Вы находитесь на странице: 1из 111

This article was downloaded by: [USC University of Southern California]

On: 11 November 2009


Access details: Access Details: [subscription number 906867864]
Publisher Taylor & Francis
Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House, 37-
41 Mortimer Street, London W1T 3JH, UK

Advances in Physics
Publication details, including instructions for authors and subscription information:
http://www.informaworld.com/smpp/title~content=t713736250

Phonons in carbon nanotubes


M. S. Dresselhaus a; P. C. Eklund b
a
Department of Electrical Engineering and Computer Science and Department of Physics,
Massachusetts Institute of Technology, Cambridge, MA 02139, USA. b Department of Physics,
Pennsylvania State University, University Park, PA 16802, USA.

To cite this Article Dresselhaus, M. S. and Eklund, P. C.'Phonons in carbon nanotubes', Advances in Physics, 49: 6, 705 —
814
To link to this Article: DOI: 10.1080/000187300413184
URL: http://dx.doi.org/10.1080/000187300413184

PLEASE SCROLL DOWN FOR ARTICLE

Full terms and conditions of use: http://www.informaworld.com/terms-and-conditions-of-access.pdf


This article may be used for research, teaching and private study purposes. Any substantial or
systematic reproduction, re-distribution, re-selling, loan or sub-licensing, systematic supply or
distribution in any form to anyone is expressly forbidden.

The publisher does not give any warranty express or implied or make any representation that the contents
will be complete or accurate or up to date. The accuracy of any instructions, formulae and drug doses
should be independently verified with primary sources. The publisher shall not be liable for any loss,
actions, claims, proceedings, demand or costs or damages whatsoever or howsoever caused arising directly
or indirectly in connection with or arising out of the use of this material.
A DVANCES IN P HYSICS, 2000, VOL. 49, NO. 6, 705 ± 814

Phonons in carbon nanotubes

M . S. D RESSELHAUS{* and P . C . EKLUND{*


{ Department of Electrical Engineering and Computer Science and Department of
Physics, Massachusetts Institute of Technology, Cambridge, MA 02139, USA
{ Department of Physics, Pennsylvania State University, University Park, PA 16802,
USA

[Received December 1999; revision received 14 March 2000; accepted 17 March 2000]
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

Abstract
A broad review of the unusual one-dimensional properties of phonons in
carbon nanotubes is presented, including phonons in isolated nanotubes and in
crystalline arrays of nanotubes in nanotube bundles. The main technique for
probing the phonon spectra has been Raman spectroscopy and the many unique
and unusual features of the Raman spectra of carbon nanotubes are reviewed.
Also included is a brief review of the thermal properties of carbon nanotubes in
relation to their unusual phonon dispersion relations and density of states.

Contents PAGE
1. Introduction 706
2. Background 707
2.1. Structure and notation 707
2.2. Electronic structure 712
3. Phonon modes 718
3.1. Phonon dispersion relations for nanotubes 720
3.2. Raman and infrared active modes of carbon nanotubes 725
3.3. Inter-tube or intra-bundle interactions 729
4. Raman spectra of single-wall nanotubes 734
4.1. Radial breathing mode phenomena 738
4.2. Tangential stretch modes 743
4.3. Temperature dependence of the Raman spectra 748
4.4. High pressure e€ ects on the tangential modes 750
4.5. Anti-Stokes spectra 753
4.6. Surface enhanced Raman spectra in carbon nanotubes 759
4.7. Polarization studies in Raman scattering 767
4.8. D-band and G 0-band spectra 776
4.9. Overtone and combination modes 785
4.9.1. Overtones 786
4.9.2. Combination modes 789
4.9.3. Overtones and combination modes for the D-band and
G 0-band 790
4.10. Raman studies of doped carbon nanotubes 791
5. Thermal properties 796
5.1. Speci® c heat 796
5.2. Thermal conductivity 800

* e-mail: millie@mgm.mit.edu; pce3@ psu.edu

Advances in Physics ISSN 0001± 8732 print/ISSN 1460± 6976 online # 2000 Taylor & Francis Ltd
http://www.tandf.co.uk/journals
706 M. S. Dresselhaus and P. C. Eklund

5.3. Thermopower 805


6. Concluding remarks 807
Acknowledgements 808
References 808

1. Introduction
Carbon nanotubes provide a remarkable model one-dimensional (1D) system,
one atom in thickness, a few tens of atoms in circumference, and many microns in
length. Although much has been written about the remarkable electronic properties
which allow a carbon nanotube to be either semiconducting or metallic depending on
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

its diameter and chirality [1, 2], the phonon properties are also remarkable, showing
unique 1D behaviour and special characteristics, such as the radial breathing mode,
the twist acoustic mode and 1D phonon subbands. The strong coupling between
electrons and phonons in this one-dimensional system furthermore gives rise to
highly unusual resonance Raman spectra, and unique features in the Stokes and anti-
Stokes spectra, as elaborated on in this article. The capability of surface-enhanced
Raman spectroscopy to give rise to large enhancement factors in the Raman signal,
likewise results in many unusual phenomena, not observed in other systems.
The article starts by brie¯ y reviewing the basic structure and notation used to
describe single-wall carbon nanotubes (SWNTs) in section 2.1, followed by a
description of the unique electronic structure of carbon nanotubes in section 2.2,
largely linked to the unique symmetry of a graphene layer and to the 1D nature of
the density of electronic states in a nanotube. With this background, the phonon
modes in carbon nanotubes are reviewed in section 3, ® rst for isolated single-wall
nanotubes section 3.1, and then for crystalline arrays of nanotubes arranged in
nanotube bundles in section 3.3. The group theory and the Raman and infrared
selection rules for single-wall carbon nanotubes (SWNTs) are reviewed in section 3.2.
The e€ ect of the inter-tube interactions which are expected from the formation of
nanotube bundles is discussed in section 3.3. Throughout, emphasis is given to the
di€ erences between the phonon modes in a carbon nanotube and the phonon modes
of a graphene sheet which conceptually gives rise to the nanotube.
The dominant experimental technique for studying phonons in carbon nanotubes
has been Raman scattering, which is reviewed in detail in section 4. Particular
emphasis is given to the phenomenon of the diameter selective resonant Raman
scattering process. The rather extensive use of the low frequency radial breathing
mode for nanotube diameter characterization is reviewed in section 4.1, including the
e€ ect of inter-tube interactions. The higher frequency modes associated with carbon
atom displacements in the cylindrical shell of the nanotube are discussed in
section 4.2 with particular emphasis given to the di€ erence in spectra between
semiconducting and metallic nanotubes. Many other topics on Raman spectroscopy
are then reviewed, including the temperature (section 4.3) and pressure (section 4.4)
dependence of the Raman spectra, the anti-Stokes spectra (section 4.5), the surface-
enhanced Raman spectra (section 4.6), polarization phenomena (section 4.7), the
special features in the Raman spectra associated with wave vectors close to the
K-point in the Brillouin zone (section 4.8), overtones and combination modes
(section 4.9), and the e€ ect of doping on the Raman spectra (section 4.10). Since
our present knowledge of many of these topics is still at an early stage of
Phonons in carbon nanotubes 707

understanding, emphasis is given to important open issues and insights for gaining
further understanding of phonons in this unique 1D system.
Thermodynamic and thermal transport phenomena involving phonons are
discussed in section 5, including speci® c heat (section 5.1), thermal conductivity
(section 5.2), and thermopower (section 5.3). This is an area of research which is only
now beginning to be addressed by the world-wide scienti® c community. Since
potential practical applications may depend signi® cantly on the thermodynamic
parameters, it is highly desirable for the research community to be aware of the
current status of developments on this topic. Finally, perspectives on the current
status of the ® eld are brie¯ y presented in section 6.
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

2. Background
This section provides a brief introduction to the structure and electronic
properties of carbon nanotubes, emphasizing those aspects that are particularly
relevant to understanding the behaviour of the phonons in carbon nanotubes and
experiments that are carried out to study the phonon spectra, and the phonon
density of states, with particular emphasis given to the unusual 1D properties of
carbon nanotubes.

2.1. Structure and notation


A single-wall carbon nanotube (SWNT) can be described as a single layer of a
graphite crystal that is rolled up into a seamless cylinder, one atom thick, usually
with a small number (perhaps 20± 40) of carbon atoms along the circumference and a
long length (microns) along the cylinder axis [1]. This nanotube is speci® ed by the
chiral vector Ch
Ch ˆ na1 ‡ ma2 ² …n;m †; …1†
which is often described by the pair of indices …n;m † that denote the number of unit
vectors na1 and ma2 in the hexagonal honeycomb lattice contained in the vector Ch .
As shown in ® gure 1, the chiral vector Ch makes an angle ³, called the chiral angle,
with the zigzag or a1 direction. The vector Ch connects two crystallographicall y
equivalent sites O and A on a two-dimensional (2D) graphene sheet where a carbon
atom is located at each vertex of the honeycomb structure [3]. The axis of the zigzag
nanotube corresponds to ³ ˆ 08, while the armchair nanotube axis corresponds to
³ ˆ 308, and the general nanotube axis corresponds to 0 µ ³ µ 308. The seamless
cylinder joint of the nanotube is made by joining the line AB 0 to the parallel line OB
in ® gure 1. In terms of the integers …n;m †, the nanotube diameter dt is given by
dt ˆ Ch =p ˆ 31=2 aC¡C …m 2 ‡ mn ‡ n2 †1=2 =p; …2†
where aC¡C is the nearest-neighbou r C± C distance (1.421 A Ê in graphite), Ch is the
length of the chiral vector Ch and the chiral angle ³ is given by
³ ˆ tan ¡1 ‰ 31=2 m =…m ‡ 2n†Š: …3†
Thus, a nanotube can be speci® ed by either its …n;m † indices or equivalently by dt
and ³. Next we de® ne the unit cell OBB 0A of the 1D nanotube in terms of the unit
cell of the 2D honeycomb lattice de® ned by the vectors a1 and a2 (® gure 1).
In ® gure 2 we show (a) the unit cell and (b) the Brillouin zone of two-dimensional
graphite as a dotted rhombus and shaded hexagon, respectively, where a1 and a2 are
708 M. S. Dresselhaus and P. C. Eklund

y
B
B x

T ?
R A
a1
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

O Ch
a2
Figure 1. The unrolled honeycomb lattice of a nanotube. When we connect lattice sites O
¡! ¡!
and A, and sites B and B 0, a nanotube can be constructed. OA and OB de® ne the
chiral vector Ch and the translational vector T of the nanotube, respectively. The
rectangle OAB 0 B de® nes the unit cell for the nanotube. The ® gure is constructed for
an …n; m† ˆ …4; 2† nanotube [2].

