ARTICLE pubs.acs.org/JPCC
DFT Study of 1,3-Dimethylimidazolium Tetrafluoroborate on Al and Cu(111) Surfaces T. P. C. Klaver,*,† M. Luppi,† M. H. F. Sluiter,† M. C. Kroon,‡ and B. J. Thijsse† † ‡
Department of Materials Science and Engineering, Delft University of Technology, Mekelweg 2, 2628 CD Delft, Netherlands Department of Chemical Engineering and Chemistry, Eindhoven University of Technology, Den Dolech 2/P.O. Box 513, 5600 MB Eindhoven, Netherlands ABSTRACT: We present density functional theory results of the ionic liquid 1,3-dimethylimidazolium tetrafluoroborate ([mmim][BF4]) adsorbed on Al and Cu(111) surfaces. Results comprise both relaxed configurations and constrained ab initio molecular dynamics simulations of up to 444 atoms for 19 ps. Relaxation results show that for submonolayer coverage many aspects of adsorption energies, electron transfer, and bond length variation can be explained from a simple bond saturation picture. The ions have a relatively weak energetic preference to interact with each other over interacting with the surface. Electron density accumulation and bond length changes in the [mmim+] ion are mostly independent of what the ion interacts with. In many cases, the shape of the surface on which a [mmim][BF4] pair is situated is also of little importance. At submonolayer coverage, [BF4] ions have a stronger interaction with Al and Cu surfaces than [mmim+] ions do, and as a result the latter have greater mobility on the surface. When [mmim][BF4] pairs move across an Al surface, the migration energy is determined mainly by how close the [BF4] ion can nestle itself against the surface. On Cu this is not the case because [BF4] interacts less strongly with the surface than on Al. Both on Al and Cu, the energy required to move a [mmim][BF4] pair across the surface is low. Molecular dynamics results show that while relaxation results can be useful in understanding some aspects of the behavior of ionic liquids on surfaces there are clear limitations to their usefulness. Even in low-temperature dynamics simulations, [mmim+] ions on Al spend much of their time in positions very different from their energy minima. At low coverage, this leads to a relative [BF4] enrichment that suggests some degree of anioncation layering at the Al surface. Simulations with more [mmim][BF4] pairs did not provide extra evidence for layering, as at greater coverage the [mmim][BF4] dewetted from the surface to form a tiny droplet. The dewetting and layering results from molecular dynamics simulations provide possibilities for experimental verification.
1. INTRODUCTION Ionic liquids (ILs) have been a very active area of research for over a decade now due to properties that make them excellent candidates for a range of applications. These properties, some of which can be tailored through the right combination of cations and anions, include negligible vapor pressure at room temperature, low melting point, high thermal (low volatility) and chemical (low flammability) stability, a broad liquidus range, ionic conductivity, a large electrochemical window,1,2 catalytic activity, and good solvent and miscibility properties for organic compounds. The low volatility enables the use of ILs as a substitute for environmentally harmful volatile organic compounds (green technology). Applications include synthetic, catalytic, extraction, and separation purposes, electrochemistry1,3 (battery electrolytes, solar cells, fuel cells, electrochemical capacitors, electrodeposition2), and lubrication.46 Experimental methods used to study IL structure have included sum frequency vibrational spectroscopy,79 (surface-enhanced) Raman spectroscopy, (Fourier transform) infrared spectroscopy,3,10 direct recoil spectrometry, nuclear magnetic resonance spectroscopy,1012 X-ray diffraction,11,13 X-ray reflectivity,14,15 X-ray absorption finestructure, X-ray photoemission spectroscopy, neutron diffraction, neutron reflectivity, (tapping mode) atomic force microscopy,16,17 r 2011 American Chemical Society
and scanning tunnelling microscopy. References 18 and 19 give overviews (not exhaustive) of experimental work done on ILs. For many applications, the interaction of the IL with a solid surface is important. While bulk ILs have been studied extensively for over a decade, only recently extensive attention has been given to ILsolid surface interaction. Many of the aforementioned experimental techniques are unsuitable to study interfaces. Atomistic simulation, with its easy options for very detailed analysis, does not suffer from this drawback, and indeed ILs have been studied with both empirical force field potentials12,2029 and electronic structure methods.3033 ILs are governed by a complex interplay of Coulomb interactions, covalent bonds, hydrogen bonding, and van der Waals forces. Electronic structure methods are capable of handling aspects of interactions that can be difficult to capture in an empirical force field potential, like polarizability,12 electron transfer, and hydrogen bonding. The higher accuracy does come at the price of much higher computational cost than for force field potentials. This limits the use of electronic structure methods for studying ILsolid Received: January 14, 2011 Revised: June 20, 2011 Published: June 25, 2011 14718
dx.doi.org/10.1021/jp200401h | J. Phys. Chem. C 2011, 115, 14718–14730
The Journal of Physical Chemistry C interfaces, as a substrate of meaningful dimensions already requires numbers of atoms that represent a considerable computational effort, without putting any adsorbates on it yet. Electronic structure studies of molecules on surfaces published so far usually involve a single or a few small adsorbed molecules only.3442 The only ILsurface system studied with electronic structure calculations that we are aware of is 1-ethyl-3-methyl imidazolium tetrafluoroborate ([emim][BF4]) on Li, by Valencia et al.43,44 In this paper, we present what is, to our knowledge, the first electronic structure study of an IL on a solid surface involving constrained ab initio molecular dynamics (MD) with up to 12 ion pairs. Given the early stages of simulating ILs on surfaces with electronic structure methods, we started off with a fairly simple system. The most common cations in ILs are imidazolium-based. We chose to put methyl side groups on the