† Corresponding author. E-mail:
The experimental knowledge on interlayer potential of graphitic materials is summarized and compared with the computational results based on phenomenological models. Besides Lennard–Jones approximation, the Mie potential is discussed, as well as the Kolmogorov–Crespy model and equation of Lebedeva et al. An agreement is found between a set of reported physical properties of graphite (layer binding energies, compressibility along c-axis in a broad pressure range, Raman frequencies for bulk shear and breathing modes under pressure), when a proper choice of model parameters is taken. It is argued that anisotropic potentials, Kolmogorov–Crespy and Lebedeva, are preferable for modeling, as they provide a better, self-consistent description. A method of fast numerical modeling, convenient for the accurate estimation of the discussed physical properties, is proposed. It may be useful in studies of other van der Waals homo/heterostructures as well.
There is recently a growing interest in artificial heterostructures formed by 2-dimensional (2D) layers of graphene and other newly discovered materials, and named van der Waals heterostructures.[1] This is due to richness of physical phenomena observed and high potential of their possible applications.[2] Moiré patterns result when two layers are rotated[3,4] and are related to the formation of van Hove singularities in the density of electronic states. In the bilayer graphene twisted at about 1.06°, an unconventional superconductivity was found with a critical temperature of 1.7 K.[5–8] By manipulating the doping levels, the van Hove singularities were observed at angles up to 31°.[9] It was shown that free-standing vdW heterostructure of graphene on hexagonal boron nitride (h-BN), where a small lattice mismatch exists, shows a buckled atomic structure formed as Moiré pattern.[10] There is already a number of laboratory prototypes of optoelectronic devices,[11] like bolometers[12] and photodetectors.[13,14] Their characteristics can be tuned by elemental doping, surface chemical doping, intercalation, electrostatic gating, strain[15] and by twisting the layers.[16]
It is believed that artificial vdW heterostructures offer a broad field for research on novel materials, such as graphene/silicene, graphene/MoS2, and silicene/MoS2 systems,[2] as well as silicene[17] or MoS2 alone.[18] Ab-initio calculations suggest the existence of magnetism in ReS2 doped with Co, Fe, and Ni.[19] The formation of quantum-well type-I heterostructure in van der Waals interacting monolayers of MoS2 and ReS2 has been demonstrated[20] and type-II band alignment was found in vdW heterojunction diodes based on InSe, GaS, and graphene.[21] WSe2 and MoS2 monolayers have been shown useful to create charge density modulation in electronic band valleys, resulting in valley polarization. Na-doped WS2 under strain was considered as a candidate for spintronics.[22] After spin injection from a ferromagnetic electrode, the transport of spin-polarized holes within the WSe2 layer was observed.[23] The h-BN and graphene vdW heterojunctions showed large potential of use in spintronics.[24] Light-induced negative differential transconductance phenomenon was realized on the graphene/WSe2 heterojunction transistor.[25] Density functional theory (DFT) calculations show that vdW interactions dominate between antimonene and graphene layers and by applying an electric field between them, one can tune the height and type of the Schottky contacts.[26] Lattice dynamical theory and ab-initio calculations indicate the existence of piezoelectricity in 2D lattices comprising h-BN, 2H-MoS2, and other transition-metal dichalcogenides.[27]
It is generally thought that van der Waals interaction, ubiquitous in nature, is caused by temporal fluctuations of electronic charge that induces dipoles of random amplitude.[28] It plays a role in friction and adhesion between materials, and in absorption of molecules on surfaces. These relatively weak forces are responsible for binding between separate sheets of graphene.