(a) y (b)
x b1
K
A B
G M
a1 ky

a2 kx b2
Figure 2. (a) The unit cell and (b) Brillouin zone of two-dimensional graphite are shown as
the dotted rhombus and the shaded hexagon, respectively. ai , and bi , …i ˆ 1; 2† are
basis vectors and reciprocal lattice vectors, respectively. Energy dispersion relations
are obtained along the perimeter of the dotted triangle connecting the high symmetry
points, G, K and M [4].

basis vectors in real space, and b1 and b2 are reciprocal lattice basis vectors. In the
x;y coordinates shown in ® gure 2, the real space basis vectors a1 and a2 of the
hexagonal lattice are expressed as

31=2 a 31=2 a
a1 ˆ … 2
a; ;
2 † a2 ˆ … 2
a; ¡ ;
2 † …4†

Ê is the lattice constant of two-dimen-


where a ˆ ja1 j ˆ ja2 j ˆ 1:42 £ 31=2 ˆ 2:46 A
sional graphite. Correspondingly the basis vectors b1 and b2 of the reciprocal lattice
are given by:
Phonons in carbon nanotubes 709

b1 ˆ… 32pa ; 2pa †;
1=2
b2 ˆ … 32pa ; ¡ 2pa †
1=2
…5†

corresponding to a lattice constant of 4p =31=2 a in reciprocal space. The direction of


the basis vectors b1 and b2 of the reciprocal hexagonal lattice are rotated by 308 from
the basis vectors a1 and a2 of the hexagonal lattice in real space, as shown in ® gure 2.
By selecting the ® rst Brillouin zone as the shaded hexagon shown in ® gure 2 (b), the
highest symmetry is obtained for the Brillouin zone of 2D graphite. Here we de® ne
the three high symmetry points, G, K and M as the centre, the corner, and the centre
of the edge, respectively. The energy dispersion relations are calculated for the
triangle GMK shown by the dotted lines in ® gure 2 (b). ¡!
To de® ne the unit cell for the 1D nanotube, we de® ne OB in ® gure 1 as the
shortest repeat distance along the nanotube axis, thereby de® ning the translation
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

vector T
T ˆ t1 a1 ‡ t2 a2 ² …t1 ;t2†; …6†
where the coe cients t1 and t2 are related to …n; m † by
t1 ˆ …2m ‡ n†=dR ;
…7†
t2 ˆ ¡…2n ‡ m †=dR ;
where dR is the greatest common divisor of …2n ‡ m ;2m ‡ n† and is given by
d; if n ¡ m is not a multiple of 3d ;
dR ˆ …8†
3d ; if n ¡ m is a multiple of 3d ;
in which d is the greatest common divisor of …n;m †. The magnitude of the translation
vector T ˆ jTj is
jTj ˆ 31=2 L =dR …9†
where L is the length of the chiral vector Ch ˆ pdt and dt is the nanotube diameter.
The unit cell of the nanotube is de® ned as the area delineated by the vectors T and
Ch . The number of hexagons, N, contained within the 1D unit cell of a nanotube is
determined by the integers …n;m † and is given by
2…m 2 ‡ n2 ‡ nm †
Nˆ : …10†
dR
The addition of a single hexagon to the honeycomb structure in ® gure 1 corresponds
to the addition of two carbon atoms. Assuming a value aC¡C ˆ 0:142 nm on a
carbon nanotube, we obtain dt ˆ 1:36 nm and N ˆ 20 for a (10,10) nanotube. Since
the real space unit cell is much larger than that for a 2D graphene sheet, the 1D
Brillouin zone (BZ) for the nanotube is much smaller than the BZ for a single two-
atom graphene 2D unit cell. Because the local crystal structure of the nanotube is so
close to that of a graphene sheet, and because the Brillouin zone is small, Brillouin
zone-folding techniques have been commonly used to obtain approximate electron
and phonon dispersion relations for carbon nanotubes …n; m † with speci® c symmetry.
Whereas the lattice vector T, given by equation (6), and the chiral vector Ch ,
given by equation (1), both determine the unit cell of the carbon nanotube in real
space, the corresponding vectors in reciprocal space are the reciprocal lattice vectors
K2 along the nanotube axis and K1 in the circumferential direction, which gives
discrete k values in the direction of the chiral vector Ch . The vectors K1 and K2 are
710 M. S. Dresselhaus and P. C. Eklund

obtained from the relation Ri ¢ Kj ˆ 2p¯ij , where Ri and Kj are, respectively, the
lattice vectors in real and reciprocal space, and K1 and K2 therefore satisfy the
relations
Ch ¢ K1 ˆ 2p ; T ¢ K1 ˆ 0;
…11†
Ch ¢ K2 ˆ 0; T ¢ K2 ˆ 2p:

From equations (11) it follows that K1 and K2 can be written as:


1 1
K1 ˆ …¡t2 b1 ‡ t1 b2 †; K2 ˆ …mb1 ¡ nb2 †; …12†
N N
where b1 and b2 are the reciprocal lattice vectors of a two-dimensional graphene
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

sheet given by equation (5). The N wave vectors ·K1 (· ˆ 0;. . . ; N ¡ 1) give rise to
N discrete k vectors in the circumferential direction. For each of the · discrete values
of the circumferential wave vectors, a one-dimensional electronic energy band
appears, whereas each · gives rise to 6 branches in the phonon dispersion relations.
Because of the translational symmetry of T, we have continuous wave vectors in the
direction of K2 for a carbon nanotube of in® nite length. However, for a nanotube of
® nite length L t , the spacing between wave vectors is 2p =L t and e€ ects associated with
the ® nite nanotube length have been observed experimentally [5].
Typically, the experimental single-wall nanotube (SWNT) samples have a distri-
bution of diameters and chiral angles because of the absence of experimental
techniques at present for producing SWNTs with a unique dt and ³. The nanotube
material produced by the presently available synthesis methods, including laser
vaporization, carbon arc discharge, vapour phase deposition, and solar energy
synthesis, appears in a scanning electron microscopy (SEM) image as a mat of
carbon nanotubes bundles 10± 20 nm in diameter and up to 100 mm or more in length
(see ® gure 3) and containing between 30± 500 SWNTs. The nanotubes within a
bundle are twisted together, thereby maximizing the bonding interaction between
SWNTs within the bundles where the hexagons on adjacent nanotubes tend to be in
the same AB registry as in crystalline graphite [7]. The nanotube bundles attract one
another and wrap around each other to form ropes [7]. These nanotube ropes are
accompanied by varying amounts of amorphous carbon, residual catalyst, and other
unwanted material, from which the nanotube ropes must be separated. A number of
experimental observations have also been made on a material called `bucky paper’,
which refers to a ¯ at tangled mat of bundles of carbon nanotubes that are collected
on ® lter paper as a suspension of SWNT bundles is passed through the ® lter paper.
Under transmission electron microscopy (TEM) examination, each nanotube
rope is found to consist primarily of bundles of single-wall carbon nanotubes that
are mostly aligned along a common axis (see ® gure 3 (a)). X-ray di€ raction (which
views many ropes at once) and transmission electron microscopy (which typically
views one or two ropes) show that the diameters of the single-wall nanotubes in
typical SWNT samples have a strongly peaked narrow distribution of diameters.
Typical nanotube diameters in the ropes are between 0.9± 1.8 nm, depending on the
catalyst and growth conditions though smaller diameter nanotubes as small as
0.4 nm have been reported [8]. For the synthesis conditions used in the early work,
the diameter distribution was strongly peaked at 1.38§ 0:02 nm, very close to the
diameter of an ideal (10,10) nanotube. X-ray di€ raction measurements [6, 9] further
showed that, within these `ropes’, the single-wall nanotubes form a two-dimensional
Phonons in carbon nanotubes 711
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

(a)

(b)
Figure 3. (a) Ropes of single-wall carbon nanotubes observed by scanning electron
microscopy (SEM). The ropes are 10± 20 nm thick and ¹100 mm long. (b) At higher
magni® cation, the TEM image shows that each rope contains a bundle of single-wall
nanotubes with diameters of ¹1.4 nm, arranged in a triangular lattice (with lattice
constant ¹1.7 nm). The lower image is seen when the rope bends, so that the rope
cross section is in the image plane of the transmission electron microscope [6].

triangular lattice with a lattice constant of 1.7 nm, and an inter-tube separation of
0.315 nm at closest approach within a rope, in good agreement with prior theoretical
modelling results [10, 11]. The diameter and chiral angle of individual nanotubes are
measured by transmission electron microscopy [12] and by scanning tunnelling
microscopy [13, 14] techniques. Because of the dependence of the radial breathing
mode frequency on the inverse nanotube diameter (see sections 3.3, 4.1 and 4.4),
measurement of the radial mode frequencies at a variety of laser excitation energies
provides a convenient secondary characterization tool for the measurement of the
diameter distribution contained within a typical single-wall nanotube sample.
However, it has been very di cult to measure dt and ³ on the same SWNT that is
used for property measurements. For example, it has not yet been possible to
measure the Raman spectrum on an individual SWNT, characterized for its dt and ³.
712 M. S. Dresselhaus and P. C. Eklund
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

Figure 4. Transmission electron microscopy pictures of carbon nanotubes with three


common cap terminations: (a) a symmetric polyhedral cap, (b) an asymmetric
polyhedral cap and (c) a symmetrical ¯ at cap [15].

The Raman spectra of both single-wall and multiwall nanotubes have been
investigated, where the multiwall nanotubes can be described as capped coaxial
cylinders separated radially by ¹ 3:4 A Ê (see ® gure 4) [1]. Each of the constituent
cylinders can, in principle, be speci® ed by its chiral vector Ch through its indices
…n;m †, or equivalently by its diameter dt and chiral angle ³. Because of the strong
dependence of the Raman spectra on nanotube diameter and to a lesser degree on
the chiral angle, the structural characterization of the nanotube is necessary for the
comparison of experimental results and theoretical models, and for the comparison
of Raman spectra obtained in di€ erent laboratories, on di€ erent Raman spectro-
meters and when using di€ erent laser excitation energies Elaser .

2.2. Electronic structure


Because of the strong coupling between electrons and phonons in the resonance
Raman e€ ect, the remarkable electronic properties of carbon nanotubes play an
important role in discussing the unusual Raman spectra of these unique one-
dimensional periodic structures. In single-wall carbon nanotubes, con® nement of
the electronic wave function in the radial direction is provided by the monolayer
thickness of the nanotube in the radial direction. Circumferentially, the periodic
boundary condition applies to the enlarged unit cell that is formed in real space. The
application of this periodic boundary condition to the graphene electronic states
leads to the prediction of a remarkable electronic structure for carbon nanotubes of
small diameter.
The 1D electronic energy band structure for carbon nanotubes [16 ± 20] is related
to the energy band structure calculated for the 2D graphene honeycomb sheet used
to form the nanotube. These calculations for the electronic structure of SWNTs [2]
show that about 1/3 of the nanotubes are metallic and 2/3 are semiconducting,
depending on the nanotube diameter dt and chiral angle ³. It can be shown that
metallic conduction in a …n; m † carbon nanotube is achieved when
Phonons in carbon nanotubes 713

Figure 5. One-dimensional energy dispersion relations for (a) armchair (5,5) nanotubes, (b)
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

zigzag (9,0) nanotubes and (c) zigzag (10,0) nanotubes. The energy bands with a
symmetry are non-degenerate, while the e-bands are doubly degenerate at a general
wave vector k [4, 21, 22].