N atoms to form a small, symmetrical cation that diffuses relatively fast during the relatively short MD times. The anions in ILs are usually small inorganic ions that are all relatively uncomplicated. We chose tetrafluoroborate to arrive at 1,3-dimethylimidazolium tetrafluoroborate ([mmim][BF4]) for our choice of IL. The choice of the substrate was also informed by expediency, i.e., using fcc (111) metal substrates. For the choice of metal elements, we picked two rather different metals to see which results are similar between them and which ones differ. One is the normal p-metal Al, the other the late transition d-metal Cu. Experiments have shown1,2 that Au(111) metal surfaces can reconstruct or restructure when in contact with ILs (dependent also on bias voltage). However, for our relatively small surface areas and short MD times, the (111) surfaces remain unreconstructed substrates. The electronic structure method we have employed is density functional theory (DFT) with gradient-corrected interpolations for the exchangecorrelation functional. Standard local density approximated or gradient-corrected DFT methods do not include a correct description of dispersive interactions (van der Waals forces). McNellis et al.45,46 have looked at correction schemes for molecules on coinage metal surfaces (though not ILs, but rather azobenzene) and found that the required corrections can be significant. B€uhl et al.33 showed that for 1,3-dimethylimidazolium chloride various DFT methods reproduce second-order Møller Plesset results reasonably well, though still not perfectly. This means that while our electronic structure method offers great improvements in accuracy over empirical force field potentials in several ways there is also a known shortcoming. The choice of small methyl groups on the imidazole ring helps to limit the region of the cation for which van der Waals interaction is relatively important. The lack of a proper inclusion of dispersive interactions for a system in which they do play a role, plus other issues related to compromises we made for the sake of computational expediency (see Computational Details section), do mean that our DFT results do not have the same accuracy as can be obtained for, e.g., bulk metals.
2. COMPUTATIONAL DETAILS Results were calculated with the mainstream DFT package VASP.47,48 VASP is a plane wave code that implements the PAW method.49,50 Standard PAW potentials supplied with VASP were used, with exchange and correlation described by the PBE parametrization in the generalized gradient approximation. The number of electrons within spheres around the atom nuclei was determined, with the radii set to the default values in the PAW
ARTICLE
Table 1. Numbers of Valence Electrons nvalence and Sphere Radii r for Evaluating the Numbers of Electrons on Atoms nvalence
r (Å)
Al
3
1.402
Cu C
11 4
1.312 0.863
N
5
0.741
H
1
0.370
B
3
0.794
F
7
0.905
potential files. Table 1 lists the numbers of valence electrons and sphere radii used for the elements in our simulations. Evaluating the number of electrons within a sphere around a nucleus is not a very accurate measure for determining numbers of electrons since some electron density will be attributed to multiple atoms (when spheres overlap) while other electron density is not taken into account at all if it falls outside any sphere. Charge differences between configurations determined in this way are therefore very useful for observing trends only. Charge determination through Voronoi polyhedra avoids the problem of double sampling some density while not sampling other density at all. However, constructing Voronoi polyhedra in a system with a lot of open space can also lead to problems, as large polyhedra may wrongfully attribute unrealistically large electron densities to atoms. In our attempts to determine charges inside Voronoi polyhedra, we found that charges attributed to atoms that should have been similar (the atoms being in perfectly symmetrical positions) varied by up to almost 2/3 of an electron. We have used charge determination inside Voronoi polyhedra only to determine the total charge on cations and anions because in that case the deficiencies mostly cancel out. Voronoi polyhedra can be constructed both with or without taking into account the relative sizes of atoms. In our experience, the results in which relative sizes were taken into account were the most credible. The radii in Table 1 were used for determining the size ratios between atoms that determine where polyhedral planes are placed between atoms. For relaxation calculations, the plane wave energy cutoff was set to 450 eV, which is sufficient for convergence of energy differences. For the electronic convergence criterion, the default setting of 0.0001 eV was used, and for ionic relaxation it was set to 0.01 eV/Å. We did not use dipole corrections in our calculations since they can not be applied in the same way in all our calculations and would therefore introduce an inconsistency in our data. Seven tests of ionic pairs in vacuum and of single ions and ion pairs on both Al and Cu surfaces showed that dipole corrections amounted to hundredths of electronvolts at most. Thus, an error of that magnitude is present in our results due to periodic image interaction of the long-ranged Coulomb tails. Also, all our systems were chargeneutral, removing the requirement of dipole corrections for charged systems. Isolated BF4 and mmim radicals in vacuum were the only systems that were calculated with spin polarization. While our systems contain hydrogen, we did not apply zero-point energy corrections. In most calculations, substrates consisted of only three (111) monolayers. Tests were carried out to evaluate the change in an interaction of 0.41 eV between a fixed, unrelaxed [mmim][BF4] pair and a slab as a function of the number of monolayers. Increasing the slab thickness from three to four, five, or six monolayers changed the interaction by 23 and 10 meV at most for Al and Cu. Forces 14719
dx.doi.org/10.1021/jp200401h |J. Phys. Chem. C 2011, 115, 14718–14730
The Journal of Physical Chemistry C
ARTICLE
Figure 1. Two [mmim][BF4] pairs on a 22.87 19.80 Å2 Al(111) surface. The black Al atoms denote the periodicities over which [mmim][BF4] pairs were displaced to determine energy barriers (see Section 3.4).
