Fernando J. A. L.
Cruz
*a,
Erich A.
Müller
b and
José P. B.
Mota
a
aRequimte/CQFB, Department of Chemistry, Faculdade de Ciências e Tecnologia, Universidade Nova de Lisboa, 2829-516, Caparica, Portugal. E-mail: f.cruz@dq.fct.unl.pt
bDepartment of Chemical Engineering, Imperial College London, South Kensington Campus, London, SW7 2AZ, UK
First published on 2nd August 2011
The influence of the molecular force field upon the transport properties of ethylene is studied by molecular dynamics simulations, addressing both the bulk fluid and the confined phases within pristine single-walled carbon nanotubes. Five different molecular models with different degrees of coarse-graining were selected, spanning from a simple isotropic Lennard–Jones sphere to a fully detailed all-atom description. The fluid was probed for its self-diffusion coefficient under isochoric (0.026 ≤ ρ (mol L−1) ≤ 15.751) and isothermal (220 ≤ T (K) ≤ 340) conditions, both in the sub- and supercritical regions (Tbulkc = 282.4 K). Although the particular details of each potential model are seen to be nearly irrelevant to the bulk fluid dynamics, they are crucial to correctly describe the inhomogeneous system. The most important aspects affecting fluid transport are the existence/absence of explicit electrostatic contributions and the molecular shape. The effect of temperature on the confined fluid self-diffusion is described by the Arrhenius law, D = Aexp(−Ea/RT), and the nonlinear density dependencies of the activation energy (Ea/R) and pre-exponential factor (A) have been fitted here to empirical equations. In spite of the quasi one-dimensional confining nature of SWCNTs, the isothermal results (T = 300 K) obtained for the bulk and confined systems collapse onto a unique master curve, D = D0ρλ, suggesting that the self-diffusion coefficient of a confined fluid can be estimated from its molecular density, an easily accessible property.
As a structurally simple and well characterized model of cylindrical nanopores, single-walled carbon nanotubes (SWCNTs) are excellent candidates for theoretical studies. Their structure can be rationalized as the cylindrical folding of a graphene sheet along a particular directional vector (n,m).10,11 As the result of that folding, the carbon atoms on the solid walls may be arranged in three distinct topologies, namely armchair (n = m), zig-zag (m = 0) or chiral (n ≠ m) nanotubes. Each of those topologies exhibit different optical12 and electronic13 properties, and amongst other applications have been proposed as building blocks in composites,14 chemical sensors,15 field-effect transistors in nanoelectronics,16 biomedical devices17,18 and as separating media of organic vapors.19–21 Their hydrophobicity places them as ideal candidates to explore molecular transport through biological membranes, and therefore the dynamics of confined water has been one of the main subjects of previous work in this area.7,22,23 Recently, Liu et al.23 calculated the potential energy surfaces for zig-zag and armchair SWCNTs, and showed that those two different topologies can interact differently when put into contact with water molecules. This observation was later related to the different diffusion mechanisms observed for water confined in topologically and geometrically (10.5 < D (Å) < 22.5) distinct single-walled carbon nanotubes.
Molecular transport can be classified according to two essentially different mechanisms: i) Fickian diffusion,24 when molecules move randomly over short distances, and ii) single-file or ballistic diffusion, when molecular mobility has contributions from preferential paths. While the former is the dominant mode of transport in isotropic phases, such as bulk fluids, the latter has been observed to occur when anisotropy is introduced into the system, e.g., under the form of a nanoconfining solid surface.25 Bhatia and co-workers26 studied the diffusive and viscous transport of CH4 adsorbed into (10,10) armchair SWCNTs, and concluded that the diffusion coefficient decreases with adsorbate density. Later,2 simulations of confined H2 and CH4 indicated that the reflection of fluid molecules on SWCNT walls is essentially specular, thus mitigating momentum loss upon collision; this observation was postulated to occur when the molecular size is relatively larger than the solid lattice parameters dC–C = 1.42 Å. The dynamics of CH4 and C2H6 flowing inside a zig-zag SWCNT have been studied by Krishna et al.,27 who showed that molecules display a Fickian-type mode of transport after 100–500 ps of simulation time. Jacobtorweihen et al.28 performed extensive molecular dynamics simulations with n-alkanes (n < 6) diffusing along zig-zag nanotubes, and estimated that the corresponding self-diffusivities in the zero-loading limit decrease monotonically with the number of carbon atoms of the fluid molecule. Recently, the diffusional properties of cold argon confined inside SWCNT hexagonal bundles were probed by molecular dynamics simulations,29 and it was shown that the nanotube confined fluid exhibits Fickian diffusion, whereas molecules in the interstitial channels move according to a single-file mechanism. Bearing in mind these previous reports, it becomes clear that the diffusivity of gases and liquids within carbon nanotubes is far from being thoroughly understood and, furthermore, results are sometimes conflicting. Water and the smaller alkanes have been the generally studied fluids. However, in all those studies, there is rarely uniformity in the choice of potential models employed. We aim herein to understand the effect of the level of detail used in the description of the fluid's energetics, upon the effective observed dynamics. Much can be gained or lost in terms of computational effort and physical detail by coarse-graining molecular models; the reader is referred to recent reviews for further discussion on coarse-graining techniques.30,31 A particularly challenging issue is the mapping of the diffusivities of coarse-grained particles to those obtained from detailed atomistic descriptions.