2n ‡ m ˆ 3q; …13†
where q is an integer. All armchair carbon nanotubes (³ ˆ 308) are metallic and
satisfy the more general equation (13).
Calculated dispersion relations based on these simple zone folding considerations
for tight binding energy bands are shown for metallic nanotubes …n;m † ˆ …5;5† and
(9,0) in ® gures 5 (a) and (b), respectively, and for a semiconducting nanotube
…n;m † ˆ …10;0† in ® gure 5 (c) [21]. These results are consistent with more detailed
ab initio calculations of the band structure [23]. The calculated electronic structure
can be either metallic or semiconducting depending on the choice of …n;m † as given
by equation (13), although there is no di€ erence in the local chemical bonding
between the carbon atoms in the nanotubes, and no doping impurities are present [4].
These surprising results and the unique features of the electronic structure of
SWNTs can be understood on the basis of the electronic structure of a graphene
sheet which is a zero gap semiconductor [24] with bonding and antibonding p bands
that are degenerate at the K-point (zone corner) of the hexagonal 2D Brillouin zone
[2]. The periodic boundary conditions for the 1D carbon nanotubes of small
diameter permit only a few wave vectors to exist in the circumferential direction,
and these wave vectors k satisfy the relation n¶ ˆ pdt , where ¶ ˆ 2p =k is the de
Broglie wavelength. Metallic conduction occurs when one of these allowed wave
vectors k passes through the K-point of the 2D Brillouin zone, where the 2D valence
and conduction bands are degenerate because of the special symmetry of the 2D
graphene lattice [2]. As the nanotube diameter increases, more wave vectors become
allowed for the circumferential direction, so that the nanotubes become more two-
dimensional and the semiconducting band gap disappears. The band gap for isolated
semiconducting carbon nanotubes is proportional to the reciprocal nanotube
diameter 1 =dt . At a nanotube diameter of dt ¹ 3 nm, the bandgap becomes
comparable to thermal energies at room temperature, showing that small diameter
nanotubes are needed to observe 1D quantum e€ ects.
Of particular importance to the discussion of the resonant Raman spectra in
single-wall carbon nanotubes is the 1D density of states plots shown in ® gure 6 for:
(a) a semiconducting (10,0) zigzag carbon nanotube and (b) a metallic (9,0) zigzag
carbon nanotube. The results for the 1D electronic density of states show sharp
714 M. S. Dresselhaus and P. C. Eklund

(a) (n,m)=(10,0)
1.0

DOS [states/unit cell of graphite]

0.5

0.0
-4.0 -3.0 -2.0 -1.0 0.0 1.0 2.0 3.0 4.0
Energy/ g 0
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

(b) (n,m)=(9,0)
1.0
DOS [states/unit cell of graphite]

0.5

0.0
-4.0 -3.0 -2.0 -1.0 0.0 1.0 2.0 3.0 4.0
Energy/g 0

Figure 6. Electronic 1D density of states per unit cell of a 2D graphene sheet for two …n; 0†
zigzag nanotubes : (a) the …10; 0† nanotube which has semiconducting behaviour,
(b) the …9; 0† nanotube which has metallic behaviour. Also shown in the ® gure is the
density of states for the 2D graphene sheet (dotted line) [25].

=2
singularities associated with the …E ¡ E0 †¡1 van Hove singularities about each
subband edge at energy E0 (see ® gure 5). The electronic density of states plots in
® gure 6 show that the metallic nanotubes have a small, but non-vanishing 1D density
of states at the Fermi level (which is at E ˆ 0 in ® gure 6), and this non-vanishing
density of states is independent of energy until the energies of the ® rst subband edges
of the valence and conduction bands are reached. In contrast, for a 2D graphene
sheet (dashed curve), the 2D density of states is zero at the Fermi level (where also
E ˆ 0 in ® gure 6), and varies linearly with energy, as we move away from the Fermi
level. Furthermore, the density of states for the semiconducting 1D nanotubes is zero
throughout the bandgap, as shown in ® gure 6 (a), and their bandgap energy Eg is
equal to the energy di€ erence E11 …dt† between the two van Hove singularities in the
1D density of states that span the Fermi level, where it is noted that Eg is
proportional to the reciprocal nanotube diameter Eg / 1 =dt . Because of these
singularities in the density of states, high optical absorption is expected when the
photon energy matches the energy separation between an occupied peak in the
electron density of states and one that is empty. This situation occurs at the band gap
for the semiconducting nanotubes, but also at higher energies for transitions from an
occupied subband edge state to the corresponding unoccupied subband edge state.
Such transitions between subband edge states can occur for both semiconducting
Phonons in carbon nanotubes 715
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

Figure 7. Electronic 1D density of states (DOS) calculated with a tight binding model for
(8, 8), (9, 9), (10, 10), (11, 11) and (12, 12) armchair nanotubes and for (14, 0), (15,
0), (16, 0), (17, 0) and (18, 0) zigzag nanotubes and assuming a nearest neighbour
carbon± carbon interaction energy ® 0 ˆ 3:0 eV [27]. Wavevector conserving optical
transitions can occur between mirror image singularities in the 1D density of states,
i.e. v1 ! c1 and v2 ! c2 , etc. and these optical transitions are given in the ® gure in
units of eV. These interband transitions are denoted in the text by E11 , E22 , etc. and
are responsible for the resonant Raman e€ ect discussed extensively in this review [27].

and metallic nanotubes. Comparing the density of states curves in ® gure 6, we see
that the smallest energy separation between subband edge states for the semicon-
ducting nanotube (10, 0) is much smaller than the corresponding separation between
subband edges for the metallic (9, 0) nanotube. Because of the very large electron
density of states at the van Hove singularities (or subband edges) the intensity of the
interband optical transitions Eii …dt† are exceptionally strong, giving rise to excep-
tionally high intensities for the resonant Raman e€ ect associated with the 1D density
of electronic states. Thus we can expect the resonance Raman e€ ect for carbon
nanotubes to be much stronger than for 3D crystalline materials, based on the low
dimensionality of carbon nanotubes.
The general characteristics that are predicted for the 1D electronic density of
states of carbon nanotubes have recently been con® rmed by low temperature STM/
STS (scanning tunnelling microscopy/spectroscopy) studies carried out on isolated
single-wall carbon nanotubes [13, 14, 26], as discussed below. Insight into the
variation of the electronic density of states with nanotube diameter dt is provided
by ® gure 7, where the density of states is presented for (8, 8), (9, 9), (10, 10), (11, 11)
and (12, 12) armchair nanotubes and for (14, 0), (15, 0), (16, 0), (17, 0) and (18, 0)
zigzag nanotubes [28]. Referring to ® gure 7, we see that the lowest energy transition
M
for the armchair nanotube, denoted by E11 …dt†, varies continuously from 2.4 eV for
the (8, 8) nanotube to 1.6 eV for the (12, 12) armchair nanotube.
The interpretation of the interband transitions between van Hove singularities
for zigzag nanotubes and chiral nanotubes can be understood by plotting the
energies for the transitions between the van Hove singularities in the valence and
716 M. S. Dresselhaus and P. C. Eklund

3.0
3.0

1.96
1.92
2.0
2.0
1.8
t ) [eV]

1.58
Eii ?(dE

E S33
1.49
1.0
1.0 EM
11

ES22
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

ES11
0.0
0.0
0.0 1.0 1.35 2.0
2.0 3.0
d t [nm ]
Figure 8. Calculation [29± 31] of the energy separations Eii …dt † for all …n; m † values versus
nanotube diameter in the range 0:7 < dt < 3:0 nm using ® 0 ˆ 2:9 eV. Semiconducting
and metallic nanotubes are indicated by crosses and open circles, respectively, and the
S S M S
four lowest energy transitions are labelled by E11 …dt †, E22 …dt †, E11 …dt†, and E33 …dt †,
where S and M, respectively, refer to semiconducting and metallic nanotubes. The
® lled squares denote zigzag tubes. The vertical lines denote dt ˆ 1:35 § 0:20 nm for a
particular single-wall carbon nanotube sample [32].

conduction bands Eii …dt † of all possible …n;m † nanotubes, including not only
armchair and zigzag nanotubes, as shown in ® gure 7, but also all chiral nanotubes.
Such a plot of Eii …dt † versus dt is shown in ® gure 8. In this plot, the nearest neighbour
carbon ± carbon transfer energy ® 0 was taken to be 2.9 eV which provides a good ® t to
a variety of experiments, including resonance Raman scattering, optical absorption
and scanning tunnelling spectroscopy studies as discussed below. Figure 8 is used
extensively to interpret resonance Raman spectra in carbon nanotubes. The two
S S
lowest energy transitions E11 …dt† and E22 …dt† are for semiconducting nanotubes,
M
while the next higher energy transition is E11 …dt† for metallic nanotubes, followed by
S
E33 …dt†, as indicated in the ® gure [29, 32]. The bounds on Eii …dt† at constant dt and i
are delineated by zigzag nanotubes and are due to trigonal warping e€ ects [32]. For
example, the (15,0) zigzag nanotube with diameter dt ˆ 1:17 nm has a lower and
M
upper bound for E11 …dt† ˆ 2:0 and 2.4 eV, as shown in ® gure 7. In relating ® gure 8 to
the resonance Raman e€ ect, we note that resonance occurs when Eii …dt † is in
resonance with either the incident or scattered photon.
Since measurements of dI =dV in the STS (scanning tunnelling spectroscopy)
mode of a scanning tunnelling microscope yields a signal (shown in ® gure 9), which is
proportional to the 1D density of states, the STS technique has become a powerful
tool for studying the electronic structure of both metallic and semiconducting single-
wall carbon nanotubes [13, 14]. The top 4 traces in ® gure 9 show that the bandgap or
S
energy separation is E11 …dt† ’ 0:6 eV for the indicated semiconducting nanotubes,
M
while the lower 3 traces show energy separations of E11 …dt† ’ 1:8 eV for metallic
nanotubes. The combined STM/STS studies [13, 33] are consistent with: (1) about
2/3 of the nanotubes being semiconducting and 1/3 being metallic; (2) the density of
states exhibiting van Hove singularities, characteristic of the expectations for 1D
Phonons in carbon nanotubes 717
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

Figure 9. Derivative of the current± voltage …dI =dV † curves obtained by scanning tunnelling
spectroscopy on various isolated single-wall carbon nanotubes with diameters near
1.4 nm. Nanotubes #1± 4 are semiconducting and #5± 7 are metallic [13].