While many electronic properties of graphene-based systems are known,[29] still there are open questions concerning the detailed form and physical mechanisms leading to the inter-plane potential in these compounds. There are two types of approaches used to describe the van der Waals potentials of graphitic systems: the ab-initio one based on density functional theory and the other one based on the use of empirical potentials.[30]
The DFT calculations often give underestimated values of binding energy in case of vdW interactions [31–35] and the energy difference between bindings of (preferred) AB and AA types of stacking, EAB and EAA, seems too large. Recent diffusion quantum Monte Carlo calculations predict for instance EAA/EAB of about 0.65.[36] The results however depend strongly on the functionals used to model the vdW forces.[37,38]
In this work, we compare how well several properties of graphene/graphite can be described by using a few empirical potentials. Some of them are well based on DFT calculations. An advantage of using empirical potentials relies on ease of their implementation in numerical methods. We describe a scheme of computing several basic properties based on a simple analytical approach. The method may be easily extended to the investigation of vdW homo- and heterostructures other than graphene/graphite. We demonstrate that by careful choosing the parameters of the phenomenological potentials, we may reproduce self-consistently experimental values of several physical quantities at the same time: layer binding energy, compressibility, Raman shifts observed in shear and layer breathing modes under pressure. By using our procedure, we are able to determine well how the potential interaction of atoms changes with distance between them. Knowing that is important in modeling of a range of mechanical and electronic phenomena as well.
An often used approximation of vdW potential is the 12–6 Lennard–Jones one (LJ),
It reproduces well the interlayer distance and elastic constant of graphite at zero pressure. It was proved useful in describing several basic properties of C60 molecules[39] with ϵ=2.964 meV and σ=3.407 Å. It was used for modeling carbon nanotubes[40] with ϵ=4.656 meV and σ=3.825 Å. He et al.[41] used it to obtain an analytical approximation for vdW forces acting between nanotubes. Kitipornchai et al. extended their continuum model[42] to study vibrations of multilayered graphene sheets. It is a convenient approximation in modeling reinforcement of composite materials with carbon nanotubes.[43]
A more general form of vdW potential is the one introduced by Mie.[44,45] It often reproduces well the properties of carbon compounds:
There are two deficiencies of LJ or Mie type of potentials. The first one is that they are isotropic, i.e., the potential depends only on the distance between atoms. However, in graphitic materials, the binding between atoms on different planes is due to overlap between π orbitals from adjacent layers, as such it ought to depend on the angle between them. The second problem, as it was noticed first by Kolmogorov and Crespi (KC),[46,47] is that these potentials produce too small energy difference between bindings of AB (preferred) and AA types of stacking, EAB and EAA. Albeit there is some disagreement in literature regarding how large that energy difference is. The KC potential has an
The KC potential has been used broadly in numerical modeling (molecular dynamics)[3,4,49,50] as well in description of ballistic nanofriction.[51,52]
Lebedeva et al.,[53,54] used another analytic function of anisotropic potential which was obtained by approximating the results of DFT calculations and was found to describe well the experimental graphite compressibility and the corrugation against sliding, while Jiang[55] used the following function for modeling the thermal conductivity of FLG:
Lattice spacing in graphite in hexagonal c-direction is known well from x-ray diffraction, it is 3.3538 Å at room temperature, and it does not change strongly with temperature.[56] The equilibrium interlayer spacing is 3.34 Å in AB stacked graphene[56] and 3.44 Å in turbostratic graphene.[57] Neutron diffraction studies show that all in-plane C–C lengths are 1.422 ± 0.001 Å.[58]
There are two types of ordering of atoms on the nearest plane with respect to atoms on one plane. We will call them type I and type II orderings. The type I results when the atom is placed over another atom of the hexagon and type II is found when the atom is placed over the center of hexagon. Type I ordering is the only one present in case of AA stacking, and in case of AB (Bernal) stacking, equal numbers of atoms are in orderings of type I and type II.
We observe that atoms of both types of ordering can be grouped in rings as shown in Fig.
Our aim is to find out radii of these rings and the number of atoms in any ring.
We can create a convenient scheme of calculating numerically the potential energy for different stackings by computing it for each of these types of orderings. For that, we need distances projection (in plane) between an atom position over the plane together with the number of atoms at these distances (rings of equidistant atoms, as these in Fig.
