Jesús
Carrete
*a,
Vu
Ngoc Tuoc
b and
Georg K. H.
Madsen
a
aInstitute of Materials Chemistry, TU Wien, A-1060 Vienna, Austria. E-mail: jesus.carrete.montana@tuwien.ac.at
bInstitute of Engineering Physics, Hanoi University of Science and Technology, 1 Dai Co Viet Road, Hanoi 100000, Vietnam
First published on 6th February 2019
We propose a convenient method to characterize the acoustic phonon branches of 2D monolayer materials using measurements of the infrared- and Raman-active vibrational modes of nanotubes. The relations we employ are derived from a symmetry analysis based directly on the line groups of nanotubes. We perform extensive ab initio calculations for the MoS2 monolayer and nanotubes to evaluate the method and illustrate all our results. Specifically, we show how the low-energy phonon transmission, a determining factor in thermal transport, can be easily and successfully reconstructed by this procedure.
Among non-elemental compounds, transition-metal dichalcogenides (TMDs) have attracted significant attention. TMDs follow the formula TX2, where T is a transition metal from groups IV–VI (e.g. Mo, Ti, Nb), and X is a chalcogen (S, Se, or Te). They crystallize in several different bulk phases, all of which consist of covalently bonded layers held together by weaker interactions. This fact opens the door to exfoliation to obtain large-area single layers.1 In particular, monolayer (ML) TMDs have been extensively studied as direct-band-gap semiconductors.2 Even long before the rediscovery of graphene, early observations of the polyhedral and cylindrical structures of WSe2 led to the proposal that it could be unstable in a planar structure and spontaneously assemble into a fullerene-like or NT form.3 MoS2 NTs were synthesized three years later.4
The properties of TMD NTs depend on their diameter and chirality. Computational studies have shown that zigzag (ZZ) MoS2 NTs, like the corresponding ML, are direct-band-gap semiconductors, while armchair (AM) MoS2 NTs show an indirect gap like the bulk 2H phase.5 The gap width varies from almost 0 eV to 1.5 eV depending on the diameter and strain.6 The tunability has aroused interest for applications in nano- and optoelectronics, where absorption could be shifted from the near infrared (IR) range to the visible one. Further applications include valleytronics, as 2D field-effect transistors (FETs), logical circuits and amplifiers.7,8 The first actual device based on MoS2 NTs, with diameters of 20 nm to 100 nm and lengths 1 μm to 2 μm, is a FET with mobilities up to 0.014 cm2 V−1 s−1 and an on/off ratio9 of 60. General perspectives on the challenges and opportunities for a number of potential applications in the fields of lubrication, high-performance nanocomposites, sensors, renewable energy, energy storage, and catalysis are discussed in recent reviews.8,10–12
A central question for the characterization of 1D and 2D systems is their stability, and vibrational properties are one of the key elements determining it. Not only is a fully real phonon spectrum a requirement for mechanical stability, but the vibrational free energy can also alter the order of thermodynamic stability.13 Furthermore, in addition to their singular electronic features, 1D and 2D systems have attracted attention because of their thermal transport properties, especially in connection with the important problem of heat dissipation in nanodevices. Part of this appeal is due to the extremely high thermal conductivity of graphene14–16 and carbon NTs.17–20 As in the case of stability, characterizing the thermal transport properties of a material requires a detailed understanding of its phonon physics.
The phonon spectrum of the MoS2 ML has been characterized both using semi-empirical potentials21 and from first principles,22 with the latter article also calculating its thermal conductivity. The phonon spectra of MoS2 NTs have also been studied using both families of methods.6,23,24 However, a significant stumbling block for calculations comes from the particular features of the phonon dispersions of 1D and 2D systems. Electronic structure programs typically fall into one of two categories: those designed to work on crystals, with periodic boundary conditions along every axis, and those that focus on aperiodic systems like molecules. Regarding symmetry, the former typically implement space groups and the latter point groups. Therefore, none of them adequately supports the subperiodic groups required to deal with systems that are only periodic in one or two dimensions.