From the point of view of molecular interactions, ethylene is an exceptionally interesting and challenging molecule, hence its selection as a prototype fluid. Geometrically, it is a planar molecule; however, the presence of a quadrupole moment, resulting from a highly anisotropic charge distribution, allows different approaches to be employed in order to model the fluid. It is the purpose of this work to study the influence of the intermolecular potential on the dynamical behaviour of ethylene, in a bulk phase, and also inside a single-walled carbon nanotube, thus subject to near one-dimensional confinement. This issue has been partially addressed by Mao and Sinnott32,33 and Cruz and Müller,34 using differently coarse-grained potentials to study the dynamics of C2H6 and C2H4. The one-dimensional self-diffusivities calculated by Mao and Sinnott, corresponding exclusively to molecular translation along the nanotube main axis, showed that ethane exhibits a slightly higher coefficient than its unsaturated analogue. On the other hand, Cruz and Müller determined effective self-diffusion coefficients, considering three-dimensional translation within the confining volume, and their calculations showed that the confined fluid density plays a paramount role upon the dynamics.
Molecular dynamics simulations are here employed along with five distinct molecular potential models, whose level of coarse-graining corresponds to different degrees of detailing the C2H4 molecule. The set of molecular models include a fully atomistic representation of the ethylene molecule,35,36 which is validated by comparison against bulk experimental data,37,38 but also simpler models with varying degrees of coarse-graining ranging down to a single isotropic Lennard–Jones sphere.39 The capabilities and drawbacks of each of these different molecular descriptors will be explored, particularly in the case of confined fluids, where, as far as we are aware, no experimental results are available in the literature. The following parts of this manuscript are organized as follows: the next section introduces the intermolecular potentials used in the calculations, for both the fluid and the solid, as well as the simulation details and methodology (Section 2). We then proceed to present the results obtained, discuss them in light of the microscopic phenomena involved (Section 3), and finally summarize the work with some concluding remarks (Section 4).
Fig. 1 Pictorial representation of the five different intermolecular potentials used in the simulations; each sphere represents a pseudo-atom with a characteristic Lennard–Jones (σ,ε) pair. Partial electrostatic charges, δ, and a point molecular quadrupole, Q, are used in the AA-OPLS and 2CLJQ models, respectively, to calculate electrostatic interactions (see Sections 2.1–2.2 for a detailed description). In the two-center united-atom models (UA-OPLS, TraPPE), L is the chemical bond length between two CH2 (sp2) groups. |
In the present work, we set as benchmark a fully atomistic six-atom description of the ethylene molecule, employing partial Coulombic charges on all sites (AA-OPLS). A first level of coarse-graining adopts a model with two Lennard–Jones centers with an embedded molecular quadrupole (2CLJQ); further coarse-graining leads to the neglect of electrostatic interactions (UA-OPLS, TraPPE), and finally a simple one-site isotropic model is considered (1CLJ). All five potentials calculate the dispersive energies between two interacting particles via a (12,6) Lennard–Jones (LJ) term (eqn (1)), although the corresponding (σ,ε) pairs are obviously potential-specific.
(1) |
(2.1) |
(2.2) |
Atom | q i | ε i /kB (K) | σ i (Å) |
---|---|---|---|
C | −0.230 | 38.270 | 3.55 |
H | 0.115 | 15.107 | 2.42 |
Recently, we have successfully employed the AA-OPLS parameterization to measure the isothermal adsorption and self-diffusion of ethylene confined into an isolated SWCNT,21,34 but the systematic study of the system's dynamical behaviour under different temperature and density conditions is still lacking.