systems; (3) energy gaps for the semiconducting nanotubes that are proportional to
1 =dt . Using the approximate relation Eg ’ 2® 0 aC¡C =dt for the band gap of single-
wall nanotubes, the nearest neighbour overlap energy ® 0 (or the transfer integral of a
tight binding model) can be found. The STS experiments con® rm that the density of
electronic states near the Fermi level is zero for semiconducting nanotubes, and non-
zero for metallic nanotubes [13]. These electronic density of states curves in ® gure 6
and the plot of interband transition energies in ® gure 8 are also important for
explaining the resonance Raman experiments on carbon nanotubes discussed in
section 4 of this review.
Subsequent atomic resolution STM/STS studies coupled with tight-binding
calculations [34] were able to make a detailed identi® cation of the features in the
dI =dV versus voltage spectrum of an individual (13,7) metallic carbon nanotube
with speci® c interband transitions. Comparison to the tight-binding calculations led
to the ® rst experimental demonstration of the trigonal warping e€ ect in metallic
nanotubes [34], consistent with a systematic theoretical study [32] of this phenomena
718 M. S. Dresselhaus and P. C. Eklund

in both metallic and semiconducting nanotubes. This topic awaits further systematic
experimental studies.
A second method for determining the interband transitions Eii …dt† comes from
optical spectra, where the measurements are made on ropes of single-wall carbon
nanotubes, so that appropriate corrections should be made for inter-tube inter-
actions in interpreting the experimental data. Optical transmission spectra were
taken for single-wall nanotubes synthesized using four di€ erent catalysts [29, 35],
namely NiY (1.24± 1.58 nm), NiCo (1.06 ± 1.45 nm), Ni (1.06 ± 1.45 nm) and RhPd
(0.68± 1.00 nm), where the range in dt for each catalyst is indicated [29, 35]. For the
NiY catalyst, three large absorption peaks were observed at 0.68, 1.2 and 1.7 eV,
S
yielding a value of ® 0 ˆ 3:0 § 0:2 eV, using the approximate relations E11 ’
M
2® 0 aC¡C =dt and E11 ’ 6® 0 aC¡C =dt , for semiconducting and metallic nanotubes,
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

respectively. These approximate relations denote the centre of the width of the
Eii …dt † transition energies in ® gure 8 at constant dt [32]. Optical spectra were also
reported for nanotubes produced with the NiCo, Ni and RhPd catalysts, but the
peak values for the absorption bands were not explicitly quoted [29, 35, 36]. In
interpreting the optical transmission data, corrections for the diameter distribution
of the nanotubes, for trigonal warping e€ ects and for the non-linear k dependence of
E …k† away from the K-point in the Brillouin zone need to be considered [30, 32]. The
calculated Eii …dt † transition energies in ® gure 8 provide a ® rm basis for a detailed
analysis of the optical data [29, 35].
Optical absorption measurements on bromine and caesium-doped single-wall
nanotubes (1:24 < dt < 1:58 nm) [37] showed an upshift of EF on doping with Cs and
a downshift of EF on doping with Br2 due to charge transfer e€ ects between the
dopants and the nanotubes. From observation of the reduction in the optical
absorption intensity upon doping, the shift of EF with Br2 doping could be
monitored, showing that at full Br2 saturation concentration, EF had decreased
su ciently so that optical transitions from an occupied to an unoccupied state could
M
only occur for Elaser ¶ 2 eV, which lies above the metallic window for the E11 …dt†
transition for the NiY catalysed sample used in the experiments. Whereas one report
based on resonant Raman spectra for semiconducting nanotubes [38] claimed
reversibility in the spectra upon iodine doping and undoping, another study based
on resonant Raman spectra for metallic nanotubes upon bromine doping and
undoping claimed some irreversible behaviour [39, 40]. Further studies as a function
of Elaser and dt are needed to account for the connection between these two
observations [38± 40] (see section 4.10).

3. Phonon modes
The phonon dispersion relations in a carbon nanotube can be obtained from
those of the 2D graphene sheet by using the same zone folding approach [41± 43] as
was used to ® nd the 1D electronic dispersion relations [2, 44]. Zone folding and
symmetry-base d force constant models were used by some authors [41, 43, 45, 46] for
calculations of the phonon dispersion relations, tight binding calculations by other
authors [47± 49], and some ab initio calculations [50, 51] were also reported. Because
of the very weak interplanar interactions, the phonon dispersion relations for
graphite in the basal plane (see ® gure 10) provide a good ® rst approximation for
the 2D phonon dispersion relations of an isolated graphite plane, which is called a
graphene sheet. The calculated phonon dispersion curves of ® gure 10 were ® t to the
Phonons in carbon nanotubes 719
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

Figure 10. The phonon dispersion relations for graphite plotted along high-symmetry in-
plane directions. Experimental points from neutron scattering and electron energy
loss spectra were used to obtain values for the force constants (see table 1) and to
determine the phonon dispersion relations throughout the Brillouin zone [41].

Table 1. Force constant parameters for 2D graphite in units of


104 dyn cm ¡1 [41]. Here the subscripts r, ti and to refer to
radial (bond stretching), transverse in-plane and trans-
verse out-of-plane (bond bending) f orce constants,
respectively (see ® gures 10 and 11).

Radial Tangential

1
…† 1
…† 1
…†
¿r ˆ 36:50 ¿ti ˆ 24:50 ¿to ˆ 9:82
…2† …2† …2†
¿r ˆ 8:80 ¿ti ˆ ¡3:23 ¿to ˆ ¡0:40
…3† …3† 3
…†
¿r ˆ 3:00 ¿ti ˆ ¡5:25 ¿to ˆ 0:15
…4† …4† …4†
¿r ˆ ¡1:92 ¿ti ˆ 2:29 ¿to ˆ ¡0:58

experimental points obtained by electron energy loss spectroscopy, inelastic neutron


scattering, velocity of sound and other techniques [41, 52, 53]. The inclusion of force
constants taking into account fourth-neighbou r interaction terms (see table 1) have
been su cient for reproducing the experimental data for graphite in ® gure 10.
The three phonon dispersion curves (or branches), which originate from the G-
point of the Brillouin zone with ! ˆ 0 (see ® gure 10), correspond to acoustic modes:
an out-of-plane mode, an in-plane tangential (bond-bending) mode and an in-plane
radial (bond-stretching ) mode, listed in order of increasing energy, respectively. The
remaining three branches correspond to optical modes: one non-degenerate out-of-
plane mode and two in-plane modes that remain degenerate as we move away from
k ˆ 0.
It is noted that the out-of-plane (transverse) acoustic branch for a graphene sheet
shows a special k 2 energy dispersion relation around the G-point, while the other two
720 M. S. Dresselhaus and P. C. Eklund
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

Figure 11. The phonon density of states versus phonon energy for a 2D graphene sheet in
units of states/C-atom/cm ¡1 £ 10 ¡2 [41].

in-plane acoustic branches show a linear k dependence, as is normally seen for


acoustic modes. The optical out-of-plane transverse branch (at ! ¹ 865 cm ¡1 at the G
point) also shows a k 2 dependence. Because of this special k 2 dependence of !…k† for
the out-of-plane acoustic branch near the G-point, there is neither a phase velocity
nor a group velocity for the z component for these acoustic vibrations.
At low frequencies the in-plane acoustic branches for a graphene sheet give rise to
a density of states g…!† linear in !, while the out-of-plane k 2 acoustic branch gives
rise to a small constant contribution (independent of !) to the density of states,
which is signi® cant up to ¹400 cm ¡1 and a€ ects the thermal properties (see section 5)
up to ¹50 K, as shown in ® gures 10 and 11.

3.1. Phonon dispersion relations for nanotubes


As a ® rst approximation, the phonon dispersion relations for an isolated single-
wall carbon nanotube can be determined by zone folding the phonon dispersion
curves !m2D …k† of a two-dimensional graphene layer (see ® gure 1), where m ˆ 1;. . . ;6
labels the 3 acoustic and 3 optic modes and k is a vector in the layer plane. Since
there are 2N carbon atoms in this unit cell (see equation (10)), we will have N pairs of
bonding p and anti-bonding p ¤ electronic energy bands. Similarly the phonon
dispersion relations will consist of 6N branches resulting from a vector displacement
of each carbon atom in the nanotube unit cell.
The phonon dispersion relations of a carbon nanotube depend on the indices
…n;m † or equivalently on the diameter and chiral angle of the carbon nanotube, dt
and ³, since the phonon wave vector in the circumferential direction becomes discrete
and is described by each K1 vector (see equation (12)), in accordance with the
periodic boundary conditions of the chiral vector Ch .
In the context of zone folding, the one-dimensional phonon energy dispersion

relations for !1D …k† for SWNTs are related to the !m 2D …k† by:
Phonons in carbon nanotubes 721
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

Figure 12. (a) The calculated phonon dispersion relations of an armchair carbon nanotube
with Ch ˆ …10; 10†. The number of degrees of freedom is 120 and the number of
distinct phonon branches is 66. (b) The corresponding phonon density of states for a
(10, 10) nanotube [2]. (c) A comparison between the phonon density of states g1D …!†
for a (10, 10) nanotube shown as the solid curve and g2D …!† for a graphene sheet
shown by the points [31].

† … m· ˆ 0;.1;.....;N;6 ;¡ 1;
ˆ p p

!1D …k † ˆ !m …
2D k
K2
jK2 j
‡ ·K1 ; and ¡
T
<kµ
T
;†
…14†
where k is a one-dimensional wave vector, K2 is the reciprocal lattice vector along
the nanotube axis (see equation (12)), K1 is the reciprocal lattice vector in the
circumferential direction (see equation (12)) and T is the magnitude of the one-
dimensional translation vector T given in equation (9).

The zone folding procedure yields the appropriate one-dimensional !1D …k† for
almost all the phonon branches of a carbon nanotube. An example of the phonon
branches for an isolated SWNT is shown in ® gure 12 for a (10, 10) nanotube. In this
® gure T denotes the magnitude of the basis vector along the nanotube axis (see
equation (9)). For the 2N ˆ 40 carbon atoms per circumferential strip for the (10,
10) nanotube, there are 120 vibrational degrees of freedom, but because of mode
degeneracies there are only 66 distinct phonon branches, of which 12 modes are non-
degenerate and 54 are doubly degenerate.
722 M. S. Dresselhaus and P. C. Eklund

(a)

(b)
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

Figure 13. (a) The out-of-plane tangential acoustic modes at k ˆ 0 (left) in a single layer of
graphite give rise to a radial breathing mode in the carbon nanotube with non-zero
frequency (right). (b) An acoustic mode of a carbon nanotube whose vibration is
perpendicular to the nanotube axis (right) corresponds to a linear combination of
both in-plane and out-of-plane graphite-derived modes (left). These modes do not
couple in the case of a single graphite layer, but do couple for the nanotube because
of the curvature that is introduced by rolling up the graphene sheet [2].

