has shown to be able to describe the adsorption behavior of

water and methanol on activated carbon,83 in this system, theIAST fails to predict the adsorption correctly.

Figure 9 shows the binary adsorption isotherms as a functionof the sum of the water and alcohol fugacities up to 180 kPa at360 K at a constant water fugacity ratio of 0.2, whichcorresponds to a feed molar fraction of xw ) 0.5 and 0.2 formethanol/water and ethanol/water mixtures at 360 K, respec-tively. These conditions correspond to the feed conditions ofthe conducted pervaporation experiments, which are indicatedby the dashed lines. The loading of each component at the feedside under the pervaporation conditions can be directly readfrom the vertical axis. The calculated water loading remainslower than the alcohol loading within the range of calculatedtotal fugacities. The ethanol loading at the experimental feedcondition is lower than that of methanol, while the water loadingis equal in both systems.

Figure 10a and b show snapshots of the mixture adsorptionof methanol/water and ethanol/water, respectively. The figuresshow a higher loading of the alcohols compared to water. Thewater molecules show a preferential adsorption in the vicinityof the alcohol groups of the adsorbed methanol and ethanolmolecules with which hydrogen bonds are formed.

Diffusivities and Permeation. To elucidate the mechanismof separation of the selective water transport through thehydrophobic DD3R membrane, the simulation data are com-bined with data from permeation measurements conducted underpervaporation conditions. To assess the diffusivity of thecomponents, the feed conditions of the pervaporation experi-ments of pure-component and mixture permeation are taken asinput to calculate the loading at the feed side. The self-diffusivities and transport diffusivities, calculated by MDsimulations and the thermodynamic correction factor, are

TABLE 2: Fitted Langmuir Parameters [qi ) qisat(Kifi/(1 +

Kifi))] from the Measured Isotherms of Methanol andEthanol on DD3R at 273, 303, and 333 K

methanol ethanol

T/K qisat/mol ·kg-1 Ki

-3/×10 Pa · s-1 qisat/mol ·kg-1 Ki /×10-3 Pa · s-1

273 1.39 ( 0.03 18.18 ( 1.77 1.38 ( 0.03 16.81 ( 1.82303 1.53 ( 0.03 3.30 ( 0.29 1.51 ( 0.02 3.52 ( 0.29333 1.60 ( 0.06 0.53 ( 0.05 1.55 ( 0.05 0.70 ( 0.07

Figure 6. Water (b), methanol (9), and ethanol (2) adsorptionisotherms up to 1 GPa at 303 K.

Figure 7. Experimental (exp) pure-component and calculated (sim)methanol/water adsorption isotherms at 303 K for water fugacity ratiosof Υw ) 0, 0.1, 0.2, 0.5, and 1. The loading of (a) water and (b)methanol is shown as a function of the component fugacity. The solidlines indicate the experimental pure-component isotherms.


computed. In the calculations, a zeolite density of Fz ) 1700kg ·m-3 was used, which follows from the unit cell dimensions,15

and a membrane thickness of 2 µm was assumed.At the conditions of the pure-component permeation experi-

ments, the calculations predict a water loading at the feed sideof the membrane which is a factor of 2 to 3 lower than that forthe alcohols (Table 3). To asses the temperature dependence ofthe diffusivity, the self-diffusivities of water and methanol atconstant loading were computed. The loading was kept constantat the value corresponding to the equilibrium loading at 360 K.The values are listed as a function of the temperature in Table4 and plotted in Figure 11. The calculated feed side loadingand the self-diffusivities for the methanol/water and ethanol/water mixtures are listed in Tables 5 and 6, respectively. The

measured permeate side water molar fraction, ywperm, and the feed

composition, xwfeed, are also given, which give a measure for the

selectivity. Although the presence of alcohol enhances wateradsorption, at the feed conditions of the mixture permeationexperiments, the simulation data suggest that the molar waterloading is still only half of the alcohol loading. For water, theM-S diffusivity has been calculated and is on the same orderof magnitude as the self-diffusivity. For the alcohols, the M-Sdiffusivity could not be obtained due to poor statistics. The lowdiffusivity of ethanol was outside of the range of values whichcan be calculated using MD simulations. The ethanol moleculesdid not escape from the 19-hedra cage within the duration ofthe simulation (>20 ns). Therefore, we have estimated thisdiffusivity using TST. We have estimated the diffusivity ofethanol in DDR-type zeolite at 360 K using the methoddescribed in ref 84. We obtained a free-energy barrier of16.6kBT, corresponding to a diffusivity of 3 × 10-14 M2 · s-1.As TST provides an upper limit for the diffusivity,60,85 the actualdiffusivity of ethanol is even lower. This clearly shows that thediffusivity of ethanol is much lower than that of methanol andwater. In the remainder of this paper, we will use a diffusivityof 3 × 10-14 M2 · s-1 for ethanol. The ethanol diffusivity issimilar to the diffusivity of ethylene in high-silica DDR.86 Inthe remainder of the paper, only the self-diffusivities arereported, which are assumed to be similar to the M-Sdiffusivities. As a first model approximation, the self-diffusivitiesare only calculated for the feed side conditions of the pervapo-ration experiments and are assumed to be constant.

For mixture permeation at 360 K, the self-diffusivities aregiven as a function of the feed side composition, xw (Figure12). The calculated diffusivities of water are higher than thosefor the alcohols and increase with temperature and concentration.The increase with temperature and concentration also holds forethanol, while the calculated methanol diffusivity is practicallyconstant.