on the atoms of the [mmim][BF4] also remained very similar. Given the other sources of error in our calculations, we have accepted the small error due to nonconvergence of the slab thickness when using only three monolayers. While it may seem hard to believe that only three monolayers already give reasonable convergence, others have also reported DFT results of molecules on fcc (111) metal slabs only three to five monolayers thick34,35 and found that three was sufficient.36,40,41 In the current calculations, one monolayer contained 64 atoms. The in-plane supercell dimensions were held fixed to values corresponding to the DFT equilibrium lattice parameters for Al and Cu at 0 K. This means that for Al the surface area was 22.87 19.80 Å2 and for Cu it was 20.52 17.77 Å2. Periodic boundary conditions were used to give the impression of an infinitely large substrate in the in-plane directions. The open space between periodic slabs was 20 Å. This means that for Al and Cu supercells dimensions are 22.87 19.80 24.67 Å3 and 20.52 17.77 24.19 Å3. To give an impression of the size of the ionic pairs compared to the slab size, Figure 1 shows two pairs on an Al surface. The binding energy Eb between ions in an ion pair and between ions and surfaces is calculated according to Eb ¼ ðEðcombinedÞ Eðobject1Þ Eðobject2ÞÞ
ð1Þ
In case the separate objects are, e.g., the BF4 and mmim radicals, the binding energy is not the energy that shows how much energy it would take to separate [mmim+] and [BF4] but rather how much it would take to separate them and undo the charge transfer between them, turning them into charge-neutral radicals. Brillouin zone sampling was done using the MonkhorstPack scheme. It was found that 2 2 k-points in the slab directions were sufficient to converge absolute energies down to 0.01 electron under [mmim+]. Finally, in this section we report the adsorption energies and induced charges on the substrate layers for systems with two [mmim][BF4] pairs (as shown in Figure 1). The adsorption energies of the second [mmim][BF4] pair onto Al and Cu with already one ionic pair present are 0.23 and 0.50 eV higher (so 1.05 + 0.23 = 1.28 eV for Al, 0.76 + 0.50 = 1.26 eV for Cu) than for the first [mmim][BF4] pair (Table 5) that absorbs on a bare surface. The 0.23 and 0.50 eV values represent attractive interaction between the [mmim][BF4] pairs. This interaction is most likely due to Coulomb attraction between ions (of opposite charge) in the first and second ion pair. This leads to a stronger adsorption of the second pair. While the positioning of the [mmim][BF4] pairs in our calculations was just one of many different possibilities (and includes periodic image interaction between [mmim][BF4] pairs that are at much closer distances than for systems with just one ionic pair), the result shows that the typical attractive interaction between [mmim][BF4] pairs is on the order of a few tenths of electronvolts. The electron transfer on the [mmim][BF4] pairs 14723
dx.doi.org/10.1021/jp200401h |J. Phys. Chem. C 2011, 115, 14718–14730
The Journal of Physical Chemistry C
ARTICLE
Figure 5. Binding energy gained by merging charge-neutral molecules into an in-plane ionic pair in vacuum, adsorbing the molecules or ionic pair onto a surface, and merging the separate ions on the surface into an in-plane pair on the surface. Left: on Al. Right: on Cu.
Figure 6. [BF4] adsorbed on the Al surface. Two Al surface atoms that have been pulled up from the surface by 0.25 Å are highlighted in black. The minimum AlF distance is 2.05 Å.
is rather similar to that on a single [mmim][BF4], except of course if the [BF4] ions relax to a different orientation than for a single pair and different F atoms have binding with the neighboring [mmim+] or the substrate. The charges induced on substrate atoms were different again for Al and Cu. The numbers of substrate atoms with induced charges were comparable to the case of a single [mmim][BF4] for both Al and Cu; however, on Al the maximum induced charge was 0.16 electrons, or three times the value for a single [mmim][BF4] pair, while for Cu the maximum value was the same as for a single pair. At present we have no good explanation for the differences between Al and Cu or for the strongly nonlinear charge accumulation on Al. 3.3. Molecules on Nonflat Surfaces. The simple bond saturation picture that explains (or at least coincides with) many of the results shown so far suggests that [mmim][BF4] may not be very sensitive to which substrate atoms it transfers its electron density to or from. Thus, it would seem likely that [mmim][BF4] would be indifferent to, e.g., lying along a step edge rather than on a flat surface. To test this hypothesis, [mmim][BF4] pairs were placed near a step edge on Al and Cu in three ways and then fully relaxed (see Figure 7). To exclude the part of the binding energy differences that is the result of rotation of the [mmim+] or [BF4] ions, we also took configurations of relaxed [mmim][BF4] on flat Al and Cu and embedded them in one or two extra surrounding monolayers (see Figure 8). These configurations are very artificial and were not relaxed. The only reason for calculating them is to determine the increase in binding compared to the flat surface. Since there is no relaxation, such an increase is only due to the presence of ∼20 extra neighbors per extra coordinating monolayer, at distances of 4 Å or more. Table 6 compares the [mmim][BF4]substrate binding energies on flat surfaces and the surfaces shown in Figures 7 and 8. For Al the energy differences between different configurations are not much bigger or even far smaller than the energy difference