Translational symmetry guarantees that, for long enough wavelengths, phonons from any of the three acoustic branches in 3D periodic systems have a linear dispersion relation. In contrast, crystals that are only periodic along one spatial axis have four acoustic branches: two linear ones and two quadratic ones. Likewise, systems periodic along two axes have a single quadratic branch, corresponding to vibrations along the non-periodic direction, and its existence is guaranteed by rotational symmetry. Unfortunately, the periodic boundary conditions employed in electronic structure calculations introduce violations of this fundamental symmetry. Typical manifestations of the problem are spurious imaginary frequencies close to the Γ point.25–27 It is therefore important to correct such artifacts, which can have a significant effect on derived quantities like thermal conductivity or lead to a mischaracterization of the mechanical stability of a system. To overcome this issue, two very different approaches have been devised. A first possibility is to exploit the fact that continuum theory can provide the long-wavelength limit of the acoustic modes, and to use electronic-structure software to estimate the elastic parameters defining those.26,27 A second approach consists in finding the maximum projection of the interatomic-force-constant (IFC) tensor on a subspace of internal coordinates that cannot express rigid translations or rotations, hence removing the artifactual elements introduced by periodic boundary conditions and recovering the phonon dispersions that fulfill the physical constraints by construction.25
Here we apply a consistent, ab initio methodology to illustrate the connection between the phonon spectra of the MoS2 NTs and both the continuum limit and the spectrum of the corresponding ML. We analyze the symmetry of the NT phonon modes, both in terms of their chirality and their irreducible representations (irreps), to associate them to those of the ML. The agreement observed between the ML and large-radius NTs makes it reasonable to reconstruct the ML acoustic bands based on the phonon frequencies of the NTs at the Γ point. We exploit the connection between the phonon spectra of the NTs and that of a ML to show that IR and Raman measurements, which can only provide data about optical branches at the Γ point, on single-walled NTs can be used to extract key features of the phonon spectrum of the ML. Thereby information on the acoustic branches at q ≠ 0 can be obtained which would otherwise require data from more specialized measurement methods like inelastic neutron scattering.28 Finally, we calculate the thickness- and chirality-dependent ballistic phonon transmission and compare it to that of the ML.
The procedure for each system starts with an energetic minimization to find the equilibrium ionic positions. We use a 17 × 17 × 1 Monkhorst–Pack grid for the ML and a 4 × 1 × 1 grid for the NTs. We add 10 Å of empty space to the unit cell along each of the non-periodic directions to reduce spurious interactions between copies of the system, and we check that the results are well converged with respect to this parameter by repeating some selected calculations with 15 Å of padding. Once the minimum-energy coordinates are known, we use Phonopy34 to generate a symmetry-reduced set of displaced configurations for supercells with sizes 8 × 8 × 1 (for the ML) or 4 × 1 × 1 (for the NTs). After using VASP again to obtain the forces on atoms in each of those configurations, we rebuild the matrix of IFCs with Phonopy. In the case of NTs, we post-process that matrix to extract a minimal set of constants corresponding to the reduced motif discussed in Section 3.1. For the ML we use a force-constant projection scheme to symmetrize the force constants.25 In 1D it is straightforward to extrapolate the ZA branch quadratically towards the Γ point from points that are free from artifacts.
![]() | (1) |
![]() | (2a) |
![]() | (2b) |
![]() | (2c) |
![]() | (2d) |
Fig. 2 shows the phonon band structure of the ML together with those of the two largest-diameter NTs studied here. In the simplest picture, the phonon band structure of the NTs can be viewed as a projection of the slices of the 2D ML band structure onto the 1D Brillouin zone (BZ) of the NTs. To see this, denote the reciprocal basis of the ML by {b1,b2} and observe (see inset in Fig. 2) that the phonon wave vectors parallel to the NT axis are parallel to the b2 and b1 + 2b2 vectors of the (n,0) and (n,n) NTs respectively. In Fig. 2 we mark the m = 0 and m = n branches of the two thickest NTs with bold lines and observe a good correspondence with the corresponding branches of the ML. The main exception is the quadratic ZA branch of the ML. It is evident from the start that the ML ZA branch transforms differently from the other acoustic branches. Perhaps most notably, if one rolled up a piece of ML vibrating according to the q → 0 limit of its ZA branch, the result would not be a NT vibrating according to one of its ZA branches. Instead, all atoms would vibrate in phase inwards or outwards. In other words, one would obtain a NT vibrating according to its radial breathing mode (RBM) with a non-zero frequency. In the continuum limit this branch is expected to be flat around Γ and converge slowly towards ω = 0 with the NT radius [eqn (5)]. One can identify this branch around 4 meV for the two NTs.