For the mixture permeation at xw ) 0.2 and 360 K, the valuesof ⟨Γij⟩ were determined and combined with the loading topredict the flux through the membrane using eq 7. The valueof the ethanol diffusivity in the mixture is estimated at 10-14

m2 · s-1. From the permeate composition, the water separationfactors are predicted at 8.7 and 2029 for methanol/water andethanol/water, respectively, which correspond well to theexperimentally determined values (6.8 and 1025).

The activation energies of the single-component transportdiffusivities are calculated from the data in Table 4 using eq 9and are listed in Table 8. The same table also shows the apparentenergy of activation calculated from the permeance, as wasreported earlier,5 and the average enthalpies of adsorptioncalculated as the isosteric heats of adsorption at zero loading at348 to 373 K. The enthalpy of adsorption increases withtemperature, but the values at 348 to 373 K only differ by lessthan 6%.

Discussion

The measured water adsorption isotherms are in fair agree-ment with the data reported by den Exter et al.53 The calculatedloading underpredicts the actual loading at low fugacities. Themeasured adsorption resembles type II/IV behavior, while thecalculated isotherm corresponds to type III/V (Figures 3 and6). This can be partly attributed to the presence of crystallinedefects in the zeolite sample, such as silanol groups, which havea strong effect on the water adsorption, especially at the lowend of the loadings (and fugacities).32 This is also the regionwhere deviations of the simulations are most apparent. Further-

Figure 8. Experimental (exp) pure-component and calculated (sim)ethanol/water adsorption isotherms at 303 K for water fugacity ratiosof Υw ) 0, 0.1, 0.2, 0.5, and 1. The loading of (a) water and (b) ethanolis shown as a function of the component fugacity. The solid linesindicate the experimental pure-component isotherms.

Figure 9. Simulated mixture adsorption isotherms of water (b) andmethanol (9) in methanol/water mixtures and water (O) and ethanol(0) in ethanol/water mixtures at 360 K and a water feed fugacity ratioof Υw ) 0.2 as a function of the total fugacity.


more, a deviation from the experimental values is expected sincethe molecular models and force field parameters were takendirectly from the literature, without further adjustment. Optimiz-ing these force field parameters was beyond the scope of thiswork.

Both the calculated and the measured alcohol isothermsresemble type I adsorption. The experimentally determinedalcohol loading is fairly well described by a single-site Langmuiradsorption isotherm (Figures 4 and 5). The Langmuir adsorptionparameters of methanol and ethanol are very similar (Table 2).However, the decrease in saturation loading with temperatureshows that this value does not resemble complete saturation.

The water adsorption is enhanced by the presence of alcoholand vice versa. This adsorption behavior is comparable to thatobserved for methanol/water coadsorption on activated carbon.83,87

The increased adsorption in the mixture as compared to single-component adsorption is significantly more pronounced for

water than that for the alcohols. Such an increase in waterloading was also observed by Jia et al., who conductedmolecular simulations for ethanol/water in silicalite-1.40 How-ever, in the study on silicalite-1, it was observed that the ethanolloading decreased, while in this study, the alcohol loading atconstant total fugacity was not significantly affected by thepresence of water. This effect can be attributed to wateradsorption in the vicinity of the polar OH groups of the adsorbedalcohols (Figure 10). The shape of the water isotherm changesin the presence of the alcohols from a type III/V to a type I,similar to the alcohol adsorption (Figures 7a and 8a). Becauseof this complex behavior, the estimation of mixture adsorptionbehavior from pure-component data is not trivial, and estimatingmixture adsorption from pure-component data using IAST didnot give a good description of the mixture adsorption.

The calculated pure-component molar loadings at the feedside of the here considered pervaporation conditions (Table 3)reveal that water adsorption increases with feed temperature,whereas the alcohol adsorption decreases. This shows that theincrease in vapor pressure for water, and therefore the drivingforce for adsorption, overcompensates the decrease in thehost-guest interaction. This is in agreement with the lowerenthalpy of adsorption for water as compared to that for thealcohols (Table 8). At high fugacities (∼100 kPa), the calculatedpure-component water adsorption surpasses the alcohol adsorp-tion (Figure 6).

The calculated saturation loadings are comparable to valuesreported for LTA-type zeolites, which are 15, 6, and 4 mol ·kg-1

(204, 82, and 55 molecules per unit cell) for water, methanol,

Figure 10. Snapshots of (a) methanol/water and (b) ethanol/water mixture adsorption on DD3R at 360 K and 15 kPa.

TABLE 3: Self-Diffusivities Calculated from MDSimulations and TST (ethanol) and Pure-ComponentLoadings Calculated from GCMC Simulations at the Feedand Permeate Sides for Pure-Component Permeation ofWater, Methanol, and Ethanola

T/K qifeed/mol ·kg-1 qi

perm/mol ·kg-1 Dis/m2 · s-1

Water348 0.87 0.004 3.61 ( 0.01 × 10-10

360 0.97 0.003 6.10 ( 0.04 × 10-10

373 1.11 0.002 5.89 ( 0.04 × 10-10

Methanol348 2.46 0.09 3.36 ( 0.01 × 10-11

360 2.39 0.05 3.55 ( 0.01 × 10-11

373 2.36 0.03 4.80 ( 0.02 × 10-11

Ethanol348 1.49 0.23 <3 × 10-14

360 1.42 0.15 <3 × 10-14

373 1.39 0.10 <3 × 10-14

a The permeate side pressure was maintained at 1 kPa.