between different configurations on flat Al (see Figure 9 in Section 3.4). For both Al and Cu, adding the second extra coordinating monolayer gives much less extra binding than the first one (for Cu it is even easily within the error margin of our calculations, hence there is no significant increase in binding at all). At present we have no good explanation why on Cu there are substantial energy differences between different configurations if [mmim][BF4] is fully relaxed. 3.4. Displacing [mmim][BF4] over Flat Surfaces. Apart from knowing how large the energy differences are between different configurations, it is also useful to know how large the energy barriers are between these different configurations. To get some impression of how easily or difficult [mmim][BF4] moves across Al and Cu surfaces, a [mmim][BF4] pair was displaced one lattice period in both in-plane directions (see Figure 1). The starting positions were the fully relaxed [mmim][BF4] pairs on Al and Cu. The [mmim][BF4] pair was moved in 10 steps in both directions. Full relaxation of the [mmim][BF4] would probably have resulted in many of these 10 different configurations along the path relaxing into the same few local energy minima. Hence, the in-plane coordinates were held fixed, and only in the direction perpendicular to the surface were [mmim][BF4] atoms allowed to relax. A similar approach was followed for rotating [mmim][BF4]. [mmim][BF4] was rotated 180° in 9° steps around an axis perpendicular to the surface, situated (somewhat arbitrarily chosen) at atom C4, and then the height coordinates of the [mmim][BF4] atoms were allowed to relax. There is no symmetry between 0 and 180° and 180360°. While the 20 data points obtained for rotation are sufficient for our purposes, extending the rotation to 180360° would have yielded different values. The left figures in Figure 9 show the total energy of the Al systems with relaxed [mmim][BF4] height coordinates as a function of displacement or rotation. The correlation between the system energy and the height of the atoms of the [BF4] ion is strong. It would appear that the energy required to move [mmim][BF4] is fully determined by how closely [BF4] can nestle itself against the surface. By contrast, the average height of the C atoms changed less than 0.01 Å when moving [mmim][BF4] in the x direction. The imidazolium ring with side groups is probably too spatially extended to “feel” and sink into the open spaces between the Al surface atoms. In addition to that, it has already been shown that [mmim+] interacts less strongly with the metal surfaces than [BF4]. On Cu we did not find the strong energy[BF4] height correlation. It has already been noted in Section 3.2 that [BF4] does not pull up Cu surface atoms as far as it pulls up Al atoms and that the [BF4] adsorption energy is lower on Cu than on Al. 14724
dx.doi.org/10.1021/jp200401h |J. Phys. Chem. C 2011, 115, 14718–14730
The Journal of Physical Chemistry C
ARTICLE
Figure 7. Relaxed [mmim][BF4] near step edge on Cu. Left: configuration with the [BF4] part hanging over the step edge. Middle: configuration with the [mmim+] part hanging over the step edge. Right: [mmim][BF4] lying along the step edge. To clarify the position of the steps edge, the first and last atoms of it have been colored black.
Figure 8. Left: [mmim][BF4] on Al surface, embedded in one extra coordinating monolayer. The Al atoms in the extra layer closest to the [mmim][BF4] are highlighted in black. Top right: grazing side view of system on the left. Bottom right: side view of [mmim][BF4] on Al in two extra coordinating layers.
Table 6. Binding Energies ΔE of [mmim][BF4] in Different Configurations ΔE (eV) configuration
Al
Cu
[mmim][BF4] on flat substrate
1.05
0.76
[mmim][BF4] on flat substrate, one extra coordinating layer
1.26
0.82
[mmim][BF4] on flat substrate, two extra coordinating layers 1.38
0.83
[BF4] hanging over step edge
1.11
0.80
[mmim+] hanging over step edge [mmim][BF4] along step edge
1.11 1.14
1.36 1.06
Since [BF4]Cu interaction is weaker, it does not determine the displacement energy of [mmim][BF4] on Cu.
Whether the displacement energies for translation and rotation are dominated by [BF4]surface atom interaction or not, they are always low, on Al (hundredths of electronvolts for translation, tenth of electronvolts for rotation) and even more so on Cu (never more than hundredth of electronvolt), where strong [BF4]surface interaction is absent. This is the case even with the artificial constraints that [mmim][BF4] can only move in certain directions or that it must rotate around a particular fixed atom. If [mmim][BF4] were allowed to move along any low energy path, it could probably move almost unhindered. 3.5. Ab Initio Molecular Dynamics. The results in Section 3.3 have shown that most individual relaxed configurations are not particularly special. The results obtained about them are interesting only to the degree that they are representative of the behavior of [mmim][BF4] on Al or Cu in general. In addition, 14725
dx.doi.org/10.1021/jp200401h |J. Phys. Chem. C 2011, 115, 14718–14730
The Journal of Physical Chemistry C
ARTICLE
Figure 9. Total energy (left) and average height of F and B atoms (right) for Al surfaces with [mmim][BF4] being translated and rotated. The in-plane displacement is enforced, and the height coordinates of the [mmim][BF4] atoms are allowed to relax for each in-plane position of the [mmim][BF4]. Top: data for [mmim][BF4] being displaced one periodicity along the x-direction in Figure 1. Middle: data for [mmim][BF4] being displaced one periodicity along the y-direction in Figure 1. Bottom: data for [mmim][BF4] being rotated half a turn around atom C4 in 9° steps.