In ® gure 12 (b) we show the corresponding phonon density of states for


the (10,10) nanotube in units of states per C atom per cm ¡1 . When we integrate
the phonon density of states with respect to the energy, we get 3 states/C-atom as the
total number of states. Since we use the same units for the phonon density of states
for nanotubes and for 2D graphite, we can directly compare the phonon density of
states for the (10,10) nanotube (® gure 12 (b)) and for 2D graphite (see ® gure 11). This
comparison given in ® gure 12 (c) [31] shows that the phonon density of states for the
(10,10) nanotube is close to that for 2D graphite, since the phonon dispersion
relations are, in principle, given by the zone-folding of those for 2D graphite. The
di€ erences in the nanotube phonon density of states relative to that for 2D graphite
pertain to the one-dimensional van Hove singularities for the optical phonon
subbands and to the 4 acoustic modes of the nanotubes and their special properties
at low ! discussed below.
However, zone-folding of the graphene phonon branches does not always give
the correct dispersion relation for a carbon nanotube [2, 41], especially in the low
frequency region, and some additional physical concepts must be introduced. For
example, when the out-of-plane tangential acoustic (TA) modes of a graphene sheet
shown in ® gure 13 (a) on the left are rolled into a nanotube as shown on the right, the
radial breathing mode is formed and the resulting vibration does not have ! ! 0 as
k ! 0. Therefore the radial breathing mode is not an acoustic mode, but rather is an
optical mode with a non-zero frequency at k ˆ 0. For the radial breathing mode
shown in ® gure 13 (a) on the right [41], both the graphite radial force constant and
the in-plane tangential force constant in the circumferential direction of the
nanotube are related to the radial breathing normal mode vibration. This results
in a ® nite frequency !0RBM for the radial breathing mode at k ˆ 0, where the
superscript denotes the mode frequency for an isolated SWNT, leaving !RBM to
denote the mode frequency in the presence of inter-tube interactions which occur in
SWNTs found in nanotube bundles (see section 3.3). Since there is no vibration in
Phonons in carbon nanotubes 723

20

15
E (meV)

10

Phonon DOS
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

0 4 8 12
E (meV)

0
0 0.1 0.2 0.3 0.4
k(1/Å )
Figure 14. Low-energy phonon dispersion relations for a (10, 10) nanotube. There are four
acoustic modes: two degenerate TA modes (v ˆ 9 km s¡1 ), a `twist’ mode
(v ˆ 15 km s ¡1 ) and one TA mode (v ˆ 24 km s¡1 ). The inset shows the low-energy
phonon density of states of the nanotube (solid line) and that of graphite (dashed
line) and graphene (dot-dashed line). The nanotube phonon DOS is constant below
2.5 meV, then increases stepwise as higher subbands enter; there is a 1D singularity at
each subband edge [2, 54].

the direction of the nanotube axis for the breathing mode, the bond angle of the
hexagon network is unchanged.
While the acoustic vibrations of a carbon nanotube in the longitudinal direction
correspond to acoustic vibrations in the 2D graphene sheet, the two acoustic modes
in the directions perpendicular to the nanotube axis, do not directly correspond to
any two-dimensional graphene phonon modes. In a graphene sheet, the in-plane and
out-of-plane modes are decoupled from each other. However, when the graphene
strip is rolled up into a nanotube, the graphite-derived in-plane and out-of-plane
modes do couple to each other, as shown on the left-hand side of ® gure 13 (b), to
form the acoustic mode of the nanotube shown on the right. To avoid di culties
associated with lack of compatibility between the TA modes in a graphene sheet with
the TA modes and the radial breathing mode of SWNTs, the three-dimensional
carbon nanotube dynamical matrix for a given nanotube described by …n;m † has
been solved in the one-dimensional Brillouin zone of the nanotube [2].
The resulting lowest energy modes are shown in ® gure 14 for a (10,10) nanotube
where the 4 acoustic modes appropriate to SWNTs are displayed. Here we see the
transverse acoustic (TA) modes, which are doubly degenerate, and have x and y
vibrations perpendicular to the nanotube (z) axis (see ® gure 13 (b) right). The highest
energy acoustic mode in ® gure 14 is the longitudinal acoustic (LA) mode with
vibrations in the direction of the nanotube axis. Since the displacements of these
three acoustic nanotube modes are three dimensional, all three phonon dispersion
relations are of the form ! ˆ vk, where v is the velocity of sound appropriate for that
acoustic branch. The sound velocities of the TA and LA phonons for a (10,10)
724 M. S. Dresselhaus and P. C. Eklund

(a) (b)

E2g 17 cm-1 E1g 118 cm-1


(c) (d) (e)
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

A1g 165 cm-1 E2g 368 cm-1 E1g 1585 cm-1

(f) (g)

A1g 1587 cm-1 E2g 1591 cm-1


Figure 15. The calculated Raman mode atomic displacements, frequencies and symmetries
for selected normal modes for the (10, 10) nanotube modes. The symmetry and the
frequencies for these modes are not strongly dependent on the chirality of the
nanotube. In the ® gure, we show the displacements for only one of the two modes in
the doubly degenerate E1g and E2g modes [2].

armchair carbon nanotube are estimated as vTA ˆ 9:43 £ 103 m s¡1 and vLA ˆ
20:35 £ 10 3 m s¡1 , respectively, from the measured Young’s modulus, Y , from which
the LA phonon velocity is estimated by v ˆ …Y =»†1=2 , in which » is the density of
carbon atoms taken as » ˆ 1:28 £ 103 kg m ¡3 and Y ˆ 532 GPa. Ab initio calcula-
tions [51, 55, 56] have shown that the Young’ s modulus and Poisson ratio for typical
SWNTs are essentially the same as for a graphene layer.
In addition, there is a fourth acoustic mode with ! ˆ 0 at k ˆ 0 for isolated
single-wall carbon nanotubes, and this mode is related to a rigid rotation around the
nanotube axis at k ˆ 0 [28, 57]. Since the driving force for this wave motion is a
twisting motion of the nanotube, this mode is called the twisting mode (TW). The
sound velocity of the TW mode is taken to be vTW ˆ 15:00 £ 103 m s¡1 , which is the
same as the calculated value of the in-plane TA mode for 2D graphite, since
the displacements associated with the TW mode are in the cylindrical plane,
perpendicular to the nanotube axis. Also shown in ® gure 14 are the lowest optical
subbands obtained by the zone folding procedure for the (10,10) armchair carbon
nanotube, including an E2g mode at ¹17 cm ¡1 , an E1g mode at ¹118 cm ¡1 and the
radial breathing A1g mode at ¹165 cm ¡1 for k ˆ 0 (see ® gure 15). The diagram
(® gure 14) shows modes of like symmetry which couple to each other and show anti-
crossing behaviour, while branches with di€ erent symmetries do not interact and can
simply cross.
Phonons in carbon nanotubes 725

From the phonon dispersion relations in ® gure 12 (a) for the SWNTs, we obtain
the phonon density of states g1D …!† shown in ® gure 12 (b) and the comparison
between g1D …!† and g2D …!† shown in ® gure 12 (c). Of particular interest is the
behaviour of g1D …!† for the isolated SWNTs (single-wall nanotubes) in the limit of
small ! where the acoustic modes and the lowest phonon subbands are dominant.
The four phonon branches which follow the dispersion relations ! ˆ vk contribute a
constant term to the density of states, which is proportional to the inverse of the
velocity of sound for that branch. Additional contributions to g1D …!† for the isolated
SWNTs arise from the low energy optical phonons from the phonon subbands
=2
associated with the low dimensionality, which contribute to g1D …!† as a …! ¡ !0 †¡1
van Hove singularity at their subband edge frequency !0 . This g1D …!† is in contrast
to the g2D …!† for the graphene sheet for which the ! ˆ vk branches give rise to a
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

contribution to g2D …!† that is linear in !, while the out-of-plane ! / k 2 branch


contributes to the phonon density of states g2D …!† a small term that is constant, and
independent of !.
Other than the establishment of acoustic modes of the type shown in ® gure 14, the
e€ ect of the nanotube curvature on the in-plane and out-of-plane radial and
tangential force constants has been considered by various authors [2, 30, 31, 40,
51, 58], showing these corrections to be relatively minor, but not negligible for the
smallest diameter nanotubes. Since the magnitude of the curvature is not expected to
a€ ect the form of !…k† for the four acoustic branches in ® gure 14 very much, no
major e€ ect of curvature on the 1D density of phonon states g1D …!† is expected.
Interactions between adjacent SWNTs in a nanotube bundle (see section 3.3) are
expected to give rise to a dispersion relation with a ® nite frequency at k ˆ 0 and a k 2
dependence for the twist mode [59]. The contribution of such an inter-tube inter-
action term would result in a feature with a van Hove-type singularity in the phonon
density of states such as is introduced by a phonon subband.

3.2. Raman and infrared active modes of carbon nanotube s


The special symmetry properties of 1D carbon nanotubes results in only a few
Raman-active and infrared-active vibrational modes, as described in this section.
Among the 6N calculated phonon dispersion relations for carbon nanotubes whose
unit cell contains 2N carbon atoms, only a few modes are Raman or infrared (IR)
active, as speci® ed by the symmetry of the phonon modes. Since only k vectors very
close to k ˆ 0 are coupled to the incident light because of the energy-momentum
conservation requirements for the photons and phonons, we need only consider the
symmetry of the nanotube zone-centre vibrations at the G-point …k ˆ 0†. Point group
theory of the unit cell, predicts the number of Raman-active modes and IR-active
modes and their symmetry types [2].
The numbers of the Raman-active (A1g , E1g , E2g symmetries) and IR-active (A2u ,
E1u symmetries) modes for the nanotubes can be predicted by group theory, once the
lattice structure and its symmetry are speci® ed. We list in table 2 the number and
symmetries of the Raman-active lattice modes for all the possible di€ erent types of
carbon nanotubes.
More speci® cally, we list below the number of modes with their corresponding
symmetry types for all types of nanotubes. For armchair nanotubes (with even
n ˆ 2j having D nh symmetry) we write:
726 M. S. Dresselhaus and P. C. Eklund

Table 2. Number and symmetries of Raman-active modes for different types of carbon
nanotubes.