(3) |
Pseudo-atom | ε i /kB (K) | σ i (Å) | Q (B) | L (Å) |
---|---|---|---|---|
CH2 (sp2) | 76.95 | 3.7607 | 4.331 | 1.2695 |
Pseudo-atom, CH2 (sp2) | ε i /kB (K) | σ i (Å) | L (Å) |
---|---|---|---|
TraPPE | 85.0 | 3.675 | 1.33 |
UA-OPLS | 70.437 | 3.850 | 1.34 |
In a planar graphene sheet, carbon atoms are sp2 hybridized, with an overall charge distribution of electrical quadrupoles located perpendicular to the surface. When this sheet is bent to produce a nanotube, the carbon atoms' sp2 orbitals tend to be “squeezed-out” towards the tube’s external volume, and therefore the original quadrupole moments will be dislocated further away from the tube's internal pore.11 This effect becomes more and more pronounced as the SWCNT diameter decreases, and confined molecules will experience a reduced charge anisotropy. For the narrow nanotube diameter considered here, the neglect of quadrupolar interactions with the surface seems adequate, particularly for the highest temperature calculations.57 We have thus excluded contributions arising from the quadrupole moment that could be ascribed to the solid walls. This simplification has been frequently adopted before in studies involving either isolated nanotubes and/or bundled into membranes,58–63 and has a parallel with the approximations made when studying fluid adsorption on graphene sheets.64,65
Nanotube flexibility can play an important role when the carbon lattice is allowed to breathe, however, this usually leads to a dramatic increase of computer processing time. In order to address this issue, Jacobtorweihen et al.28 employed a modified version of the Lowe–Andersen thermostat, thus mimicking the breathing of (13,0) and (20,0) SWCNTs, with diameters D = 10.2 Å and D = 15.6 Å, respectively. They calculated the one-dimensional self-diffusion coefficient of confined n-alkanes (n ≤ 6) along the nanotube's main axis, and concluded that tube flexibility is only important for lighter molecules in the limit of zero-loading. Their calculations for the narrow tube (10.2 Å) considered only C4H10, but, nonetheless, indicate that tube diameter can also play an important role for lengthy molecules: the diffusion coefficient determined for the rigid lattice is roughly 27% higher than the corresponding quantity for the fully flexible structure. Similar findings have been observed by Chen et al.66 for methane confined into a (15,0) SWCNT. Only below a very low pressure threshold, p < 4 × 10−2 bar, did they identify meaningful differences in the self-diffusion coefficients calculated for the rigid and flexible lattices.
Considering all the aforementioned pieces of evidence, it is expected that for systems away from the zero-loading limit, postulating a priorinanotube rigidity would result in minor discrepancies compared to the fully flexible lattice. We have therefore modeled the nanotube walls as rigid, similar to previous studies of fluid transport inside SWCNTs.27,67–71 The carbon nanotube employed in this work has zig-zag simmetry,10,11 (n,0) with n = 16, and diameter D = 12.55 Å. We have already demonstrated that nanotube topology exerts a relatively minor influence on the self-diffusivities of confined quadrupolar fluids.34 The chirality dependency suggested in Liu's work23 can probably be ascribed to the strong electrical dipole of water molecules and the corresponding H-bonding mechanism. In line with previous works57,72 we define the nanotube effective diameter as Deff = D − δ, where D is the skeletal diameter, measured from center-to-center of opposite carbon atoms on the solid walls, and δ = 3.4 Å is the distance of closest approach between a C2H4 molecule and the SWCNT, thus taking into account the softness of the solid–fluid interaction potential. Recently, we determined the δ values for all the lightest alkanes and alkenes (C1–C3).49 As previously mentioned, the bending of a planar graphene sheet to produce a SWCNT squeezes out the π clouds towards the nanotube external space and increases the purely sp2 hybridized σ (C–C) bond length. We have thus considered a value of 1.42 Å for this latter property, in accordance with previous simulation studies.73,74
The effect of temperature on the fluid dynamical behavior was studied over the temperature range 220 ≤ T (K) ≤ 340, by running isothermal simulations at 10 K intervals, in the canonical ensemble (NVT). The array of molecular densities considered spans a wide range, 0.242 ≤ ρ (mol L−1) ≤ 15.751, whose limits correspond to a nanotube containing only two fluid molecules (ρ = 0.242 mol L−1) and a completely filled SWCNT: ρ = 15.751 (AA-OPLS), ρ = 15.024 mol L−1 (2CLJQ, TraPPE), ρ = 14.540 mol L−1 (UA-OPLS), ρ = 11.632 mol L−1 (1CLJ).