TABLE 4: Single-Component Self-Diffusivities for Waterand Methanol, Calculated from MD Simulations at aConstant Loading of 0.97 and 2.39 mol ·kg-1 (14 and 35molecules per unit cell) for Water and Methanol,Respectively

T/K Dws /m2 · s-1 Dm

s /m2 · s-1

303 1.43 ( 0.01 × 10-10 1.91 ( 0.03 × 10-11

348 3.58 ( 0.01 × 10-10 3.00 ( 0.03 × 10-11

360 6.10 ( 0.04 × 10-10 3.55 ( 0.01 × 10-11

373 6.76 ( 0.05 × 10-10 4.80 ( 0.02 × 10-11

Figure 11. The logarithm of the self-diffusivity, Dis, versus the

reciprocal temperature for pure-component permeation of water (b)and methanol (9). The data are also listed in Table 4. The lines indicatea linear trend, according to eq 9.


and ethanol, respectively.88,89 Adsorption of 1 mol ·kg-1 inDD3R corresponds to 14.4 adsorbed molecules per unit cell and2.4 molecules per 19-hedra cage (6 cages per unit cell). In theLTA zeolite structure, which contains only one R-cage and one�-cage per unit cell, 1 mol ·kg-1 equals 13.6 molecules per unitcell. Since in DDR only the 19-hedra cages are accessible andin LTA the complete internal void volume is accessible forwater, the saturation loading on DD3R is expected to be lower.Corma et al. reported a water loading of 0.6 mol ·kg-1 at 298K and 2 kPa on all-silica LTA,90 which is higher than themeasured 0.35 mol ·kg-1 on DD3R at 303 K and similarpressure. Coudert et al. reported a water saturation loading inthe all-silica LTA calculated by GCMC simulations of 1.5mol ·kg-1,25 suggesting the pressure at which this loading wascalculated was not sufficient to obtain full saturation.

The calculated alcohol saturations using GCMC simulationsare higher than the values found by fitting the Langmuirisotherms. This deviation and the increase of the fitted saturationloading with temperature could indicate the Langmuir adsorptionisotherm fitted at low pressures does not describe the fulladsorption behavior. The fitted saturation loadings (Table 2)are similar to reported values obtained by fitting Langmuirisotherms for methane (1.7 mol ·kg-1) and ethane (0.8 mol ·kg-1)and lower than those for carbon dioxide (2.8 mol ·kg-1) inDD3R.19,91 Both the calculated and experimental values of theethanol loading are approximately three times higher thanreported loadings for ethane.19 The loadings of methanol and

ethanol are also higher than reported loadings for methane,ethane, and ethene on ZSM-58.86 The calculated water saturationloading on DD3R is substantially higher than the 6.9 mol ·kg-1

(41 molecules per unit cell) reported for water in silicalite-1,which corresponds to a density inside of the zeolite of 85% ofthe bulk fluid value.27,29 To obtain the actual saturation loadingof water on DD3R, adsorption measurements over a widepressure range need to be conducted, as was already reportedfor water in silicalite-1.32

The enthalpies of adsorption show a small increase withtemperature, but since the values in the temperature rangeconcerned differ by less than 6%, an average value is used(Table 8). The calculated values for the alcohols (-40 and -47kJ ·mol-1 for methanol and ethanol, respectively) are comparableto values reported for adsorption on silicalite-1.92 For water,the computed value of -26 kJ ·mol-1 was comparable to the-20 kJ ·mol-1 reported by Fleys et al. for silicalite-1. Den Exterobtained an enthalpy of adsorption of -50 kJ ·mol-1 at zeroloading by fitting the experimentally measured water isothermson DDR (which is in agreement with the data obtained in thisstudy) to a BET isotherm.81

The water and methanol self-diffusivities at constant loadingincrease with temperature, yielding a positive energy of activa-tion. The sum of the energy of activation and the enthalpies ofadsorption, obtained by the GCMC simulations, correspond wellwith the apparent energy of activation calculated from the pure-component permeances (Table 8). This indicates that thetemperature dependency of neither the adsorption nor thediffusion is dominating.

At equal component fugacities, the water loading (computedusing the Tip5pEw force field) is similar to that of the alcohols.However, despite the improved water adsorption in the presenceof the alcohols, the calculations indicate that the water loadingat the feed side during pervaporation experiments is at most50% of the alcohol loading (Tables 5 and 6). This shows thatpreferential adsorption is not the cause of the observed separa-tion selectivities of the membrane. The high water selectivitiesobtained in the pervaporation experiments should be attributedto the difference in diffusivity.

For methanol and ethanol, the adsorption behavior is con-sistent with Langmuir adsorption. Moreover, the size of theDD3R cage window is similar to the molecular diameters ofthe alcohols.5 The calculated pure-component transport diffu-sivity of water is 1 order of magnitude higher than the methanoldiffusivity and even 3 orders of magnitude higher than theethanol diffusivity. This identifies the difference in diffusivity

TABLE 5: Methanol/Water Feed and Permeate Side Composition and Loading and Feed Side Self-Diffusivitiesa

T/K xwfeed (ref 5) qw

feed/mol ·kg-1 qmfeed/mol ·kg-1 yw

perm (ref 5) qwperm/mol ·kg-1 qm

perm/mol ·kg-1 Dws /m2 · s-1 Dm

s /m2 · s-1

348 0.20 0.69 2.15 0.59 0.011 0.067 1.49 ( 0.02 × 10-10 1.48 ( 0.01 × 10-11

360 0.10 0.42 2.18 0.45 0.062 0.006 1.43 ( 0.01 × 10-10 2.02 ( 0.01 × 10-11

360 0.20 0.59 2.12 0.63 0.031 0.004 2.04 ( 0.01 × 10-10 1.91 ( 0.01 × 10-11

360 0.50 1.00 1.94 0.84 0.015 0.005 2.55 ( 0.02 × 10-10 1.48 ( 0.01 × 10-11

373 0.20 0.59 2.08 0.64 0.002 0.018 1.05 ( 0.03 × 10-10 3.14 ( 0.03 × 10-11

a The permeate side pressure in the pervaporation experiments is 1.5 kPa.