The (n,0) NTs belong to the L(2n)nmc group and four one-dimensional, qA0, qAn, qB0, and qBn, and 2n − 2 two-dimensional irreps, qEm with 1 ≤ |m| ≤ n − 1, need to be considered for each value of q in the BZ. Among the one-dimensional irreps, modes belonging to A are even and modes belonging to B are odd with respect to inversion across a symmetry plane containing the NT axis. The subscript denotes the value of m in each case. At the Γ point, modes transforming according to 0A0 and 0E1 are IR-active, while those transforming according to 0B0 and 0E2 are Raman-active.43 Due to time-reversal symmetry, the frequencies only depend on |m|.
The (n,n) NTs belong to the L(2n)n/m group. For a q not at the center or the edge of the BZ, all irreps are two-dimensional, −qqEm with 0 ≤ |m| ≤ n. At Γ each of those can be further reduced according to parity with respect to inversion across the plane perpendicular to the NT axis, giving rise to the one-dimensional irreps 0A+m and 0A−m. Hence, IR and Raman activity analyses must be performed in terms of one-dimensional irreps. Modes transforming according to 0A−0 and 0A+1 are IR active, while those transforming according to 0A+0, 0A−1 and 0A+2 are Raman active.43Table 1 summarizes this information for both kinds of NTs.
ZZ nanotubes | AM nanotubes | ||
---|---|---|---|
m = 0 | IR | 0A0 | 0A−0 |
Raman | 0B0 | 0A+0 | |
|m| = 1 | IR | 0E1 | 0A+1 |
Raman | ∅ | 0A−1 | |
|m| = 2 | IR | ∅ | ∅ |
Raman | 0E2 | 0A+2 |
The symmetry analysis makes it straightforward to identify the acoustic branches which can be related to the continuum limit and the modes of the ML. Specifically, the frequencies of the two lowest-lying branches with m = 0 become linear functions of qz, as the LA and TA modes of a beam. In light of the parities introduced above, for (n,0) NTs the lowest-lying branch in qA0 tends to the longitudinal acoustic branch, whereas the lowest-lying branch in qB0 converges to the transverse acoustic branch. For (n,n) NTs, in a neighborhood of Γ the LA and TA branches both emerge from −qqE0. Exactly at Γ the LA branch will connect to the zero-frequency vibrational mode in 0A0− and the TA branch will connect to the zero-frequency vibrational mode in 0A+0. For both kinds of NT, the next m = 0 branch in order of frequency has a nonzero horizontal asymptote in the q → 0 limit, corresponding to the RBM of a beam, rod or tubule, whereas the two degenerate lowest-lying branches with |m| = 1 converge to the quadratic ZA branches. In (n,0) NTs both ZA branches belong to qE1, whereas for (n,n) NTs one ZA branch emerges from each of the −qqE1. They are still degenerate because of the time-reversal symmetry, which is not part of the line group.
To illustrate the advantages of this method for calculating the phonon spectrum of a NT over the more usual approach that only takes into consideration the space group of the three-dimensional crystal formed by the NT and the simulation cell, in Fig. 3 we compare the results of both techniques for the case of the (17,17) MoS2 NT, focusing on the low-energy part of the spectrum. At least six kinds of artifacts can be identified, all of which are avoided by the symmetry-based analysis. The 3D calculation enforces the homogeneity but not the isotropy of free space, meaning that one of the acoustic branches is misidentified as optical, with a non-zero frequency at Γ. Furthermore, the two ZA branches are incorrectly described: in the case shown, they contain “pockets” of imaginary frequencies in the long-wavelength limits which mistakenly point to a mechanical instability of the system. More generally, the basic parameters of the acoustic branches, including the speed of sound, are grossly mispredicted. Finally, branches which should be degenerate because of the discrete rotational symmetries are split.