Nanotube structure Point group Raman-active modes IR-active modes

armchair (n; n) n even D nh 4A1g ‡ 4E1g ‡ 8E2g A2u ‡ 7E1u


armchair (n; n) n odd D nd 3A1g ‡ 6E1g ‡ 6E2g 2A2u ‡ 5E1u
zigzag (n; 0) n even D nh 3A1g ‡ 6E1g ‡ 6E2g 2A2u ‡ 5E1u
zigzag (n; 0) n odd D nd 3A1g ‡ 6E1g ‡ 6E2g 2A2u ‡ 5E1u
chiral (n; m) n 6ˆ m 6ˆ 0 CN 4A ‡ 5E1 ‡ 6E2 4A ‡ 5E1

Gvib ˆ 4A1g ‡ 2A1u ‡ 4A2g ‡ 2A2u ‡ 2B1g


Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

2j

‡ 4B1u ‡ 2B2g ‡ 4B2u ‡ 4E1g ‡ 8E1u ‡ 8E2g


‡ 4E2u ‡ ¢ ¢ ¢ ‡ 4E… j¡1†g ‡ 8E… j¡1†u : …15†
In equation (15), we assume that j is even. If j is odd [such as for …n;m † ˆ …6;6†], the
4 and 8 are interchanged in the last two terms in equation (15). For zigzag nanotubes
(with even n ˆ 2j having D nh symmetry) we write:

Gvib
2j ˆ 3A1g ‡ 3A1u ‡ 3A2g ‡ 3A2u

‡ 3B1g ‡ 3B1u ‡ 3B2g ‡ 3B2u


‡ 6E1g ‡ 6E1u ‡ 6E2g ‡ 6E2u
‡ ¢ ¢ ¢ ‡ 6E… j¡1†g ‡ 6E… j¡1†u : …16†
For armchair and zigzag nanotubes (with odd n ˆ 2j ‡ 1 having D nd symmetry) we
write:

Gvib
2j‡ 1 ˆ 3A1g ‡ 3A1u ‡ 3A2g ‡ 3A2u

‡ 6E1g ‡ 6E1u ‡ 6E2g ‡ 6E2u


‡ ¢ ¢ ¢ ‡ 6Ejg ‡ 6Eju : …17†
And ® nally for chiral nanotubes we write:
Gvib
N
ˆ 6A ‡ 6B ‡ 6E1 ‡ 6E2 ‡ ¢ ¢ ¢ ‡ 6E…N=2†¡1 : …18†
Using the information in equations (15) ± (18), we obtain the information
summarized in table 2 for the number and symmetry type of the Raman and IR
active modes for achiral (i.e. armchair and zigzag) and chiral nanotubes. Among the
achiral nanotubes, only armchair nanotubes with even number indexes …2j ;2j† have
di€ erent numbers of Raman and IR modes from the others, such as armchair
nanotubes with odd number indexes …2j ‡ 1;2j ‡ 1† or zigzag nanotubes …n;0†. From
table 2 we deduce the remarkable Raman and infrared selection rules for SWNTs,
that the numbers of Raman and infrared-active modes do not depend on the
nanotube diameter and chirality, though the total number of ® nite frequency phonon
modes …6N ¡ 4† is very di€ erent for di€ erent chiralities and diameters. Group theory
selection rules indicate that there are only 15 or 16 Raman-active modes and 6 to 9
IR-active modes for a single-wall carbon nanotube, despite the large number of
Phonons in carbon nanotubes 727
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

Figure 16. The armchair index n versus mode frequency for the Raman-active modes of
single-wall armchair …n; n† carbon nanotubes [28]. The nanotube diameter can be
found from n using equation (2).

vibrational modes. Even though group theory may indicate that a particular mode is
Raman-active, this mode may nevertheless only have a small Raman cross-section.
In fact, we have only six or seven intense Raman-active modes for any nanotube
chirality.
Since the number of Raman-active and infrared-active modes for a given
symmetry category is independent of nanotube diameter, the dependence of a
particular vibrational mode on nanotube diameter can be investigated. Several of
the mode frequencies and their Raman cross-sections are found to be highly sensitive
to the nanotube diameter, while others are not. Figure 16 shows the dependence of
the frequency of the Raman-active modes on the nanotube diameter for armchair
nanotubes, expressed in terms of their …n;n† indices [60]. Here it is seen that the A1g
mode, which occurs at about 165 cm ¡1 for an isolated (10,10) nanotube, is strongly
dependent on nanotube diameter, while the modes near 1580 cm ¡1 are not. A
diagram similar to ® gure 16 can be constructed for the infrared-active modes for
the armchair nanotubes, and also for Raman and infrared-active modes for zigzag
and chiral nanotubes [1, 60].
Another factor which simpli® es the Raman spectra for SWNTs is the low
intensity of many of the Raman-active modes in ® gure 16. For example, ® gure 15
shows only the particular Raman-active modes that are expected to have signi® cant
intensity, based on calculations involving a bond polarizability model [2, 28].
The normal mode displacements for the seven Raman modes for a (10,10)
nanotube which have a relatively large Raman intensity are shown in ® gure 15.
The honeycomb lattice shown in ® gure 1 can be described in terms of two sublattices
consisting of A and B atoms. In the higher frequency Raman-active A1g mode
(1587 cm ¡1 in ® gure 15), the A and B atoms move in opposite directions (out-of-
phase) in the unit cell, while in the lower frequency Raman-active A1g mode
728 M. S. Dresselhaus and P. C. Eklund

1000
E2g

A1g
[ cm ] E1g
-1

100
E2g
w
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

10
1 10
r[ ]
Figure 17. Log± log plot of the lower Raman mode frequencies (below 500 cm¡1 for (10, 10)
nanotubes), as a function of carbon nanotube radius r ˆ dt =2 [2, 43].

(165 cm ¡1 in ® gure 15), the A and B atoms move in the same direction (in-phase). It
is clear from ® gures 15 (e) to (g), that the higher-frequency modes are out-of-phase
between nearest neighbour carbon atoms, while the lower-frequency modes of
® gures 15 (a) to (d ) show in-phase motion. The out-of-phase motions observed in
® gures 15 (e) to (g) are similar to the motion of the Raman-active E2g mode of
graphite at 1582 cm ¡1 , which corresponds to C¡C bond stretching motions for one
of the three nearest neighbour bonds in the unit cell.
The basic motions of the atomic displacements of the normal modes shown in
® gure 15 are independent of the chirality of the nanotube [2]. The normal mode
displacements illustrated in ® gure 15 show that the A1g , E1g and E2g modes have
zero, two and four nodes around the nanotube z axis, respectively (see ® gures 15 (e),
( f ) and (g) for out-of-phase motion, and (a), (b) and (c) for in-phase motion). These
normal mode patterns are very useful for the interpretation of the observed Raman
spectra, especially with regard to polarization phenomena (see section 4.7).
For the lower frequency Raman-active modes (below 500 cm ¡1 for (10,10)
nanotubes), the mode frequencies ! shift systematically with increasing diameter,
as shown in the log± log plot of ! in ® gure 17 as a function of the carbon nanotube
radius r ˆ dt =2 for …n;m † in the range …8 µ n µ 10; 0 µ m µ n†. Figure 17 clearly
shows straight line dependences of log ! on log dt for all four low frequency Raman
modes on this log± log plot, thus indicating a power law dependence of !…dt† on dt .
No signi® cant chirality dependence is found for the mode frequencies for these
modes, which is consistent with the fact that the energy gap of a semiconducting
nanotube and the strain energy depend primarily on the nanotube radius and are
only weakly dependent on the chiral angle ³ [61, 62]. From the slopes of !…dt† for this
range of dt (see ® gure 17), we conclude that, except for the lowest E2g mode, the
mode frequencies are approximately proportional to 1 =dt . The frequency !2g …dt † of
the lowest E2g mode has a predicted dependence of dt¡1:95§0:03 , which is approxi-
Phonons in carbon nanotubes 729

mately quadratic (i.e. 1 =dt2 ). The power law predicted for the A1g radial breathing
mode frequency !0RBM …dt† for an isolated SWNT valid in the range 0.6 nm µ dt µ
1.4 nm is
1:0017§0:0007
d…10;10†
!0RBM …dt † ˆ !0…10;10† ; …19†
dt

which is very close to linear (i.e. 1=dt ) and this relation is used by experimentalists as
a secondary characterization tool for the diameter distribution in isolated SWNT
samples. When the SWNTs are in bundles, inter-tube interactions become important
(see section 3.3 and section 4.1) and modify the dispersion relation for !RBM …dt †. In
equation (19) !0…10 ;10† and d…10 ;10† are, respectively, the mode frequency and diameter
of isolated (10,10) armchair nanotubes, which have been given by !0…10 ;10†= 169 cm ¡1
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

and d…10;10†= 1.375 nm [2] and are further discussed in section 4.1, in connection with
radial breathing mode measurements.

3.3. Inter-tube or intra-bundle interactions


To date, almost all the experimental data on SWNTs, and especially Raman
spectra, have been collected on bundles of nanotubes that are produced either in an
electric arc (EA) or by the pulsed laser vaporization (PLV) method. High resolution
TEM and X-ray di€ raction studies on these nanotube bundles show that they
contain on the order of 100 well-aligned nanotubes in a close-packed triangular (or
honeycomb) lattice [6]. The intra-bundle or inter-tube interactions that arise in this
lattice are usually assumed to be weak and are approximated by a van der Waals
interaction, similar to the coupling between adjacent graphene layers in 3D crystal-
line graphite. However, the e€ ects of this coupling on the vibrational [63± 65] and
electronic states [66 ± 69] have nevertheless been shown theoretically to be of su cient
strength to signi® cantly a€ ect the physical properties of SWNTs, as measured in the
laboratory.
The e€ ect of these interactions on the electronic structure has been calculated [66,
68] in the local density approximation. These calculations [66± 68], however, did not
consider a possible structural distortion, or hexagonal faceting of the nanotube
cross-section due to the inter-tube interactions, but the calculations did allow for
charge redistribution within a three-dimensional unit cell containing adjacent
nanotubes. For semiconducting and metallic nanotubes with diameters near that
of a (10,10) nanotube, a small perturbation is found in the energy di€ erence between
the mirror image singularities in the one-electron density of states (DOS) in the
valence and conduction bands (Eii …dt† in ® gure 8), and for metallic nanotubes, a
pseudo-gap ( ¹100 meV) was predicted to open in the DOS at the Fermi energy [66,
68, 69]. An absorption band at ¹120 cm ¡1 has been reported recently for a `bucky
paper’ sample made from PLV-derived nanotubes [70]. The authors cite this
observed absorption band as evidence for this predicted pseudo-gap, although the
observed gap is a factor of eight smaller [70] than theoretically predicted. Any shift in
the spacing between the 1D electronic singularities in the valence and conduction
bands (see section 2.2) will, of course, also a€ ect the quantitative interpretation of
the resonant Raman scattering spectra (see section 4). That is, the laser excitation
energy to promote resonant Raman scattering from a particular nanotube will be
shifted depending on whether that nanotube is isolated or is located in a nanotube
bundle [66, 71].
730 M. S. Dresselhaus and P. C. Eklund

The e€ ect of inter-tube interactions on the vibrational modes of SWNTs was ® rst
reported by Venkateswaran et al. [64] and this e€ ect was used to interpret the
pressure-induced shift in both the Raman-active tangential and radial breathing
modes (see section 4.4). They used a generalized tight-binding molecular dynamics
(GTBMD) scheme [72], including the e€ ects of an externally applied pressure, to
determine the structural relaxation, or faceting, within the bundle. Next, using the
coordinates obtained from this calculation, and an optimized set of van der Waals
interaction parameters between C atoms on neighbouring nanotubes, the e€ ect of
the inter-tube interactions and applied pressure on the vibrational spectrum was
calculated, as discussed in section 4.4. Here, we focus on the theoretical results
relevant to the corrections to the radial breathing mode (RBM) at zero applied
pressure due to inter-tube interactions [63, 64, 73, 74].
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