TABLE 6: Ethanol/Water Feed and Permeate Side Composition and Loading and Feed Side Self-Diffusivitiesa

T/K xwfeed (ref 5) qw

feed/mol ·kg-1 qmfeed/mol ·kg-1 yw

perm (ref 5) qwperm/mol ·kg-1 qm

perm/mol ·kg-1 Dws /m2 · s-1 De

s/m2 · s-1

348 0.10 0.80 1.63 0.993 0.0056 0.013 4.45 ( 0.06 × 10-11 <3 × 10-14

360 0.20 0.83 1.59 0.996 0.0017 0.007 5.75 ( 0.04 × 10-11 <3 × 10-14

360 0.30 0.94 1.56 0.998 0.0004 0.004 7.79 ( 0.20 × 10-11 <3 × 10-14

373 0.20 0.83 1.53 0.997 0.0043 0.001 1.44 ( 0.20 × 10-11 <3 × 10-14

a The permeate side pressure in the pervaporation experiments is 1.5 kPa.

Figure 12. The logarithm of the self-diffusivity, Dis versus the water

feed molar fraction, xw, for pure-component permeation of water (b)and methanol (9), and water (O) and methanol (0) in methanol/watermixtures, and water (gray circle) in ethanol/water mixtures at 360 K.The data are also listed in Tables 3, 5, and 6. The dashed lines indicatea guide to the eye.


as the major separating mechanism for the dewatering ofalcohols using DD3R membranes. The calculated diffusivity ofwater is lower than the reported values for water in silicalite-1.26 The diffusivity found for methanol is about 1 order ofmagnitude higher than the reported value for methane.86 Forethanol, TST was used to obtained a more accurate value ofthe ethanol diffusivity as the MD approach is inadequate dueto the long required simulation time.

To investigate the influence of the thermostat, we computedthe diffusivity of water and methanol at 360 K using MD, bothin the NVT ensemble (using the Nose-Hoover thermostat) andin the NVE ensemble. We found that the diffusivity com-puted in the NVE ensemble was 25% higher than that in theNVT ensemble. This clearly shows that in our case, thethermostat only plays a minor role in the dynamics.

To investigate the possible effect of framework flexibility,we have performed additional MD simulation using the flexiblezeolite model of Demontis and Suffritti.93 In this model,interactions between framework zeolite atoms are described byharmonic springs, using equilibrium bond lengths taken fromthe crystal structure of DDR-type zeolite. Electrostatic interac-tions between framework atoms are not taken into account.Electrostatic interactions between guest molecules and guestmolecules and the framework are handled using the Wolfmethod.94 At 373 K, we found that the diffusivity of water andmethanol in a flexible framework is ∼50% lower than that fora rigid zeolite. We also verified that the use of the Wolf methodinstead of the Ewald summation does not change the diffusivity.These preliminary simulations clearly show that frameworkflexibility indeed influences the diffusion of water in DDR-typezeolite but that the main conclusion of our work (water isdiffusing much faster than methanol and methanol is diffusingmuch faster than ethanol) is unaffected.

In methanol/water mixture permeation, the methanol diffu-sivities are very similar to the values found for pure-componenttransport, while for water, the diffusivities are a factor of 2 to3 lower than those for pure-component diffusion (Tables 3, 5,and 6). In the ethanol/water system, the diffusivity of water isreduced by 1 order of magnitude as compared to pure-component permeation. This trend is in accordance with themeasured component fluxes under pervaporation conditions.5

As the adsorption of the components is increased in the mixture,

competitive adsorption does not explain this reduced mixturepermeance. The stronger interactions, leading to higher adsorp-tion in the mixtures, combined with steric hindrance of theadsorbed molecules blocking the cage windows are likely toreduce the diffusivity, leading to a lower permeance in mixtures.The seemingly unaffected permeance of methanol may beexplained by a positive effect of the increased adsorption,balanced by a negative effect of a reduced diffusivity.

The calculated thermodynamic correction factors show thatthe presence of the alcohol at permeate side conditions has anegative impact on the water flux and vice versa. This showsthe same trend as the calculated self-diffusivities (Figure 12)and agrees with the experimental observation of a decrease influx of both alcohol and water in mixtures as compared to pure-component permeation.5

The predicted values for the flux (Table 7) are 1 order ofmagnitude higher than the experimentally measured values.However, the calculated permeate composition deviates less than10% from the experimental value. The discrepancy in the fluxescan be partly attributed to the assumption of a constant valuefor Γij. Since the diffusivity of the alcohols is much lower thanthat for water, it is expected that the average value of Γij willbe closer to the permeate side conditions.