(m,n) | R (Å) | ν TA (Å ps−1) | ν LA (Å ps−1) | ω RBM (rad ps−1) | |
---|---|---|---|---|---|
(9,0) | 5.3 | 32.3 (39.1) | 54.5 (64.3) | 249 (254) | 14.2 (12.4) |
(6,6) | 5.8 | 29.9 (39.1) | 56.4 (64.3) | 270 (276) | 11.7 (11.5) |
(12,0) | 6.7 | 33.8 (39.1) | 58.9 (64.3) | 313 (317) | 10.6 (9.90) |
(9,9) | 8.3 | 35.6 (39.1) | 56.7 (64.3) | 387 (390) | 8.04 (8.00) |
(12,12) | 10.7 | 36.4 (39.1) | 61.9 (64.3) | 504 (508) | 6.22 (6.13) |
(24,0) | 12.5 | 41.5 (39.1) | 63.0 (64.3) | 530 (588) | 5.25 (5.28) |
(17,17) | 15.1 | 36.1 (39.1) | 61.1 (64.3) | 310 (708) | 4.42 (4.38) |
The decomposition in terms of irreps makes it possible to identify the lowest m = 0 bands as the TA, LA and RBM and the lowest m = 1 band as the ZA of the NTs. Consequently, the group velocities and second-order coefficients can be extracted directly from the phonon spectra and compared to the continuum predictions [Table 2]. For the thinner NTs, the deviations are considerable and in general the agreement gets better with increasing diameter. Two different types of circumstances may contribute to this: local strain not completely captured by elasticity theory, and nonlocal interactions with parts of the sheet brought closer by the process of rolling up the NT. Thinner NTs deviate the most from the cylindrical geometry with uniform curvature, and have a shorter average interatomic distance that promotes interactions between different segments. The effect is more noticeable in the case of the TA branch, which involves displacements in the azimuthal direction, while the LA branch involves displacement along the z direction, which is not affected by the rolling-up. The frequencies of the RBM are predicted fairly accurately and their order is perfectly aligned with the ω ∝ R−1 prediction of eqn (2c). An exception to the general trend of better agreement between the continuum and atomistic predictions for thicker radii happens for the parabolic ZA branches, whose second derivative seems to be severely mispredicted for the (24,0) and (17,17) NTs. However, this disagreement essentially reflects that the quadratic part of the ZA branches covers only a small interval in the neighborhood of Γ, after which they hybridize with the other acoustic branches [Fig. 2].
The different IR and Raman activity, but overall agreement between the ML and large-radius NT bands [Fig. 2] makes it interesting to investigate the relation further. As mentioned above, phonon wave vectors parallel to the NT axis are parallel to the b2 and to b1 + 2b2 vectors for the (n,0) and (n,n) NTs, repectively. The m ≠ 0 modes thus correspond to phonon modes in the ML with a projection of along an axis perpendicular to the NT axis, which are defined by b2 and b1 + 2b2 for the (n,n) and (n,0) NTs. In Fig. 4 we plot the frequencies of all IR- and Raman-active modes at the Γ point of the NTs, and compare them with the acoustic branches of the MoS2 ML plotted along the corresponding directions, with the relevant wave number (m, or the corresponding component of q) reduced to the same scale. Given the values of |m| < 3 of all Raman- and IR-active irreps, the
value of the detectable modes depends fundamentally on the value of n and thereby on the NT radius. This explains why points towards the right of both panels in Fig. 4 deviate more clearly from the bulk phonon bands. They correspond to NTs with smaller diameters, and thus are more affected by the curvature and inter-segment interactions. For smaller values of q the active vibrational modes of the NTs trace the linear acoustic branches of the bulk very closely. This is, however, not the case for the ZA branch of the ML, whose curvature is much higher than what a naive reading of the NT results would suggest, even if restricted to the same range of q values that give good results for the linear branches. Hence, understanding and accounting for this deviation is the crucial step missing if one wants to reconstruct the most relevant features of the acoustic branches of ML MoS2.
We have already mentioned how the RBM has full rotational symmetry (m = 0) while the ZA branches of the NT belong to irreps with m = 1. Hence, the approximation to the ZA branch of the ML reconstructed from a single NT by sampling the corresponding modes at Γ and taking will not have a minimum at m = 0, but two minima at m = ±1, making it a poor estimate of the actual ML phonon branch. Moreover, the frequency of the RBM at q = 0 does not depend on the flexural rigidity of the ML [eqn (2d)], which determines the curvature of the ZA branch of the ML [eqn (2c)], but on the speed of sound of the linear branches. This suggests the possibility of devising a scheme to eliminate that effect and extract a corrected curvature.