In order to establish a comparison with the confined fluid dynamics, bulk phases were also monitored at 300 K using a supercell of 30 Å × 30 Å × 250 Å containing the pure fluid. In both cases, confined and bulk fluid, MD simulations were run for 500 ps, and, in order to avoid any non-Fickian regime,34 statistical sampling was performed after an initial 50–100 ps of simulation time.
(4) |
Fig. 2 Self-diffusion coefficients, D, calculated for the bulk fluid using different potential models at T = 300 K (data for AA-OPLS, ρ < 3 mol L−1, and 2CLJQ from ref. 34). Symbols correspond to simulation (●) or experimental results (▲):37,44 AA-OPLS (black), 2CLJQ (blue), TraPPE (red), UA-OPLS (purple) and 1CLJ (green). Lines are least-squares fittings of the data using Liu's equation79 (eqn (5)) corresponding to the fully atomistic model (black, AA-OPLS) and to the one Lennard–Jones particle (green, 1CLJ). |
From the inspection of Fig. 2, some general trends can be established: i) the 1CLJ model leads to the overall slowest dynamics observed in the MD calculations, and ii) regardless of the specific model employed, the fluid behaves regularly, e.g., the self-diffusion coefficient decreases monotonically with increasing density. As this latter property increases, the free volume available to molecular displacement decreases, and random motion becomes progressively more restricted. Furthermore, in the low-density region, D scales linearly with (1/ρ) in agreement with the Chapman–Enskog theory.80 Jonas and co-workers38 have used spin-echo NMR to measure self-diffusivities of supercritical ethylene (ρ < 21.43 mol L−1), however, the temperatures of their experiments were far to high (323.15–398.15 K) to establish a direct comparison with our MD data. Nonetheless, their findings confirm the general behavior observed in our calculations, clearly showing a monotonical decrease of self-diffusivity with increasing bulk density. From Fig. 2 it can also be observed that, besides minor scattering of simulation results around the overall variation with ρ, the nature of the potential model does not seem to drastically influence the bulk fluid mobility under isothermal conditions. This is not surprising, bearing in mind that the models employed here have been developed from thermodynamical data for the pure bulk fluid. It is worth noting the change in slope exhibited by the experimental data, as molecular density surpasses a threshold of ρ ≈ 11–12 mol L−1. This presumably is a consequence of the enhanced importance of the π–π interactions in the very dense fluid, as C2H4 possess a significant quadrupole moment (Q = 1.3–4 B).81,82 Arends and Trappeniers37 associated this behavior with a departure from the Enskog model for hard-spheres.
The quest for a simple model to relate the self-diffusion coefficient with the corresponding fluid physical properties has been the subject of intense investigation, and is of paramount importance for both theoretical and practical applications. Amongst several others (recent discussions on the topic can be found elsewhere83,84), we have chosen the approach developed by Liu et al. because its was tested for several different pure substances and also some binary mixtures (rare gases, alkanes, alkenes, polar organics),79,85 but essentially due to its mathematical simplicity (eqn (5)). The model builds-upon the Chapman–Enskog transport theory for purely repulsive hard-spheres, DHS, as parameterized by Cohen and Turnbull,86,87 and introduces an exponential term to include the attractive interactions of a Lennard–Jones fluid, D = DHSexp(−αLJ/T*); αLJ = 0.2786 is an empirical constant fitted from molecular dynamics data and T* is the reduced temperature (T* = kBT/εLJ). Under isothermal conditions, only two parameters are required to evaluate the fluid self-diffusivity, namely the Lennard–Jones collision diameter and potential well depth, σLJ and εLJ respectively; with these two quantities at hand, Liu et al. claim that eqn (5) is able to estimate the fluid self-diffusivity under a wide range of densities (0.773–24.47 mol L−1).79 See ESI for further details.†
(5) |
As previously discussed, the degree of potential coarse-graining appears to play a minor role in the bulk fluid dynamics. We have thus applied eqn (5) to the MD data obtained using two limiting potential models, AA-OPLS and 1CLJ, and the corresponding findings are indicated in Fig. 2 and Table 5. Although Liu's approach was not specifically developed for low-density fluids, it can still reasonably predict the ethylene self-diffusivities at very dilute conditions (ρ < 0.1 mol L−1). Nevertheless, its range of validity seems limited, for it slightly underestimates both experimental and simulated self-diffusion data, either with or without electrostatics. In fact, those discrepancies become significant as the fluid density increases beyond ρ > 5.6 mol L−1, and eqn (5) then predicts a much faster decrease of self-diffusivity than the experimentally observed trend. At high densities, eqn (5) becomes insensitive to the particular molecular potential details, and therefore the results obtained for the AA-OPLS and 1CLJ models are observed to converge into similar values (Fig. 2). Table 5 indicates several values for the (σLJ, εLJ) pair, obtained from different sets of thermodynamic data, as well as our own results fitted to the self-diffusivities calculated with the AA-OPLS model. It is worth noting the relatively large σLJ = 4.728 Å value obtained in this work, suggesting that molecular shape is in fact a vital property to accurately describe the fluid's dynamics.