The GTBMD calculations [63] show that for bundles of nanotubes having
individual nanotubes with diameters in the range 0:7 < dt < 1:5 nm, the simple
1 =dt frequency dependence of the radial breathing mode frequency that was derived
for isolated nanotubes can be approximately amended by the addition of a nearly
constant 7% upshift, independent of dt in this range of dt [42, 63]. The frequency
upshift is the result of an additional tube-wall restoring force due to next-neighbour,
nanotube ± nanotube interactions. The relation between !RBM and dt is important for
the characterization of the diameter distribution of actual nanotube samples, since
Raman spectroscopy provides a quick and convenient technique for such sample
characterizations (see section 4.1).
The general features of the model calculations of this perturbation e€ ect [63]
starts with the calculation for the relaxed structure and the vibrational spectrum of
the (10,10) nanotube lattice at zero applied pressure [42]. The (10,10) nanotube
lattice with D 10h symmetry relaxes by a distortion of the circular nanotube symmetry
due to inter-tube interactions to a monoclinic system with space group P2 =m. By
putting the uncoupled (10,10) tubes on a triangular lattice associated with the
symmetry of the nanotube bundle, a formal lowering of the symmetry (orthorhombic
space-group Cmmm ) occurs relative to the D 10h symmetry of the individual
nanotubes. The resulting lowering of the symmetry to P2 =m symmetry, changes,
in principle, the Raman activity of the vibrational modes of the SWNTs. However,
since the inter-tube coupling is weak, it is not expected that many new Raman modes
will actually be observed, and the spectral intensity should still be dominated by the
selection rules for the isolated SWNT (see section 3.1). The calculations show that
the relaxed coupled nanotube lattice di€ ers only slightly from the lattice found for
uncoupled nanotubes, with lattice constants for the coupled lattice being a ˆ 16:68 A Ê
and b ˆ 16:72 A Ê (perpendicular to the nanotube axis) and c ˆ 2:453 A Ê (parallel to
the nanotube axis) and cell angles ¬ ˆ ­ ˆ 908 and ® ˆ 59:908, using the usual
notation for the monoclinic system. The shortest carbon± carbon inter-tube distance
is found to be relatively small (between 3.17 and 3.26 A Ê ) and this small distance is
attributed to the curved nanotube surface and the variation in inter-tube distance is
attributed to the missing 6-fold symmetry of the individual nanotubes. (Within the
same theoretical model, the shortest carbon± carbon inter-layer distance in graphite is
calculated to be 3.38 A Ê as compared to the experimental value of 3.35 A Ê [63].)
The e€ ect of the van der Waals inter-tube coupling on the vibrational spectrum
!…k† is displayed in ® gure 18 for (10, 10) SWNTs on a triangular lattice. As can be
seen by the presence of weakly negative ! eigenvalues in ® gure 18 (a), the relaxation
procedure did not result in an overall stable solution to the eigenvalue problem,
Phonons in carbon nanotubes 731
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

Figure 18. Phonon dispersion curves in the low frequency region for (a) the 3-dimensional
(10, 10) nanotube triangular lattice and (b) an isolated (10, 10) nanotube. G± A is the
k-space direction parallel to the nanotube axis, while the dispersion relations !…k† for
wave vectors perpendicular to the nanotube axis are shown in the G± M± K± G portion
of (a). Solid arrows point to the G-point frequency of the breathing modes in the
respective systems, while open arrows point to the E4g modes, as discussed in the text.
The carbon atom displacements corresponding to the modes labelled ¬, ­ and ® in
the dispersion curves are given in (c) [63].

insofar as some of the normal modes in the G± M ± K± G plane (wave vectors


perpendicular to the tube axis), appear as decaying modes (complex frequencies)
which are displayed in ® gure 18 (a) as modes with slightly negative frequencies. This
weak instability is a sign that the chosen set of basis atoms is too small to allow for a
full relaxation to a stable con® guration. By doubling the unit cell along an in-plane
lattice vector, part of the instabilities (around the M-point) were removed. However,
relaxing the structure with the doubled unit cell did not noticeably change (except for
a folding of the phonon branches) the dispersion relations for the modes with
frequencies >20 cm ¡1 that were obtained using the smaller unit cell. The authors [63]
therefore concluded that the main e€ ects of inter-tube coupling are well described
within the small set of basis atoms.
The inter-tube coupling was found to lead to additional dispersion in the plane
perpendicular to the nanotube axis, and this is shown along the G± M ± K± G
directions in ® gure 18 (a). The dispersion curves !…k† for the uncoupled-tube lattice,
which corresponds to the isolated nanotube discussed in section 3.1 and is shown for
732 M. S. Dresselhaus and P. C. Eklund

comparison in ® gure 18 (b), would be ¯ at (zero slope) in the region G± M ± K± G.


Compared to the uncoupled or isolated nanotube, the frequencies of some of the
modes for a nanotube within a bundle are shifted, and modes that were 2-fold
degenerate for the isolated nanotube are split for a nanotube in a bundle. At the K
or M points, where neighbouring nanotubes vibrate with opposite phases, the
frequencies of some of the modes are close to the G-point frequencies of the
corresponding vibrations in the isolated nanotube and this occurrence gives rise to
coupled modes. As expected, the in¯ uence of the inter-tube coupling on the
vibrational spectrum was found to diminish with increasing phonon frequency,
and this coupling is essentially negligible above ! ˆ 500 cm ¡1 . The lowest, non-zero
frequency mode at the G-point with ! º 10± 12 cm ¡1 is the librational mode (rigid
small angle rotations) of the individual nanotubes in the lattice, obtained within a
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

density functional theory framework [66, 68].


The A1g symmetry breathing mode for the individual nanotube is marked as ¬ in
® gure 18 (b) at its zone centre frequency value (which was found to be 156 cm ¡1 for
this calculation for a (10, 10) nanotube [63]). The van der Waals coupling between
the nanotubes was found to upshift this mode by ¹14 cm ¡1 (9%) as marked by ­ in
® gure 18 (a) and the corresponding normal mode displacements are shown in
® gure 18 (c). However, a frequency shift is not the only e€ ect that occurs when the
triangular lattice describing a nanotube bundle is formed. The lower symmetry of the
triangular crystalline lattice allows the A1g mode of the isolated nanotube to mix
with one partner of a doubly degenerate E4g pair of the normal modes for the (10,10)
nanotube in equation (15). As a result, an additional pair of modes is introduced,
(­ ;® ), which resemble the displacements of the breathing mode for the isolated
nanotube. The corresponding displacement patterns are shown in ® gure 18 (c). While
mode ® has larger displacement amplitudes in the region between two neighbouring
nanotubes, mode ­ shows larger displacements towards the (open) channels between
the isolated nanotubes and thus has the lower frequency. The ® mode is, therefore,
more strongly a€ ected by the van der Waals interaction. We further note that the
second partner of the E4g pair (open arrows in the region between ® gures 18 (a) and
(b)) is not a€ ected by the inter-tube coupling. According to these calculations [63] the
mixing of the A1g breathing mode with an E`g mode occurs for all …n; m † nanotube
lattices, with ` assuming the largest possible even numbered value for the speci® c
nanotube type. For example, from equation (15) we see that ` ˆ 4 for a (10, 10)
nanotube. However, calculations on larger nanotubes show that the amount of
mixing and the e€ ect of the inter-tube coupling decreases with increasing nanotube
diameter (see ® gure 19) [63]. Even though the basis atoms of the nanotube triangular
lattice relax slightly away from their corresponding positions in the individual
nanotube, the projections h­ j¬i and h® j¬i of the normal displacement vectors can
be de® ned in terms of the coupled modes projected onto the zero coupling breathing
mode. It is found that h­ j¬i2 ‡ h® j¬i2 º 1 with h­ j¬i2 ˆ 0:89. If it is assumed that
the Raman intensity of the breathing mode is carried over from the case of the
isolated nanotube to the case of the coupled nanotubes, then it was proposed that the
ratio h® j¬i2 =h­ j¬i2 corresponds roughly to the intensity ratio one expects for the two
Raman peaks for ® and ­ (up to a frequency factor !­ =!® ) and, in the case of
resonance Raman scattering, a small di€ erence in the resonance condition for ­ and
® is expected to occur. As stated above, the intensity ratio I …!­ †=I…!® † decreases with
increasing nanotube diameter (as shown in the inset to ® gure 19 (b)).
Phonons in carbon nanotubes 733

Frequency [cm -1]

o
dt [A]
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

(a)
o
tube diameter (dt ) [A]
Frequency [cm -1]

o
dt [A]

(b)
o
tube diameter (dt ) [A]
Figure 19. Nanotube diameter dependence of various radial breathing mode frequencies. (a)
Comparison between !0RBM …dt † for isolated nanotubes, !RBM…dt † for the same
nanotubes in the triangular lattice of a bundle, and the same nanotubes with an inter-
tube van der Waals coupling equal to that between adjacent graphene sheets in the
graphite lattice. The inset gives the upshift D ! …dt† of the radial breathing mode in the
bundle relative to the isolated nanotube as a function of dt . (b) Comparison between
the dt dependence of the ­ and ® modes in the nanotube bundle (see ® gure 18) and of
!E`g …dt† for the largest ` value possible for a given …n; m † nanotube (see section 3.2).
The inset gives the predicted Raman scattering intensity ratio of the ­ (solid line) and
of the ® (dashed line) radial breathing modes for the nanotubes in a nanotube bundle
as a function of dt [63].

The frequency shift and the mode mixing is found to be a general feature of the
breathing mode in these model calculations and to be dependent on the nanotube
diameter [63]. Figure 19 (a) shows the calculated mode frequency as a function of
nanotube diameter dt for an isolated (10, 10) nanotube in comparison to the same
nanotube in a triangular lattice of (10, 10) nanotubes in a nanotube bundle. The
dotted curve gives the results for a simple spring model calculation that uses the
radial breathing mode frequency for an isolated nanotube and an inter-tube coupling
734 M. S. Dresselhaus and P. C. Eklund

of 120 cm ¡1 which is set equal to the coupling energy between adjacent graphene
layers in graphite. The inset shows the frequency upshift D ! of the radial breathing
mode relative to that for an isolated (10, 10) nanotube as a function of nanotube
diameter dt . Also shown in ® gure 19 (a) is !RBM …dt † for an isolated (10, 10) nanotube
to which the van der Waals interaction potential appropriate to graphite is added. It
is seen that the graphite van der Waals potential underestimates the tube± tube
interaction for small dt but greatly overestimates the interaction for large dt .
Figure 19 (b) shows the nanotube diameter dependence of the frequencies of the
breathing modes ­ and ® for a variety of individual nanotubes and nanotube lattices
of the armchair, zigzag and chiral type. The indices of the respective nanotubes are
listed on top of the ® gure centred along the x axis at the value of the corresponding
diameter. With and without the inter-tube coupling, the !RBM for ­ and ® display
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

only a dependence on the diameter dt , not on the chirality of the nanotubes. The
diameter dependence of the frequency of the ­ and ® modes for the nanotube
bundles shown in ® gure 19 (b) indicates that the frequency di€ erence between the ­
and ® modes decreases as the nanotube diameter increases. The frequency of the ®
mode for small diameter nanotubes is close to that of the E`g mode, where ` is the
maximum allowed integer for the nanotube modes discussed above and in section 3.2.
The inset to ® gure 19 (b) gives the relative intensities of the ­ mode and the ® mode as
a function of nanotube diameter dt , and shows that for dt < 1:2 nm, it is only the ­
mode that should be observed, while for large diameter nanotubes dt ¶ 2:0 nm, the ­
and ® modes are expected to have equal intensity. Another calculation [74], based on
a pair potential approach, has con® rmed that an upshift in frequency is expected,
and agreement was obtained for the approximate magnitude of the upshift for a
(10,10) nanotube lying within a nanotube bundle. These authors [74] predict the
upshift to be small for small diameter nanotubes and larger as the nanotube diameter
increases.
The e€ ect of inter-tube interactions on the mode frequencies over a wide range of
nanotube diameters and chiralities has been calculated on the basis of a tight binding
formalism [65] showing very large e€ ects for the lowest frequency modes with radial
mode displacements, and much smaller e€ ects for high frequency modes with
predominantly tangential displacements. The largest e€ ects of the inter-tube inter-
action are predicted for the lowest frequency E2g mode (see ® gure 15).
In summary, the inter-tube interactions are expected to lower the symmetry of
the individual isolated SWNTs, which gives rise to mode frequency shifts associated
with distortion e€ ects so that the radial breathing mode may show some chirality
dependence. Line broadening e€ ects in the Raman spectra can arise from these inter-
tube interactions, because of the di€ erence in local environments between a
nanotube in the centre of the bundle and a nanotube at the edge of a nanotube
bundle.