It should be noted that the diffusivities obtained in the MDsimulations assume a perfect crystalline membrane withoutmicro- and macroscopic defects. This will not resemble theactual membrane, which consists of complexly intergrowncrystals with internal and external grain boundaries. Moreover,the DD3R membrane synthesized by NGK insulators consistedof a thin zeolite layer deposited on an R-alumina support. Duringthe membrane synthesis, aluminum can be leached out of thesupport material and be incorporated into the zeolite structure,making the membrane material less hydrophobic than theseparately synthesized crystals and the simulated zeolite frame-work. Furthermore, despite the high calcination temperature ofthe DD3R zeolite,11 silanol groups can still be present in thezeolite as crystalline defects. In silicalite-1, it has been observedthat up to 8% of the Si is present as Si-OH.95 These groupslinearly increase the framework polarity and therewith thehydrophilicity.95,96

Considering that no force field parameters were fitted here,the calculated adsorption isotherms show a reasonable fit withthe experimental adsorption data. To obtain a quantitative resultthat better matches the experimental isotherms, a refinementof the force field would be required, while taking proper accountof the material imperfections, such as silanol groups. Thesimulation results could also be improved by accounting forframework flexibility. Including framework flexibility has shownto increase the ethanol/water loading in silicalite-1 withoutaffecting the sorption selectivity.40

TABLE 7: Measured (Niexp) and Predicted (Ni

pred) Component Fluxes Using Equations 7 and 8 and the ThermodynamicCorrection Factors at the Feed and Permeate Side Conditions in Methanol/Water and Ethanol/Water Pervaporation at xw ) 0.2and T ) 360 Ka

component Γiifeed Γij

feed Γiiperm Γij

perm Niexp/mol ·m-2 · s-1 Ni

pred/mol ·m-2 · s-1 yiperm,exp yi

perm,pred

Methanol/Waterwater 2.31 0.42 0.91 -0.01 0.023 0.40 0.63 0.69methanol 9.20 0.99 0.23 -0.05 0.013 0.19 0.37 0.31

Methanol/Waterwater 1.74 1.65 1.00 -0.30 0.015 0.25 0.996 0.998ethanol 6.15 0.48 1.00 -0.04 0.59 × 10-4 4.9 × 10-4 0.004 0.002

a The permeate side pressure in the experiments was maintained at 1.5 kPa.

TABLE 8: Average Enthalpies of Adsorption at ZeroLoading Computed by GCMC at 348 to 373 K, the Energyof Activation for the Transport Diffusivity, and theApparent Activation Energies of Pervaporation as Reportedin Reference 5 for Water, Methanol, and Ethanol on DD3R

component ∆adsH/kJ ·mol-1 EA/kJ ·mol-1 Eapp5/kJ ·mol-1

water -26.3 21.3 -9.1methanol -39.6 11.9 -32.4ethanol -46.6 -23.6


Conclusions

Adsorption isotherms for water, methanol, and ethanol onall-silica DD3R have been measured by single-componentvapor-phase adsorption and calculated by GCMC simulations.The measured alcohol adsorption can be described by a single-site Langmuir adsorption isotherm. The MC simulations wereable to qualitatively reproduce the adsorption behavior of theexperimental isotherms. For water, the adsorption resembles typeII adsorption. The water loading is underpredicted by thecalculations at pressures up to 2.5 kPa. The calculated loadingsat high pressures of all of the components are higher thanreported values for gases in DDR-type zeolites and arecomparable to loadings on LTA-type zeolites. The molecularmodels and force field parameters were taken directly fromliterature without further adjustment. Nevertheless, the orderof magnitude and the shape of the pure-component isothermsgive a good resemblance of the experimental data.

Mixture adsorption isotherms and diffusivities have beencalculated and are compared with permeation data measuredunder pervaporation conditions. The calculated mixture iso-therms show that the loading of both alcohols and water atconstant component fugacity increases as compared to pure-component adsorption. Moreover, the shape of the waterisotherm changes from type IV to type I. The decrease in waterand ethanol permeance in the mixture as compared to pure-component permeation is not caused by competitive adsorption.The increase in loading in mixture adsorption is significantlymore profound for water than that for the alcohols, but this doesnot lead to adsorption selectivity for water at the feed conditionsof the pervaporation experiments.

The self-diffusivities calculated by MD simulations and thecalculated mixture isotherms show that the water diffusivity isat least 1-3 orders of magnitude higher than those of thealcohols. Although component fluxes predicted from the simula-tion data overpredict the experimentally observed values by 1order of magnitude, the permeate composition corresponds wellwith experimental data. The selective water transport throughDD3R zeolite membranes can, therefore, be explained by thehigher diffusivity of water in the hydrophobic DD3R.

Acknowledgment. We acknowledge the Delft ResearchCenter for Sustainable Industrial Processes (DCSIP) and theEuropean Science Foundation (ESF) for the financial support.T.J.H.V. acknowledges financial support from The NetherlandsOrganization for Scientific Research (NWO-CW) through aVIDI grant. Sander Brouwer is gratefully acknowledged for theadsorption measurements.