σ LJ (Å) | ε LJ (K) | Optimized property | Reference |
---|---|---|---|
4.232 | 205 | viscosity | Hirschfelder80 |
4.523 | 199.2 | second virial coefficient | Hirschfelder80 |
4.163 | 224.7 | viscosity | Reid88 |
4.047 | 215.3 | self-diffusivity | Zhu89 |
4.155 | 225.6 | viscosity | Mourits90 |
4.124 | 215.8 | critical point | Müller91 |
4.048 | 169.1 | self-diffusivity | Liu79 |
4.140 | 224.2 | molecular density | Calado92 |
4.257 | 201.9 | viscosity | Tee93 |
4.433 | 202.5 | second virial coefficient | Tee93 |
4.589 | 200.8 | vapor pressure | Cuadros39 |
4.728 | 172.3 | self-diffusivity | this work |
Fig. 3 Influence of the intermolecular potential upon bulk and confined C2H4 dynamics at T = 300 K: top) Self-diffusion coefficients for ethylene in the whole density domain (data for AA-OPLS, ρ < 3 mol L−1, and 2CLJQ from ref. 34); bottom) projection onto the (D,ρ) plane, showing the two areas corresponding to a bulk (ρ ≤ 2.969 mol L−1) and confined fluid (ρ ≥ 0.242 mol L−1). Symbols are MD data and lines correspond to a power-law, D = D0ρλ (Table 6); note that, for each individual potential, both bulk and confined self-diffusivities seem to collapse onto the same master-curve (see text for details). Data points and lines are colored according to the molecular potential employed: AA-OPLS (black), 2CLJQ (blue), TraPPE (red), UA-OPLS (purple) and 1CLJ (green). In snapshots, ethylene molecules are represented as a two-center Lennard–Jones fluid and the SWCNT by its graphitic carbon mesh. |
Molecular potential | D 0 × 107 | λ |
---|---|---|
AA-OPLS | 2.247 | 1.229 |
2CLJQ | 1.784 | −1.218 |
TraPPE | 4.720 | −0.943 |
UA-OPLS | 2.899 | −1.169 |
1CLJ | 1.368 | −1.269 |
Overall, it can be observed from Fig. 3 that the models with explicit electrostatic interactions (AA-OPLS and 2CLJQ) systematically lead to lower values of self-diffusivities, while on the other hand, the TraPPE and UA-OPLS that neglect Coulombic energies, exhibit the highest values. The isotropic 1CLJ potential seems to behave in a spurious fashion, leading to the lowest observed self-diffusivities. It was already demonstrated that the 1CLJ model led to the slowest dynamics observed in the bulk fluid (Fig. 2), and a similar conclusion can be extrapolated to the confined phase, where molecular displacement becomes severely restricted.
Considering the two cases in which electrostatic interactions are explicitly considered (AA-OPLS and 2CLJQ), it is observed that the differences in the corresponding self-diffusivities are rather small and approximately constant over the entire density range (Fig. 3); however, the 2CLJQ potential somehow results in smaller calculated D values. As clearly pointed out by Vrabec and co-workers,42 their parameterization of the ethylene molecule leads to a quadrupole moment which is ca. 0.331 B higher than the upper limit of the range of reported experimental values (Q = 1.3–4 B),81,82 thus artificially enhancing fluid–fluid electrostatics and possibly resulting in a slowing down of molecular translation. Because electrostatic interactions are long-ranged, the effect just mentioned should influence the results obtained not only for the dense but also for the very dilute fluid, thus explaining the constant ΔD differences between the AA-OPLS and 2CLJQ models.