Positions of carbon atoms in a graphene lattice may be generated with the following equation:
Accordingly, the plane projections of distances from an atom located at (x,y)=(0,0) are given by
We do not know how to find out the number of atoms in any ring of equidistant atoms by using an analytical formula. These numbers can however be found numerically in a few ways by using a proper algorithm, as will be described in a separate publication. The results are shown in Table
In the numerical simulations, e.g., performed in LAMMPS,[59] it is a standard procedure to introduce a cut-off distance in equations on potential energy assuming it equal to zero when the distance between particles exceeds that value, while the function given by these equations is appropriately smoothed out in order to avoid possible discontinuities in computational results. That cut-off is taken usually at around 12 Å in case of graphene modeling. Limiting any approximation to atoms listed in Table
Potential energy
In case of AB planes, there is equal number of atoms that are in orderings of type I and type II. The equation on energy for type II atoms,
Figure
Hence, an average potential energy for an atom at distance z from the surface of bulk graphite is given as a sum; for AA, AB, and ABC stackings, it is respectively
The proposed method of computing interlayer potential is convenient for quick testing of potentials other than those of Mie or Lennard–Jones, since this requires only replacement of the function U(z) in Eq. (
In case of perfect AA or AB stacking, due to the symmetry, the π-orbitals must point out in directions normal to the planes. In this case, in Kolmogorov–Crespi equations (
In case of Lennard–Jones potential, the results presented here are computed with ϵ=2.4 meV and σ=3.4322 Å. For the models of Kolmogorov–Crespi and Lebedeva et al., we used values of parameters listed in Tables
Equation (
Figure
The pressure dependence of c can be found or deduced from several measurements.[58,61–63,65] The experimental results in Fig.
There is much controversy around prevailing stacking order in graphite and in a few layers graphene (FLG). It is known that there are three types of stacking, an AA one, which is simple hexagonal (it is not found in natural graphite and exists only in intercalated compounds such as
Lee et al.[66] were able to obtain AA structured graphene on diamond surface, with an interplanar spacing of ∼3.55 Å. The value is between those of the AB graphite (3.35 Å) and the lithium intercalated AA graphite 3.706 Å.[67] Norimatsu and Kusunoki[68] observed ABC-stacked graphene on the SiC substrate and interpreted their results in terms of possible modification of second-plane interactions by the substrate, as explained by the Slonczewski–Weiss–McClure model.[69] Yoshizawa et al.[70] proposed that the AB stacking of layers in graphite is a consequence of orbital interactions between layers rather than that of generally accepted van der Waals forces.
The energy difference between AA and AB was reported to be 17.31 meV/atom,[36] in favor of AB, while between ABC and AB it was reported as 0.11 meV/atom,[71] which are both based however on DFT calculations only.
It was observed that in the case of graphene, its bilayer exhibits AB stacking while trilayer prefers ABC stacking.[72] When the number of layers increases, again AB stacking is favored. The binding energies are found to increase from 23 meV/atom in bilayer to 39 meV/atom in pentalayer, while the interlayer distance decreases from 3.37 Å in bilayer to 3.35 Å in pentalayer. Our models do not reproduce so strong change of binding energy with number of layers. In 2-layer graphene, we find around 90% of binding energy per atom of that in bulk graphite when the LJ model is used (Fig.
If we assume that the interaction between next nearest planes (NNP) is negligible, then there is no reason for difference in stability of ABC and AB structures. Hence the NNP interaction must play a role.
The explanation to the AB:ABC ratio observed of about 80:14[74] could be found by adding to already existing isotropic attractive term of vdW potential a small anisotropic contribution to equations like those of KC or Lebedeva. Since interplane binding is caused by the strongly anisotropic interaction between π orbitals, one would expect that not only repulsive but also attractive component of that interaction is anisotropic.