Appendix

List of Symbols

symbol, name, unitc, concentration, mol ·m-3

Ds, self-diffusivity, m2 · s-1

DT, transport diffusivity, m2 · s-1

D, M-S diffusivity, m2 · s-1

EA, energy of activation for diffusion, J ·mol-1

Eapp, apparent energy of activation, J ·mol-1

f, fugacity, Pa∆adsH, adsorption enthalpy, J ·mol-1

K, Langmuir adsorption parameter, Pa-1

kB, Boltzmann constant, m2 ·kg · s-2 ·K-1

n, molar flux, mol ·m-2 · s-1

M, molar mass, kg ·mol-1

p, pressure, Papsat, vapor pressure, Paq, equilibrium loading, mol ·kg-1

qsat, saturation loading, mol ·kg-1

R, gas constant, J ·mol-1 ·K-1

T, temperature, KVm, partial molar volume, m3 ·mol-1

x, liquid molar fractiony, vapor molar fractionz, length coordinate, mδ, membrane thickness, mε, Lennard-Jones energy parameter, JΓ, thermodynamic correction factorγ, activity coefficientσ, Lennard-Jones size parameter, mθ, fractional loadingµ, chemical potential, J ·mol-1

F, density, kg ·m-3

Π, permeance, mol ·m-2 · s-1 ·Pa-1

φsat, fugacity coefficientΥ, fugacity ratio

sub and superscripts

l, liquidexp, experimentalfeed, feed sideperm, permeate sidepred, predicteda, alcohole, ethanoli, component iM, membranem, methanolw, waterz, zeolite

References and Notes

(1) van den Bergh, J.; Zhu, W. D.; Kapteijn, F.; Moulijn, J. A.; Yajima,K.; Nakayama, K.; Tomita, T.; Yoshida, S. Res. Chem. Intermed. 2008,34, 467–474.

(2) van den Bergh, J.; Zhu, W.; Gascon, J.; Moulijn, J. A.; Kapteijn,F. J. Membr. Sci. 2008, 316, 35–45.

(3) van den Bergh, J.; Zhu, W.; Groen, J.; Kapteijn, F.; Moulijn, J. A.;Yajima, K.; Yajima, K.; Nakayama, K.; Tomita, T.; Yoshida, S. Stud. Surf.Sci. Catal. 2007, 170A, 1021–1027.

(4) Yajima, K.; Nakayama, K.; Niino, M.; Tomita, T.; Yoshida, S.Proceedings of the 9th International Conference on Inorganic Membranes,Lillehammer, Norway; SINTEF: Oslo, Norway, 2006; pp 401-404.

(5) Kuhn, J.; Yajima, K.; Tomita, T.; Gross, J.; Kapteijn, F. J. Membr.Sci. 2008, 321, 344–349.

(6) Khajavi, S.; Kapteijn, F.; Jansen, J. C. J. Membr. Sci. 2007, 299,63–72.

(7) Khajavi, S.; Jansen, J. C.; Kapteijn, F. J. Membr. Sci. 2009, 326,153–160.

(8) Morigami, Y.; Kondo, M.; Abe, J.; Kita, H.; Okamoto, K. Sep.Purif. Technol. 2001, 25, 251–260.

(9) Caro, J.; Noack, M. Microporous Mesoporous Mater. 2008, 115,215–233.

(10) Gies, H. J. Inclusion Phenom. Macrocyclic Chem. 1984, 2, 275–278.

(11) Kuhn, J.; Gascon, J.; Gross, J.; Kapteijn, F. Microporous Meso-porous Mater. 2009, 120, 12–18.

(12) Stewart, A. Zeolite Sigma-1; 1987; 8717794, GB 2193202 A.(13) Ernst, S.; Weitkamp, J. Chem.-Ing.-Tech. 1991, 63, 748–750.(14) Zhu, W.; Kapteijn, F.; Moulijn, J. A.; Jansen, J. C. Phys. Chem.

Chem. Phys. 2000, 2, 1773–1779.(15) Baerlocher, C.; McCusker, L. B.; Olson, D. H., Atlas of Zeolite

Framework Types, 6th revised edition ed.;Elsevier; Amsterdam, TheNetherlands, 2007.

(16) Nakayama, K.; Suzuki, K.; Yoshida, M.; Yajima, K.; Tomita, T.Method for preparing DDR type zeolite film, DDR type zeolite film, and


composite DDR type zeolite film, and method for preparation thereof. U.S.Patent 10796890, 09.03.2004, European Patent EP 2002765562, 25.03.2004,2003.

(17) Tomita, T.; Nakayama, K.; Sakai, H. Microporous MesoporousMater. 2004, 68, 71–75.

(18) Krishna, R.; van Baten, J. M. Chem. Phys. Lett. 2007, 446, 344–349.

(19) Zhu, W.; Kapteijn, F.; Moulijn, J. A.; den Exter, M.C.; Jansen,J. C. Langmuir 2000, 16, 3322–3329.

(20) den Exter, M.J.; Jansen, J. C.; van Bekkum, H.; Zikanova, A.Zeolites 1997, 19, 353–358.

(21) Krishna, R.; van Baten, J. M.; Garcıa-Perez, E.; Calero, S. Chem.Phys. Lett. 2006, 429, 219–224.

(22) Himeno, S.; Takenaka, M.; Shimura, S. Mol. Simul. 2008, 34, 1329–1336.

(23) Jakobtorweihen, S.; Hansen, N.; Keil, F. J. Mol. Phys. 2005, 103,471–489.