The endoadsorption of quadrupolar molecules in zig-zag carbon nanotubes (7.86 ≤ D (Å) ≤ 14.89) has been addressed before using grand canonical Monte Carlo simulations.95 It was observed that, due to the strongly confining nature of such one-dimensional geometries, a strong molecular quadrupole such as CO2 (Q = 1.64–4.87 B)46 causes alignment of the fluid in a staggered configuration. In face of this, one can hypothesize the existence of molecular clusters held together by strong electrostatic coupling, which induces some local structure in the fluid that inhibits molecular translation inside the nanopore, and thus results in smaller self-diffusion coefficients compared to the potentials where electrostatics are omitted (TraPPE and UA-OPLS). In Fig. 4 we plot the radial distribution function8 of the confined dense fluids (eqn (6)), calculated for the individual carbon centres, for three representative coarse-grained potentials: AA-OPLS, TraPPE, and 1CLJ.
(6) |
Fig. 4 Radial distribution functions of dense confined ethylene, calculated for the individual carbon centres, at T = 300 K: AA-OPLS (black, ρ = 15.024 mol L−1), TraPPE (red, ρ = 15.024 mol L−1) and 1CLJ (green, ρ = 11.632 mol L−1). |
The simple one-center Lennard–Jones fluid exhibits a very sharp peak at r = 4.73 Å and a second (less intense) long-range peak located at r = 8.73 Å, indicating that molecules are densely packed together with almost no free volume between them (σ1CLJ ≈ 4.6 Å). If the fluid potential is to be increasingly refined, up to the atomistic level, a different picture can be obtained from Fig. 4. When the C2H4 molecule is described with the TraPPE model, not only does the initial g(r) peak becomes much broader, with an half-width length of Δr ≈ 1.95 Å, but it also appears closer at r = 4.48 Å; the second peak observed in the 1CLJ model is no longer clearly observable. If electrostatics are now explicitly added to the fluid molecules, via atomic partial charges (AA-OPLS), a further refinement of the initial peak takes place, as indicated by the magnification inset of Fig. 4. The r = 4.48 Å peak of the TraPPE potential essentially gets split into two contributions, r = 4.03 Å and r = 4.78 Å, suggesting the insurgence of a very short-range ordered arrangement of molecules in the AA-OPLS model, that is not present for the other potentials.
Fig. 5 Temperature dependence of the confined fluid self-diffusion coefficient, in the range 220 ≤ T (K) ≤ 340. Symbols represent molecular dynamics results and lines correspond to an Arrhenius fit, D = Aexp(−Ea/RT). For comparison purposes, the Arrhenius plots obtained for the AA-OPLS fluid are indicated as dotted lines in both the TraPPE and 1CLJ charts. Note that the three different densities indicated include two limiting situations: when the nanotube contains only two fluid molecules (0.242 mol L−1), and when it is filled with fluid (11.632 mol L−1; 15.024 mol L−1). |
The isochors in Fig. 5 confirm our observations that potential coarse-graining plays a paramount role when studying the dynamics of confined ethylene. At constant density, it is observed that the self-diffusion coefficients obey a general trend: D (1CLJ) < D (AA-OPLS) < D (TraPPE). Simulation data show that a temperature rise in the system has a direct effect on the fluid dynamics, with a monotonical relationship between self-diffusivity and temperature. This effect can be interpreted on the basis of kinetic theory, considering that a temperature increase in the system has a direct consequence in terms of molecular kinetic energy, Ekin = (3/2)kBT, and thus molecular velocities respond to applied temperature according to .96 We consider the temperature dependence of the self-diffusional process to be determined by an Arrhenius law:
D = Aexp(−Ea/RT) | (7) |
The results obtained from fitting our data to the Arrhenius equation are recorded in Table 7. It is quite interesting to observe that for each potential model studied, regardless of its own specificities and degree of molecular detail, the activation energy (Ea/R) decreases monotonically with increasing molecular density. Bhatia2 studied the transport of small molecular fluids (H2, CH4) confined in armchair nanotubes, and observed that reflection of fluid molecules on the solid walls was dominantly specular, suggesting that momentum transfer occurs essentially via fluid–fluid collisions. Because molecules in the dense fluid are highly packed, with little free volume between them, exchange of momentum via molecular collisions is highly efficient, thus dramatically increasing the self-diffusional process as a response to temperature. On the other hand, as the fluid density decreases, molecules are increasingly far away from each other, with long mean free paths between them, and thus momentum transfer is restricted. The results recorded in Table 7 highlight the hydrophobic nature of carbon nanotubes. Liu and co-workers23 determined the activation energy for water diffusion (ρ = 55 mol L−1) inside a Deff = 15.39 Å zig-zag carbon nanotube as Ea ≈ 1941 K. At an ethylene density of the same order of magnitude as Liu's, ρ(C2H4) > 11 mol L−1, the self-diffusion of ethylene has an activation energy which is ca. 5–8 times lower than that of confined water.