Early neutron scattering measurements confirmed the existence of the low-frequency acoustic mode in bulk pyrolitic graphite,[75] that is, the longitudinal waves in the direction of the hexagonal axis, at 3.84 ± 0.06 THz. The transverse waves (in shear mode) were less pronounced, at 1.3 ± 0.3 THz. Based on these data, the elastic constants have being deduced, C33=39±4 GPa (for direction perpendicular to the graphite planes) and C44=4.2±2 GPa (shear mode parallel to the planes). These values correspond to Raman shift energies of 130 cm−1 and 43 cm−1, in agreement with several DFT calculations.[76] The experimental evidence of the longitudinal mode was reported in ultrafast laser pump–probe spectroscopy[77] on FLG at ∼120 cm−1, demonstrating also the presence of the phonon and electron coupling there through the Breit–Wigner–Fano resonance, as at more pronounced G-mode.[78]
Position of peaks in both breathing and shear modes is a geometrical effect and it depends strongly on the number of layers in FLG. Moreover, unique multipeak features are observed, characteristic for the number of layers investigated. Their frequencies in case of breathing mode[79] are described well using a simple linear-chain model based on nearest-neighbor couplings between the layers,[80]
For the shear mode, the nature of interactions between planes, their collective behavior, leads to quantitatively similar dependence of Raman frequencies on the number of layers in FLG as in the case of breathing mode: the frequency in the bulk is
In Fig.
By having the curvature of the parabolic potential wells at low-values of departure from the optimal position of both layers (i.e., from AB stacking position; we used data for x less then 0.05 Å), we are able to compute Raman frequencies in shear mode in a broad range of pressure. The curvature k of the parabolic fit of data in Fig.
The results on Raman shift are shown in Fig.
For the breathing mode, the results on Raman shift are only known at P = 0 GPa and our calculations reproduce that value well.[79] The scheme described in this section offers a convenient way of numerical computing the frequency of oscillations as a function of pressure, by starting with Eq. (
Modeling of interlayer interaction in graphitic materials was performed based on phenomenological van der Waals potentials (Lennard–Jones, Kolmogorov–Crespi, and Lebedeva’s). The results have been compared self-consistently with existing ab-initio calculations and experimental data on compressibility and Raman shifts in breathing and shear modes under pressure, favoring strongly the anisotropic models of potential. Computation was done by using molecular dynamics package LAMMPS and a proposed convenient, extendable scheme suitable for fast numerical modeling of several physical quantities. The method is useful for studying other 2-dimensional homo- and heterostructures with van der Waals type interaction between layers. In case of homostructures, the application of the proposed method is straightforward. In case of twisted homostructures or heterostructures, such as h-BN/graphene, some approximations would need to be introduced due to the lattice mismatch between components. The proposed method could possibly be extended to studies of other lattices, different than hexagonal and to other physical phenomena where interactions decaying quickly with distance are involved.
1 | |
2 | |
3 | |
4 | |
5 | |
6 | |
7 | |
8 | |
9 | |
10 | |
11 | |
12 | |
13 | |
14 | |
15 | |
16 | |
17 | |
18 | |
19 | |
20 | |
21 | |
22 | |
23 | |
24 | |
25 | |
26 | |
27 | |
28 | |
29 | |
30 | |
31 | |
32 | |
33 | |
34 | |
35 | |
36 | |
37 | |
38 | |
39 | |
40 | |
41 | |
42 | |
43 | |
44 | |
45 | |
46 | |
47 | |
48 | |
49 | |
50 | |
51 | |
52 | |
53 | |
54 | |
55 | |
56 | |
57 | |
58 | |
59 | |
60 | |
61 | |
62 | |
63 | |
64 | |
65 | |
66 | |
67 | |
68 | |
69 | |
70 | |
71 | |
72 | |
73 | |
74 | |
75 | |
76 | |
77 | |
78 | |
79 | |
80 | |
81 | |
82 |