For each of the low-lying triangles in Fig. 4 corresponding to the NT branches with |m| = 2 that we would like to use to reconstruct the ZA branch of the bulk, we define a “corrected q” as Q = 2π(|m| − m0)f/n. Here, f is a predefined factor with the values 1 and for the AM and ZZ NTs, respectively, and m0 ∈ (0,1) is a uniform offset. The f prefactor ensures that all wave numbers are measured in the same units, in keeping with the reciprocal-space basis discussed above. m0 accounts for the nonzero RBM frequency, and becomes less relevant with increased n, consistent with the continuum limit [eqn (2d)]. We then fit the frequencies of those points to a parabola of the form
, with k and m0 as fitting parameters. The result is shown in Fig. 5. The fitted value of m0 is 0.69, and the curvature of the ZA branch close to Γ is in excellent agreement with the actual calculation. It is important to emphasize that the curve is not fitted to the ML data, but to the Raman-active frequencies of the NTs. The introduction of f allows us to obtain better statistics for the fit by including both types of NTs in the same plot, and is consistent with the prediction of an isotropic curvature in eqn (2c).
![]() | ||
Fig. 5 Reconstruction of the ZA branch of monolayer MoS2 based on the low-lying points with |m| = 2 from Fig. 4, after applying the correction described in the main text. Note that the blue curve is fitted to the orange points and not to the reference bulk band (gray) but manages to reproduce the curvature of the latter. |
An interesting question when trying to apply this set of techniques to obtain information about the spectrum of other 2D crystals is which NTs to choose. Is it likely that some preexisting constraints exist: for instance, NTs of a particular chirality might be easier to synthesize. However, some general guidelines can be formulated. Firstly, although large NTs should, in principle, provide a better approximation of bulk-like behavior, and sample the monolayer dispersions at points closer to Γ, our example shows that NTs with diameters in the nm scale can afford very good results. Using smaller NTs has the advantage that the frequencies to be measured are larger and more spaced. Moreover, the isotropy predicted by eqn (2c) can be exploited by combining data from NTs with different chiralities to obtain more points in the same range of frequencies and therefore a better characterization of the ZA branch of the monolayer.
![]() | (3) |
Fig. 6 shows the result for all transmissions per unit length. The NTs with the largest diameters show an almost perfect ML-like transmission in the part of the spectrum below ∼30 meV (corresponding to the acoustic branches of the bulk ML) thanks to the contributions of higher values of |m|. This is to be expected since thicker NTs are locally closer to a planar structure and also less affected by the discretization of m, but it is still remarkable that the (9,9) is already close to the bulk in this region. In contrast, even for those thicker NTs the high-energy region of the phonon spectrum has marked discrepancies with the ML and a clear dependence on size and chirality. This phenomenon is most clearly observable by focusing on the region from 30 meV to 36 meV, a gap devoid of allowed phonon modes in the ML: in the (12,12) NT the gap is shifted to the left, while in the (12,0) NT it appears shifted to the right, with a second gap at a slightly lower frequency, and in the (9,9) NT it almost disappears due to the appearance of a new bundle of branches in that region of the spectrum.
The pieces of information obtained above (speeds of sound of the TA and LA branches, and second derivative of the ZA branch) can be used to build a Debye-like model to obtain a first approximation to the phonon transmission. The contribution to the density of states per unit cell from a phonon branch whose dispersion follows a power law of the form ω = αqn is
![]() | (4) |
![]() | (5) |
Based on these results and on the connections between the phonon branches of quasi-2D materials and the continuum theory of elastic waves, we have proposed a method to reconstruct the slopes of the two linear branches and the second derivative of the quadratic acoustic branch of a 2D ML based on measurements of the corresponding NTs. The proposed method has the advantage of relying only on Raman and infrared measurements of modes at the Γ point, which are more easily accessible than specialized techniques like inelastic neutron scattering that can offer a more comprehensive picture of the ML phonon bands. We have tested the proposed method on MoS2, showing a very good agreement with direct ab initio calculations. Finally, we have used the reconstructed branches to create an estimate of the phonon transmission of single-layer MoS2, which provides a reasonable level of accuracy in the low-energy region.
Our method and results provide an alternative method to access information about the low-frequency vibrational modes of 2D materials that is difficult to measure directly. Those pieces of information act as a first fundamental building block and a way to test models of their thermal transport properties, which are crucial in several applications. Furthermore, we have provided a detailed example of a symmetry-based approach for the phonon spectrum of a relatively complex family of nanotubes, which significantly reduces the complexity of the problem and provides important additional information about the results.
This journal is © the Owner Societies 2019 |