Molecular potential | ρ (mol L−1) | A (m2s−1) | E a/R (K) |
---|---|---|---|
AA-OPLS | 0.242 | 5.160 × 10−6 | 636.95 |
5.331 | 2.710 × 10−7 | 533.96 | |
15.024 | 2.254 × 10−8 | 242.85 | |
TraPPE | 0.242 | 1.870 × 10−5 | 678.29 |
5.331 | 7.508 × 10−7 | 614.64 | |
15.024 | 5.995 × 10−8 | 310.05 | |
1CLJ | 0.242 | 9.379 × 10−7 | 545.81 |
5.331 | 5.913 × 10−8 | 437.35 | |
11.632 | 1.867 × 10−8 | 380.44 |
The activation energies and pre-exponential factors of the Arrhenius law are plotted in Fig. 6 as a function of molecular density, and a relationship seems to exist between those two properties; a more minute mapping of the [(Ea/R), A, ρ] space would be required to fully confirm this observation, ideally in the very low and ultra high density regions. Nonetheless, in the density domain accessed in the present study, both Arrhenius parameters exhibit nonlinear decays with ρ, which can be reasonably estimated with the empirical equations:
(Ea/R) = (Ea/R)0 − bρ3/2 | (8.1) |
lnA = lnA0 − bρ1/2 | (8.2) |
Fig. 6 Dependence of the Arrhenius parameters, activation energy (Ea/R) and pre-exponential factor (A), with the confined fluid molecular density. Symbols are the results of fitting MD data with the Arrhenius law and lines correspond to nonlinear functions, (Ea/R) = (Ea/R)0 − bρ3/2 and lnA = lnA0 − bρ1/2. The dashed line is an extrapolation of the 1CLJ data with the parameters ((Ea/R)0,b) calculated for the upper (ρ = 5.331 mol L−1) and lower (ρ = 11.632 mol L−1) data point error intervals. |
The curves indicated in Fig. 6 suggest that there is an initial (small) region of molecular densities where (Ea/R) is roughly constant, corresponding to an infinitely dilute fluid, (Ea/R)0 = 546 K (1CLJ) < (Ea/R)0 = 628 K (AA-OPLS) < (Ea/R)0 = 685 K (TraPPE). After that initial region, the activation energies decrease more noticeably with density; the precise boundary is a function of the particular potential model employed. When the pre-exponential factor is considered, the data plotted in Fig. 6 indicate that the former decreases monotonically as the nanotube becomes completely filled with fluid. If we consider the slope of eqn (8.2) as a measure of susceptibility regarding molecular density (bTraPPE = 6.4, b1CLJ = 6.6, bAA-OPLS = 6.7), it can be concluded that the three potentials respond in a similar manner, within the errors associated with the corresponding fittings.
As previously mentioned, we are unaware of any experimental work reporting self-diffusion data for confined ethylene. Nonetheless, and in order to probe the simultaneous effect of confinement and temperature on the fluid dynamics, we compare in Fig. 7 our results with the bulk experimental data measured by Arends et al.37 and Baker et al.38 using the NMR spin-echo technique. Because the AA-OPLS potential is the one that best reproduces the bulk experimental measurements (Fig. 2), we proceed our discussion using this model as a benchmark for the simulation data. In the limit of complete tube filling, entropic effects are usually dominant21 and thus packing efficiency rules over energetic considerations. It is clear from Fig. 7 that, when the solid is saturated with fluid (15.024 mol L−1), the self-diffusion coefficients show a similar response to applied temperature, both in the bulk phase and under confinement. In fact, the activation energies obtained from the corresponding Arrhenius plots are (Ea/R)bulk = 214.12 K and (Ea/R)SWCNT = 242.85 K, suggesting that confinement exerts a major influence on the absolute values of the mobility constants, but not so much on their intrinsic dependence on temperature. To a reasonable approximation, the self-diffusivity differences between the bulk and confined phases can be considered as roughly constant over the entire temperature range, (Dbulk/DSWCNT)T ∈ [220K–340K] = 2.52 ± 0.13.