(24) Smit, B.; Maesen, T. L. M. Chem. ReV. 2008, 108, 4125–4184.(25) Coudert, F. X.; Vuilleumier, R.; Boutin, A. ChemPhysChem 2006,

7, 2464–2467.(26) Demontis, P.; Stara, G.; Suffritti, G. B. J. Phys. Chem. B 2003,

107, 4426–4436.(27) Desbiens, N.; Demachy, I.; Fuchs, A. H.; Kirsch-Rodeschini, H.;

Soulard, M.; Patarin, J. Angew. Chem., Int. Ed. 2005, 44, 5310–5313.(28) Beauvais, C.; Boutin, A.; Fuchs, A. H. Adsorption 2005, 11, 279–

282.(29) Puibasset, J.; Pellenq, R. J. M. J. Phys. Chem. B 2008, 112, 6390–

6397.(30) Fleys, M.; Thompson, R. W.; MacDonald, J. C. J. Phys. Chem. B

2004, 108, 12197–12203.(31) Fleys, M.; Thompson, R. W. J. Chem. Theory Comput. 2005, 1,

453–458.(32) Trzipit, M.; Soulard, M.; Patarin, J.; Desbiens, N.; Cailliez, F.;

Boutin, A.; Demachy, I.; Fuchs, A. H. Langmuir 2007, 23, 10131–10139.(33) Vigne-Maeder, F.; Auroux, A. J. Phys. Chem. 1990, 94, 316–322.(34) Pelmenschikov, A. G.; Morosi, G.; Gamba, A. J. Phys. Chem. 1993,

97, 11979–11986.(35) Hinderer, J.; Keil, F. J. Chem. Eng. Sci. 1996, 51, 2667–2672.(36) Plant, D. F.; Maurin, G.; Bell, R. G. J. Phys. Chem. B 2007, 111,

2836–2844.(37) Gupta, A.; Clark, L. A.; Snurr, R. Q. Langmuir 2000, 16, 3910–

3919.(38) Lu, L.; Shao, Q.; Huang, L.; Lu, Xiaohua Fluid Phase Equilib.

2007, 261, 191–198.(39) Takaba, H.; Koyama, A.; Nakao, S. J. Phys. Chem. B 2000, 104,

6353–6359.(40) Jia, W.; Murad, S. Mol. Phys. 2006, 104, 3033–3043.(41) Furukawa, S.; Goda, K.; Zhang, Y.; Nitta, T. J. Chem. Eng. Jpn.

2004, 37, 67–74.(42) Yang, J. Z.; Chen, Y.; Zhu, A. M.; Liu, Q. L.; Wu, J. Y. J. Membr.

Sci. 2008, 318, 327–333.(43) Yang, J. Z.; Liu, Q. L.; Wang, H. T. J. Membr. Sci. 2007, 291,

1–9.(44) Kox, M. H. F.; Stavitski, E.; Groen, J. C.; Perez-Ramirez, J.;

Kapteijn, F.; Weckhuysen, B. M. Chem.sEur. J. 2008, 14, 1718–1725.(45) Castillo, J. M.; Dubbeldam, D.; Vlugt, T. J. H.; Calero, S. Mol.

Simul. 2009, doi: 10.1080/08927020902865923.(46) Gascon, J.; Blom, W.; van Miltenburg, A.; Ferreira, A.; Berger,

R.; Kapteijn, F. Microporous Mesoporous Mater. 2008, 115, 585–593.(47) Krishna, R.; Wesselingh, J. A. Chem. Eng. Sci. 1997, 52, 861–

911.(48) Krishna, R. Chem. Eng. J. 2001, 84, 207–214.(49) Krishna, R.; Paschek, D. Chem. Eng. J. 2002, 85, 7–15.(50) Krishna, R.; van Baten, J. M. Ind. Eng. Chem. Res. 2006, 45, 2084–

2093.(51) Karger, J.; Ruthven, D. M. Diffusion in Zeolites and other

microporous solids; John Wiley & Sons, Inc.: New York, 1992.(52) Kuhn, J.; Stemmer, R.; Kapteijn, F.; Kjelstrup, S.; Gross, J. J.

Membr. Sci. 2009, 330, 388–398.(53) Kapteijn, F.; Moulijn, J. A.; Krishna, R. Chem. Eng. Sci. 2000, 55,

2923–2930.(54) Krishna, R.; van Baten, J. M.; Garcıa-Perez, E.; Calero, S. Ind.

Eng. Chem. Res. 2007, 46, 2974–2986.(55) Smith, J. M.; van Ness, H. C.; Abbott, M. M. Introduction to

chemical engineering thermodynamics; McGraw-Hill: New York, 2001.(56) Gmehling, J.; Onken, U. Vapor-liquid equilibrium data collections

Aqueous-organic systems; Chemistry Data Series; DECHEMA: Frankfurt,Germany, 1977; Vol 1.

(57) de Bruijn, F.; Gross, J.; Olujic, Z.; Jansens, P.; Kapteijn, F. Ind.Eng. Chem. Res. 2007, 46, 4091–4099.

(58) Kapteijn, F.; Zhu, W.; Moulijn, J. A.; Gardner, T. Q. Zeolitemembranes: modeling and application. Structured catalysts and reactors;CRC Taylor & Francis: Boca Raton, 2006, Chap. 20, pp 700-746.

(59) Frenkel, D.; Smit, B. Understanding molecular simulation: fromalgorithms to applications; Academic Press: New York, 1996.

(60) Frenkel, D.; Smit, B. Understanding Molecular Simulation: fromAlgorithms to Applications, 2nd ed.; Academic Press: San Diego, CA, 2002.

(61) Vlugt, T. J. H.; Krishna, R.; Smit, B. J. Phys. Chem. B 1999, 103,1102–1118.

(62) Dubbeldam, D.; Calero, S.; Vlugt, T. J. H.; Krishna, R.; Maesen,T. L. M.; Smit, B. J. Phys. Chem. B 2004, 108, 12301–12313.

(63) Vlugt, T. J. H.; Garcıa-Perez, E.; Dubbeldam, D.; Calero, S.J. Chem. Theory Comput. 2008, 1107–1118.