results in Section 3.4 show that [mmim][BF4] pairs can easily move from one position into another. For many aspects of the behavior of [mmim][BF4] on Al or Cu surfaces, more can probably be learned from letting [mmim][BF4] pairs move over the surfaces by themselves in a MD simulation than from studying individual relaxed configurations. Ab initio MD simulations of 1, 2, 4, and 12 [mmim][BF4] pairs on Al were run up to 25 ps. Fast vibrations within [mmim][BF4] were frozen so that a large 5 fs time step could be used. As the bond lengths and angles between all C and N atoms in the [mmim+] are rather similar irrespective of what [mmim+] is bound to, the C and N bond lengths and angles were frozen. Hydrogen atoms were allowed to rotate freely around their C atoms but at a fixed distance. The BF bond lengths do change, depending on the coordination of [BF4], but it is not known a
priori when which bond should be longer or shorter. We have therefore opted to freeze [BF4] as if it were a charge-neutral, perfectly symmetrical, tetrahedral BF4 molecule in vacuum. Simulation temperatures ranged from well below (100 K) to well above (1000 K) the experimental bulk melting temperature for [mmim][BF4] of 377 K.10 Due to the limited number of freely moving particles in the system, the concept of temperature should not be taken too strict for our simulations. In cases where the entire substrate was frozen, then the only moving parts are the [mmim+] and [BF4] ions with the former having a few rotational degrees of freedom for the H atoms. In cases where the atoms in the two upper monolayers were allowed to move, there were 128 more moving atoms, but the transfer of energy between the Al atoms and [mmim][BF4] is not always very efficient. Ten MD simulations were done with one, two, or four [mmim][BF4] 14726
dx.doi.org/10.1021/jp200401h |J. Phys. Chem. C 2011, 115, 14718–14730
The Journal of Physical Chemistry C
ARTICLE
Figure 10. [mmim][BF4] pairs on Al surfaces with some [mmim+] ions detaching from the Al surface, remaining connected only through [BF4] ions. Left: single pair. Right: four pairs.
pairs, at temperatures ranging from 100 to 1000 K. The combined total simulation time of these runs was 130 ps. The combined simulation time and the numbers of ion pairs are insufficient to obtain statistically sound numbers. Hence, we will only give a qualitative description of some noteworthy observations. All H atoms in [mmim+] ions are allowed to rotate around their C atoms, but not all do so. The single H atoms attached to atoms C2, C4, and C5 were practically always observed roughly in-plane with the imidazolium ring. The three H atoms on atoms C6 and C7 were often observed rotating. The distance between the three H atoms never changed much: the three rotate around their C atoms as a mostly fixed unit. The Al surface area is large enough for one to four [mmim][BF4] pairs not to exceed monolayer coverage. However, somewhat contrary to our expectations, the simulations did not show all ions moving only two-dimensionally over the surface. In accordance with the strong [BF4]surface interaction found in relaxations, [BF4] ions did stick to the surface and moved only twodimensionally if they were initially placed on it. If they were initially placed some distance above the surface, they quickly found their way to it and then moved over it mostly twodimensionally. An exception was when during a 1000 K simulation a [mmim][BF4] pair desorbed altogether (in an experiment thermal decomposition would likely have occurred already before evaporation52). By contrast to the close [BF4]surface proximity, [mmim+] ions did not remain firmly fixed to the surface the whole time. While they spent part of the time in positions close to the relaxed ones (i.e., imidazolium rings parallel and close to the surface), they also often occupied positions where they were under an angle to the surface. The [mmim+] ions could even completely detach from the Al surface and remain attached only to [BF4] ions (see Figure 10). With [BF4] attaching firmly to the surface most of the time and [mmim+] ions often detaching from it, there is on average a considerable enrichment of [BF4] at the surface. Several experimental papers have reported cationanion layering when ILs were deposited on surfaces of Si(111),15 mica, amorphous silica, and oxidized Si(110)16 and charged sapphire.14 The enrichment of [BF4] on Al in MD simulations suggests that there may be a degree of layering in [mmim][BF4] on Al(111) too, with the first layer consisting primarily of [BF4] ions. Confirming the earlier picture that the ions are relatively indifferent to what they bond with, [mmim+] ions could not only detach from the Al surface while remaining close to [BF4] ions but also detach from the [BF4] ions and move over the surface. [mmim+] ions were mostly found in close vicinity to [BF4] ions, but in simulations with multiple ionic pairs it was sometimes observed that a [mmim+] ion would detach from its nearest [BF4] ion and move over to another [BF4] ion. In a simulation with only one [mmim][BF4] pair, the ions split up, and the [mmim+]
ion moved across the periodic boundary to link up with the periodic image [BF4] ion. At submonolayer coverage, [mmim+] ions are more mobile than [BF4] ions. To study the behavior of [mmim][BF4] on Al at greater than submonolayer coverage, a simulation was run with 12 ionic pairs. These 12 pairs occupied much of the open space between the periodic Al slabs (see starting configuration in Figure 11). The simulation was run for 19 ps with the temperature fluctuating around 400 K. Atoms in the two top Al monolayers were allowed to move. There is a considerable amount of open space between the ionic pairs in the initial configuration. We had expected that this open space would disappear from between the ions and that these would form a dense liquid film covering the Al, thereby increasing the distance between the highest situated ions and the next periodic image of the Al slab. We had also expected the film to show layering. However, the system evolved in a rather different way. The ions did form into a denser substance, but this substance did not cover the entire Al surface. Instead, the IL appeared to dewet from the Al surface, leaving part of its surface area exposed. The dewetting went so far that at some point in the simulation the IL had formed a tiny droplet between the periodic Al slabs. In that configuration only a few ions were in close contact with either of the two Al surfaces (see Figure 12). This confirms the observation from earlier relaxation results that the interaction between ions is a bit stronger than the interaction between ions and the Al surface. Once it had formed, the droplet did not remain as compact as shown in Figure 12 throughout the simulation. However, even when some open volume appeared between the ions and the droplet disintegrated to some degree, there was no strong reattachment of ions to the surfaces, and interaction of ions remained mostly with other ions. 3.6. Splitting off HF? Experimentally, it has been shown that the H atom connected to atom C2 is acidic and can detach relatively easily53 at temperatures above ∼400 K. For imidazoliumhexafluorophosphate-based ILs it has been shown that the H atom attached to atom C2 can combine with a F atom to form molecular HF.54 A similar decomposition reaction could be suggested to occur for imidazolium tetrafluoroborate based ILs. In our ab initio MD simulations, all H atoms were constrained to be at a constant distance from their C atoms, and hence any potential HF formation was artificially disabled. To see if the CH bond length constraints prevented any easy HF formation, we performed a number of relaxations to determine energy differences between different states. We looked at HF formation not within an IL but rather when H and F split off from separate mmim and BF4 radicals or from just a single ionic pair. Due to the somewhat unnatural, unstable state of the radicals, the process of splitting off HF and forming BF3 and “H-amputated” mmim 14727
dx.doi.org/10.1021/jp200401h |J. Phys. Chem. C 2011, 115, 14718–14730
The Journal of Physical Chemistry C
ARTICLE
Figure 11. Starting configuration for a MD run of 12 [mmim][BF4] pairs on Al(111). Left: side view, including periodic image of the Al slab. Right: top view.