Fig. 7 Temperature dependence of the fluid self-diffusion coefficient (220 ≤ T (K) ≤ 398.15). Symbols represent experimental data measured for bulk ethylene37,38 and lines correspond to the Arrhenius equation (eqn (7)) for the bulk (dashed) and confined (full) phases. The confined ethylene results were obtained using the AA-OPLS potential at two different densities. |
A totally different picture results when density is lowered, and the existence of free volume becomes a relevant aspect. The upper two isochores of Fig. 7 correspond to a bulk and confined fluid at ρ = 5 mol L−1 and ρ = 5.331 mol L−1. At those densities, free volume inside the solid corresponds to more than 50% of the total available volume, and therefore a sharp discrepancy can be observed in the slopes of the isochores, originating in activation energies that are quite different between the bulk and confined fluid, (Ea/R)bulk = 301.64 K and (Ea/R)SWCNT = 533.96 K. For those lower densities, the self-diffusivities ratio increases dramatically to (Dbulk/DSWCNT)T ∈ [220K–340K] = 10.2 ± 0.5. One can extrapolate the previous observations to a very dilute fluid, and postulate that as the free volume inside the solid increases, so do the differences in the corresponding activation energies between the bulk and confined fluid. Unfortunately, this could not be validated by comparison with experimental findings, for as far as we are aware literature data do not reach such a low density region as the one studied in the present work (ρ = 0.242 mol L−1).
It is generally argued that coarse-graining intermolecular potentials results in increased self-diffusivities, an effect which is commonly ascribed to a smoothing of the overall molecular shape. In the present case, if the AA-OPLS model is used as a benchmark, it becomes clear (cf.Fig. 2–3) that the effect of coarse-graining is unpredictable, and an appropriate and accurate parameterization may faithfully reproduce the self-diffusion data. Furthermore, the existence of explicit electrostatic details in the potential seems to induce short-range order in the confined fluid (Fig. 4), resulting in a slowing down of molecular mobility. In spite of the marked anisotropy introduced by the solid, both the bulk fluid and anisotropic systems diffusion data are observed to roughly collapse onto a same master curve, which in this work was described as a simple power-law dependency of the self-diffusion coefficient with molecular density, D = D0ρλ (Table 6).
In order to monitor the influence of temperature upon the confined fluids, three representative intermolecular potentials were chosen, AA-OPLS, TraPPE and 1CLJ, corresponding to different degrees of coarse-graining. It was demonstrated that the temperature induced response of the self-diffusivites, although clearly potential dependent, could be reasonably described by the Arrhenius law. The activation energies, (Ea/R), and pre-exponential factors, A, thus calculated, indicate non-linear relationships between those properties and molecular density (eqns (8.1) and (8.2)), which, as far as we are aware, have previously been unnoticed. Using the AA-OPLS potential, we determined the activation energy for ethylene self-diffusion inside the SWCNT as 242.8 ≤ (Ea/R) (K) ≤ 636.95, roughly 5–8 times smaller than the corresponding value for water.23 Finally, considering the AA-OPLS model as a benchmark for simulation data and comparing with bulk fluid experimental measurements, one concludes that: i) under isochoric conditions, the influence of temperature upon molecular diffusion can be reasonably described by the Arrhenius law, both in the bulk and confined systems, and ii) as molecular density approaches that of a very dilute fluid, the effect of confinement upon the fluid's dynamics increases, leading to a dramatic increase of the activation energies compared to the isochoric bulk phase.
F.J.A.L. Cruz gratefully acknowledges financial support from FCT/MCTES (PT) through grant SFRH/BPD/45064/2008.
Footnote |
† Electronic supplementary information (ESI) available: Liu's equation. See DOI: 10.1039/c1ra00019e |
This journal is © The Royal Society of Chemistry 2011 |