(64) Bird, R. B.; Lightfoot, W. E.; Stewart, E. N. Transport Phenomena;John Wiley & Sons: New York, 2002.

(65) Kjelstrup, S.; Bedeaux, D. Non-equilibrium thermodynamics ofheterogeneous systems; Series on Advnaces in Statistical Mechanics; WorldScientific Publishing: Singapore, 2008, Vol 16.

(66) Dubbeldam, D.; Snurr, R. Q. Mol. Simul. 2007, 33, 305–325.(67) Rick, S. J. Chem. Phys. 2004, 120, 6085–6093.(68) Chen, B.; Potoff, J. J.; Siepman, J. I. J. Phys. Chem. B 2001, 105,

3093–3104.(69) Moise, J. C.; Bellat, J. P.; Methivier, A. Microporous Mesoporous

Mater. 2001, 43, 91–101.(70) Chen, H. B.; Sholl, D. S. Langmuir 2006, 22, 709–716.(71) Gies, H. Z. Kristallogr. 1986, 175, 93–104.(72) Vlugt, T. J. H.; Schenk, M. J. Phys. Chem. B 2002, 106, 12757–

12763.(73) Beerdsen, E.; Smit, B.; Calero, S. J. Phys. Chem. B 2002, 106,

10659–10667.(74) Beerdsen, E.; Dubbeldam, D.; Smit, B.; Vlugt, T. J. H.; Calero, S.

J. Phys. Chem. B 2003, 107, 12088–12096.(75) Calero, S.; Dubbeldam, D.; Krishna, R.; Smit, B.; Vlugt, T. J. H.;

Denayer, J. F. M.; Martens, J. A.; Maesen, T. L. M. J. Am. Chem. Soc.2004, 126, 11377–11386.

(76) Calero, S.; Lobato, M. D.; Garcıa-Perez, E.; Mejias, J. A.; Lago,S.; Vlugt, T. J. H.; Maesen, T. L. M.; Smit, B.; Dubbeldam, D. J. Phys.Chem. B 2006, 110, 5838–5841.

(77) Allen, M. P.; Tildesley, D. J. Computer Simulation of Liquids;Clarendon Press: Oxford, U.K., 1987.

(78) Kiselev, A. V.; Lopatkin, A. A.; Shulga, A. A. Zeolites 1985, 5,261–267.

(79) Bezus, A. G.; Kiselev, A. V.; Lopatkin, A. A.; Du, P. Q. J. Chem.Soc., Faraday Trans. 2 1978, 74, 367–379.

(80) Pascual, P.; Ungerer, P.; Tavitian, B.; Pernot, P.; Boutin, A. Phys.Chem. Chem. Phys. 2003, 5, 3684–3693.

(81) den Exter, M.J., Zeolites .Ph.D. Thesis, Delft University ofTechnology, 1996.

(82) Wang, H. J.; Xi, X. K.; Kleinhammes, A.; Wu, Y. Science 2008,322, 80–83.

(83) Finqueneisel, G.; Zimny, T.; Weber, J. V. Carbon 2005, 43, 1093–1095.

(84) Dubbeldam, D.; Beerdsen, E.; Vlugt, T. J. H.; Smit, B. J. Chem.Phys. 2005, 122, 224712.

(85) Wahnstrom, G.; Haug, K.; Metiu, H. Chem. Phys. Lett. 1988, 148,158–163.

(86) Hedin, N.; DeMartin, G. J.; Roth, W. J.; Strohmaier, K. G.; Reyes,S. C. Microporous Mesoporous Mater. 2008, 109, 327–334.

(87) Linders, M. J. G.; van den Broeke, L. J. P.; Kapteijn, F.; Moulijn,J. A.; van Bokhoven, J. J. G. M. AICHE J. 2001, 47, 1885–1892.

(88) Okamoto, K.; Kita, H.; Horii, K.; Tanaka, K.; Kondo, M. Ind. Eng.Chem. Res. 2001, 40, 163–175.

(89) Kristof, T.; Csanyi, E.; Rutkai, G.; Merenyi, L. Mol. Simul. 2006,32, 869–875.

(90) Corma, A.; Rey, F.; Rius, J.; Sabater, M. J.; Valencia, S. Nature2004, 431, 287–290.

(91) Himeno, S.; Tomita, T.; Suzuki, K.; Yoshida, S. MicroporousMesoporous Mater. 2007, 98, 62–69.

(92) Chen, H. L.; Li, Y. S.; Yang, W. S. J. Membr. Sci. 2007, 296,122–130.

(93) Demontis, P.; Suffritti, G. B.; Quartieri, S.; Fois, E. S.; Gamba, A.J. Phys. Chem. 1988, 92, 867–871.

(94) Fennell, C. J.; Gezelter, J. D. J. Chem. Phys. 2006, 124, 234104.(95) Hunger, M.; Karger, J.; Pfeifer, H.; Caro, J.; Zibrowius, B.; Bulow,

M.; Mostowicz, R. J. Chem. Soc., Faraday Trans. 1 1987, 83, 3459–3468.(96) Kuhn, J.; Gross, J.; Jansen, J. C.; Jansens, P. J. Studies in surface

science and catalysis: From zeolites to porous MOF materials; Elsevier:Amsterdam, The Netherlands, 2007, Vol. 170A, pp 942-948.

JP901869D


adsorption and diffusion of water, methanol, and …...the adsorption and diffusion of water and...

Documents