Figure 12. Configuration from MD run of 12 [mmim][BF4] pairs on periodic Al(111) slabs. Left: top view showing dewetting of part of the Al surface. Right: side view showing the formation of a tiny [mmim][BF4] droplet between periodic Al slabs.
(denoted as mmim-H from here on) is highly exothermic, as it lowers the total energy by 3.54 eV. A considerable part of this lower energy is due to the relaxation of BF3 into a planar molecule. Removing an F atom from a BF4 radical and then letting the remaining BF3 molecule relax into its trigonal planar ground state (with BF bond lengths of 1.33 Å, according to DFT) lowers the energy by 1.74 eV. By contrast, relaxing mmim-H after the H atom has
been removed from atom C2 does not change the overall shape of the imidazole ring and side groups very much, and the relaxation only lowers the system energy by 0.03 eV. While it would be energetically favorable for HF to split off from mmim and BF4 radicals, we recall here from Section 3.1 that forming an [mmim][BF4] ionic pair from these radicals in the inplane or stacked configurations lowers the energy by 5.60 or 14728
dx.doi.org/10.1021/jp200401h |J. Phys. Chem. C 2011, 115, 14718–14730
The Journal of Physical Chemistry C
Figure 13. HF molecule that has been artificially split off from a stacked [mmim][BF4] pair. The HF pair is held at equilibrium bond length, and only the three F atoms near the B atom are allowed to relax. Despite HF being held at equilibrium distance, BF3 being allowed to form a planar molecule, and the FB and HC2 distances being 2.78 and 2.42 Å, HF is not in a stable minimum, and once atoms are allowed to move, HF splits up and the system relaxes back to the stacked [mmim][BF4] pair. In the configuration shown, the energy is 2.17 eV higher than in the fully relaxed stacked configuration.
6.09 eV. The charge transfer that is primarily responsible for this lowering of the energy has to be undone for HF to form, since HF, BF3, and mmim-H are not charged ions. This means that forming HF from an ion pair is highly endothermic. Indeed, when we forced the H and F atoms away from atoms C2 and B in a stacked pair and brought them together at the HF equilibrium distance (which turned out to be 0.94 Å for isolated HF, according to DFT) while keeping all other atoms locked in place, the H and F atoms relaxed back to form [mmim+] and [BF4] ions and restore the ionic pair. There did not appear to be a local minimum in which HF was stable. As the HF pair was artificially placed further and further from the BF3 and mmim-H molecules (that were locked in place), the energy increased asymptotically and monotonously. This was true even when only the B atom was held fixed and the three F atoms were allowed to form a mostly planar BF3 molecule that oriented itself more or less perpendicular to the F atom in HF (see Figure 13). It will be clear from the previous results that forming HF from the H atom attached to atom C2 is energetically very unfavorable. It should be mentioned here that the configuration in Figure 13 consists of molecules in vacuum, whereas on a surface the resulting mmim-H, BF3, and HF could have absorbed and thereby stabilized the system, making the splitting off of HF a little easier. Still, in light of the very endothermic nature of HF formation, any experimentally observed HF formation is more likely due to hydrolysis53,55 of BF4 ions with water contamination in the IL. The constraints imposed in our MD simulations would probably not have prevented any natural occurrences of HF formation.
4. CONCLUSIONS We have performed DFT relaxations and ab initio molecular dynamics calculations to study the interaction between the ionic liquid 1,3-dimethylimidazolium tetrafluoroborate ([mmim][BF4]) and Al and Cu(111) surfaces. Using the detailed analysis that
ARTICLE
such atomistic simulation allows, we have looked at adsorption energies, electron transfer, and bond length changes of fully relaxed structures to obtain a general picture of how [mmim+] and [BF4] ions interact with Al and Cu surfaces and each other. Many of the results for submonolayer coverage can be explained from a simple bond saturation picture, in which the ions exchange electron density with either other ions or the surface atoms, having some preference for the former over the latter. Given the generic nature of this picture, it is not surprising that many results are qualitatively similar for Al and Cu surfaces. An exception to this similarity is the strength of the [BF4]surface interaction, which is clearly stronger for Al than for Cu. As a result, moving an ion pair across an Al surface takes more energy than moving it across a Cu surface, though for both Al and Cu the energy needed is quite low. MD simulations on Al surfaces show the limited usefulness of relaxations for investigating the behavior of ILs on surfaces. While fully relaxed configurations of a few [mmim][BF4] pairs show both [mmim+] and [BF4] ions sticking closely to the surface, only [BF4] ions remain close to the surface during MD simulations of submonolayer coverage, even at low temperature. [mmim+] ions can assume positions some distance from the surface, sometimes held to the surface only by [BF4] ions between the surface and the [mmim+] ion. Since [BF4] ions stick firmly to the surface and [mmim+] are located some distance away from it much of the time, there is on average an enrichment of [BF4] ions at the surface. This would seem to suggest a degree of anioncation layering at the surface for very low coverage. However, MD simulations with more ionic pairs did not provide additional evidence that this is indeed the case. Rather than forming anioncation layers, the 12 ionic pairs in our largest simulation dewetted from the surface, forming a tiny droplet between the periodic Al slabs. This result shows that attraction between the ions is stronger than between the ions and the Al surface. This result would have been impossible to obtain from relaxations or MD runs with only 14 ionic pairs. The droplet formation in the simulation with 12 ionic pairs does not necessarily undo the prediction of layering. The tiny droplet formed in our simulation has an extremely high surface/volume ratio (forming a strong external influence that may not be so important for larger droplets), and the constraints imposed in the simulation may inhibit the adsorption of ions somewhat. Another small factor may be that in a real system a larger droplet would be subject to gravity. In a real experiment, a larger drop of [mmim][BF4] on an Al surface would have to form an ILsurface interface somehow. While ILIL interaction is stronger than ILsurface interaction, it is still possible that in a real experiment some layering might still occur at the interface. This means that the simulation results allow us to make some predictions that are open to experimental verification: [mmim][BF4] should dewet from an unoxidized Al(111) surface, and at the contact surface of any drops of [mmim][BF4] on Al the IL may show some degree of layering, with the first layer containing more [BF4] ions. Looking at the results obtained in the sequence of calculations of relaxations w MD with 14 ionic pairs w MD with 12 ionic pairs, each subsequent step has shown results impossible to obtain in the previous step(s) and has even produced results that point in a different direction than the previous steps (e.g., the dewetting observed in the 12 pairs MD simulation vs adhesion seen for one or two relaxed pairs). This shows that MD simulations with considerable numbers of ionic pairs are the minimum requirement for understanding some aspects of the behavior of ILs on 14729
dx.doi.org/10.1021/jp200401h |J. Phys. Chem. C 2011, 115, 14718–14730
The Journal of Physical Chemistry C surfaces. Given that our simulations only went up to 12 ionic pairs on a 22.87 19.80 Å2 surface, more could probably be discovered by simply scaling to larger systems (presently beyond our computational capabilities).
’ AUTHOR INFORMATION Corresponding Author
*E-mail:
[email protected].
’ ACKNOWLEDGMENT We thank the Delft Centre for Materials (DCMat) for support and acknowledge support through grant 02EMM32 co-funded by M2i (www.m2i.nl) and the Foundation for Fundamental Research on Matter (FOM) of The Netherlands. Help from Tomas Bucko for carrying out some of the calculations is warmly acknowledged, as are the computational resources made available by Peter Bloom and reading of the manuscript by Tristan Youngs. ’ REFERENCES (1) Lin, L. G.; Wang, Y.; Yan, J. W.; Yuan, Y. Z.; Xiang, J.; Mao, B. W. Electrochem. Commun. 2003, 5, 995–999. (2) Borisenko, N.; Zein El Abedin, S.; Endres, F. J. Phys. Chem. B 2006, 110, 6250–6256. (3) Nanbu, N.; Sasaki, Y.; Kitamura, F. Electrochem. Commun. 2003, 5, 383–387. (4) Mu, Z.; Zhou, F.; Zhang, S.; Liang, Y.; Liu, W. Tribol. Int. 2005, 38, 725–731. (5) Liu, W; Ye, C.; Gong, Q.; Wang, H.; Wang, P. Tribol. Lett. 2002, 13, 81–85. (6) Liu, X.; Zhou, F.; Liang, Y.; Liu, W. Wear 2006, 261, 1174–1179. (7) Romero, C.; Baldelli, S. J. Phys. Chem. B 2006, 110, 6213–6223. (8) Fitchett, B. D.; Conboy, J. C. J. Phys. Chem. B 2004, 108, 20255–20262. (9) Rollins, J. B.; Fitchett, B. D.; Conboy, J. C. J. Phys. Chem. B 2007, 111, 4990–4999. (10) Holbrey, J. D.; Seddon, K. R. J. Chem. Soc., Dalton Trans. 1999, 2133–2139. (11) Consorti, C. S.; Suarez, P. A. Z.; et al. J. Phys. Chem. B 2005, 109, 4341–4349. (12) Bagno, A.; D’Amico, F.; Saielli, G. J. Mol. Liq. 2007, 131132, 17–23. (13) Triolo, A.; Russina, O.; Bleif, H. J.; Di Cola, E. J. Phys. Chem. B 2007, 111, 4641–4644. (14) Mezger, M.; Schr€oder, H.; Reichert, H.; Schramm, S.; et al. Science 2008, 322, 424–428. (15) Carmichael, A.; Hardacre, C.; Holbrey, J. D.; Nieuwenhuyzen, M.; Seddon, K. R. Mol. Phys. 2001, 99, 795–800. (16) Bovio, S.; Podesta, A.; Lenardi, C.; Milani, P. J. Phys. Chem. B 2009, 113, 6600–6603. (17) Bovio, S.; Podesta, A.; Milani, P.; Ballone, P.; Del Popolo, M. G. J. Phys.: Condens. Matter 2009, 21, 424118. (18) Aliaga, C.; Santos, C. S.; Baldelli, S. Phys. Chem. Chem. Phys. 2007, 9, 3683–3700. (19) Dupont, J.; Consorti, C. S.; Spencer, J. J. Braz. Chem. Soc. 2000, 11, 337–344. (20) LyndenBell, R. M.; Del Popolo, M. G.; Youngs, T. G. A.; Kohanoff, J.; Hanke, C. G.; Harper, J. B.; Pinilla, C. C. Acc. Chem. Res. 2007, 40, 1138–1145. (21) Hanke, C. G.; Price, S. L.; LyndenBell, R. M. Mol. Phys. 2001, 99, 801–809. (22) Liu, L.; Li, S.; Cao, Z.; Peng, Y.; Li, G.; Yan, T.; Gao, X. P. J. Phys. Chem. C 2007, 111, 12161–12164. (23) Maolin, S.; Fuchun, Z.; Guozhong, W.; Haiping, F.; Chunlei, W.; Shimou, C.; Yi, Z.; Jun, H. J. Chem. Phys. 2008, 128, 134504.
ARTICLE
(24) Sieffert, N.; Wipff, G. J. Phys. Chem. C 2008, 112, 19590–19603. (25) Yan, T.; Burnham, C. J.; Del Popolo, M. G.; Voth, G. A. J. Phys. Chem. B 2004, 108, 11877–11881. (26) Yan, T.; Li, S.; Jiang, W.; Gao, X.; Xiang, B.; Voth, G. A. J. Phys. Chem. B 2006, 110, 1800–1806. (27) Hunt, P. A. Mol. Simul. 2006, 32, 1–10. (28) de Andrade, J.; B€oes, E. S.; Stassen, H. J. Phys. Chem. B 2002, 106, 3546–3548. (29) Del P opolo, M. G.; Kohanoff, J.; LyndenBell, R. M.; Pinilla, C. Acc. Chem. Res. 2007, 40, 1156–1164. (30) Del Popolo, M. G.; LyndenBell, R. M.; Kohanoff, J. J. Phys. Chem. B 2005, 109, 5895–5902. (31) Turner, E. A.; Pye, C. C.; Singer, R. D. J. Phys. Chem. A 2003, 107, 2277–2288. (32) Bhargava, B. L.; Balasubramanian, S. Chem. Phys. Lett. 2006, 417, 486–491. (33) B€uhl, M.; Chaumont, A.; Schurhammer, R.; Wipff, G. J. Phys. Chem. B 2005, 109, 18591–18599. (34) Price, D. L.; Halley, J. W. J. Chem. Phys. 1995, 102, 6603–6612. (35) Vassilev, P.; van Santen, R. A.; Koper, M. T. M. J. Chem. Phys. 2005, 122, 054701. (36) Mattsson, T. R.; Paddison, S. J. Surf. Sci. 2003, 544, L697–L702. (37) Izvekov, S.; Mazzolo, A.; VanOpdorp, K.; Voth, G. A. J. Chem. Phys. 2001, 114, 3248–3257. (38) Lanzani, G.; Martinazzo, R.; Materzanini, G.; Pino, I; Tantardini, G. F. Theor. Chem. Acc. 2007, 117, 805–825. (39) Hass, K. C.; Schneider, W. F.; Curioni, A.; Andreoni, W. Science 1998, 282, 265–268. (40) Janik, M. J.; Neurock, M. Electrochim. Acta 2007, 52, 5517–5528. (41) Taylor, C.; Kelly, R. G.; Neurock, M. J. Electrochem. Soc. 2006, 153, E207–E214. (42) Taylor, C. D.; Kelly, R. G.; Neurock, M. J. Electrochem. Soc. 2007, 154, F217–F221. (43) Valencia, H.; Kohyama, M.; Tanaka, S.; Matsumoto, H. Phys. Rev. B 2008, 78, 205402. (44) Valencia, H.; Kohyama, M.; Tanaka, S.; Matsumoto, H. J. Chem. Phys. 2009, 131, 244705. (45) McNellis, E. R.; Meyer, J.; Reuter, K. Phys. Rev. B 2009, 80, 205414. (46) Mercurio, G.; McNellis, E. R.; Martin, I.; et al. Phys. Rev. Lett. 2010, 104, 036102. (47) Kresse, G.; Hafner, J. Phys. Rev. B 1993, 47, 558–61. (48) Kresse, G.; Furthmuller, J. Phys. Rev. B 1996, 54, 11169. (49) Bl€ochl, P. E. Phys. Rev. B 1994, 50, 17953. (50) Kresse, G.; Joubert, D. Phys. Rev. B 1999, 59, 1758. (51) Yoshitake, M.; Karas, I.; Houfek, J.; Madeswaran, S.; Song, W.; Matolín, V. J. Vac. Sci. Technol., A 2010, 28, 152. (52) Kroon, M. C.; Buijs, W.; Peters, C. J.; Witkamp, G. J. Thermochim. Acta 2007, 465, 40–47. (53) Welton, T.; Wasserscheid, P. In Ionic Liquids in Synthesis; Wiley VCH: Weinheim, 2008. (54) Shen, H. Y; Judeh, Z. M. A.; Ching, C. B. Tetrahedron Lett. 2003, 44, 981. (55) Banyai, I.; Conte, V.; Pettersson, L.; Silvagni, A. Eur. J. Inorg. Chem. 2008, 34, 5373.
14730
dx.doi.org/10.1021/jp200401h |J. Phys. Chem. C 2011, 115, 14718–14730