4. Raman spectra of single-wall nanotubes


Raman spectroscopy provides a particularly valuable tool for examining the
mode frequencies of carbon nanotubes with speci® c diameters and thereby evaluat-
ing the merits of various theoretical models for the 1D phonon dispersion relations
for carbon nanotubes. The status of this research direction is summarized. Because
of the remarkable properties of SWNTs, the Raman spectra for these 1D systems
give rise to new physical phenomena, not seen previously in Raman spectroscopy.
Phonons in carbon nanotubes 735

The status of research along these lines is also reviewed. The use of Raman
spectroscopy to characterize nanotube samples in terms of the diameter distribution
of the nanotubes in the sample is reviewed in section 4.1, for studying the 1D electron
density of states in resonant Raman (section 4.2) experiments through the electron±
phonon coupling mechanism [28], and in selectively exciting resonances in metallic or
semiconducting nanotubes (section 4.2). The temperature dependence (section 4.3)
and the pressure dependence (section 4.4) of the Raman spectra are then reviewed,
followed by a comparison of the unique di€ erences between the Stokes and anti-
Stokes spectra (section 4.5) and how this comparison is used to elucidate the
di€ erences between metallic and semiconducting SWNTs, exploiting phenomena
linked to low dimensionality. The use of surface enhanced Raman spectroscopy
(SERS) to elucidate the Raman spectra of SWNTs and new aspects of the SERS
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

technique are reviewed in section 4.6. Polarization phenomena and spectral features
associated with K-point phonons are reviewed in sections 4.7 and 4.8, respectively,
while section 4.9 treats overtones and combination modes, and section 4.10 reviews
progress on the study of doped SWNTs.
Most of the early experiments on the vibrational spectra of carbon nanotubes
were carried out on multi-wall carbon nanotubes, which were too large in diameter
to observe detailed quantum e€ ects associated with the 1D electronic dispersion
relations and observed through the resonant Raman e€ ect [75, 76]. The ® rst Raman
study to show a clear signature for single-wall carbon nanotubes [77] was carried out
on samples containing only a small concentration of single-wall nanotubes, and
having a wide distribution of diameters and chiralities, so that only features pertinent
to phonon modes near 1580 cm ¡1 , which are only weakly dependent on nanotube
diameter, could be observed [77]. This early work successfully identi® ed character-
istic features in the Raman spectra for single-wall nanotubes because of the relatively
large scattering cross-section of the nanotubes due to a resonant Raman process
involving van Hove singularities in the 1D density of electronic states discussed in
section 2.2.
The ® rst de® nitive study of the Raman spectrum of single-wall carbon nanotubes
(SWNTs) was carried out on a sample containing ropes of single-wall carbon
nanotubes with a narrow diameter distribution in the 1.2± 1.4 nm range, and
prepared by a laser vaporization technique [6]. A Raman spectrum taken on this
sample at a 514.5 nm laser excitation wavelength is shown in ® gure 20 [28, 50, 78].
Prominent in this spectrum are a number of features near 1580 cm ¡1 (see inset to
® gure 20), and a strong feature at ¹186 cm ¡1 . From ® gure 15, we see that there are
three mode frequencies near 1580 cm ¡1 which have mode symmetries A1g , E1g and
E2g for armchair nanotubes, each mode frequency being almost independent of
nanotube diameter. A similar behaviour is found for the Raman band near
1580 cm ¡1 for zigzag and chiral nanotubes [41]. The atomic displacements associated
with the normal modes near ¹186 cm ¡1 and ¹1580 cm ¡1 for a (10, 10) nanotube are
shown in ® gure 15 [2].
In contrast to the high frequency band near 1580 cm ¡1 , the feature near
¹186 cm ¡1 , which is identi® ed with an A1g radial breathing mode (!RBM ), is strongly
dependent on the nanotube diameter, as shown in ® gure 16. Calculations [42, 43]
show high intensities for this radial breathing mode and for the tangential modes
(!tang ) near 1580 cm ¡1 , in agreement with experimental Raman spectra. The other
Raman-active modes (see ® gure 16) are predicted to have low Raman cross-sections
(see ® gure 21), also in agreement with experiment (® gure 20). Bond polarizability
736 M. S. Dresselhaus and P. C. Eklund
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

Figure 20. Experimental Raman spectrum taken with 514.5 nm (2.41 eV) laser excitation
from a sample consisting primarily of single-wall nanotube bundles with diameters dt
near that of the (10, 10) nanotube (dt ˆ 1:36 § 0:20 nm) [28].

calculations further predict that the relative intensities of the weaker Raman-active
features in the experimental spectra of ® gures 20 and 21 can be increased by making
measurements on carbon nanotubes of small length (e.g. 100 nm), small compared to
an optical wavelength for laser excitation [79]. The weak features in the Raman
spectrum in ® gure 20 at about 1350 cm ¡1 (associated with resonant Raman
scattering of phonons near the K-point in the 2D Brillouin zone) and at about
1740 cm ¡1 are discussed further in sections 4.8 and 4.9, respectively.
One-dimensional quantum e€ ects are observed in the Raman spectra of single-
wall carbon nanotubes through the resonant Raman enhancement e€ ect between
incident or scattered photons and the electronic transition between the van Hove
singularities in the 1D density of states in the valence and conduction bands of the
nanotubes (see section 2.2). This resonant Raman e€ ect can be seen experimentally
by measuring the Raman spectra at a number of laser excitation energies, as shown
in ® gure 22 [28].
By comparing the various Raman spectra in ® gure 22, which were taken at
di€ erent laser excitation energies Elaser , we see large di€ erences in the vibrational
frequencies and intensities of the strong A1g radial breathing mode, consistent with a
resonant Raman e€ ect involving nanotubes of di€ erent diameters. These experi-
mental observations [28] also provided the ® rst clear con® rmation for the theoretical
Phonons in carbon nanotubes 737
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

Figure 21. Raman spectra (top) of a rope of single-wall carbon nanotubes taken with
514.5 nm excitation at ¹2 W cm ¡2 . The features in the spectrum denoted by the
symbol `*’ are assigned to second-order Raman scattering. The four bottom panels
are the calculated Raman spectra (based on a bond polarizability model) for
armchair …n; n† nanotubes, n ˆ 8 to 11 and the strongest Raman-allowed features are
indicated by vertical bars. The arrows in the panels indicate the calculated positions
of the remaining weak, Raman-active modes [28].

predictions about the singularities in the 1D electronic density of states of carbon


nanotubes through study of both the radial breathing mode features (section 4.1)
and the tangential mode features (section 4.2). This con® rmation was soon corrob-
orated by a more direct measurement of the 1D electronic density of states by STM/
STS spectroscopy, as discussed in section 2.2 [13, 14].
Because of this strong resonant enhancement e€ ect, only a small concentration of
single-wall nanotubes in a sample containing other carbon forms can give rise to
spectral features showing the characteristic sharp doublet structure in the 1570±
1600 cm ¡1 range [77]. Resonant enhancement in the Raman scattering intensity from
carbon nanotubes occurs when the energy of the incident or the scattered photon
corresponds to a transition between the sharp features in the one-dimensional
electronic density of states of the carbon nanotubes, as shown in ® gure 7. The
resonant enhancement e€ ect is so strong that it has been possible to observe up to
® fth order Raman scattering in SWNTs (e.g. up to Raman shifts of 6885 cm ¡1 using
488 nm laser excitation) [80].
Since the energies of these sharp features in the 1D electronic density of states are
strongly dependent on the nanotube diameter, a change in the laser frequency brings
into resonance a di€ erent carbon nanotube with a di€ erent diameter that satis® es the
new resonance condition, as shown in ® gure 8. For example, the model calculation in
® gure 7 shows that the (10, 10) armchair nanotube would be expected to be resonant
738 M. S. Dresselhaus and P. C. Eklund
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

Figure 22. An experimental room temperature Raman spectra for puri® ed single-wall
carbon nanotubes excited at ® ve di€ erent laser excitation wavelengths. The laser
wavelength and power density for each spectrum are indicated, as are the vibrational
frequencies (in cm ¡1 ) [28]. The equivalent incident photon energies for the laser
excitation are : 1320 nm ! 0.94 eV; 1064 nm ! 1.17 eV; 780 nm ! 1.58 eV; 647.1 nm !
1.92 eV; 514.5 nm ! 2.41 eV.

at a laser frequency of 1.9 eV, while the (9, 9) nanotube would be resonant at 2.1 eV,
considering only the incident photon.

4.1. Radial breathing mode phenomena


Theoretical calculations have shown that the frequency !0RBM of the perfectly
symmetric radial breathing mode of an isolated single-wall carbon nanotube
(section 3.1) has a particularly simple dependence 1 =dt on the nanotube diameter,
as given by equation (19) [2, 41, 50, 51, 81]. This fact, coupled with its large
(resonantly enhanced) Raman cross-section for the radial breathing mode, the
sensitivity of !RBM to charge transfer and to tube± tube interactions makes the radial
breathing mode a valuable probe for the structure and properties of SWNTs and
SWNT-based materials.
Phonons in carbon nanotubes 739
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

Figure 23. Calculated dependence of the frequency of the A1g radial breathing mode on
nanotube diameter dt plotted on a log± log plot. This plot is for isolated single-wall
carbon nanotubes [81]. It has been shown theoretically that the tube± tube interactions
within a nanotube bundle upshift these radial breathing mode frequencies by ¹7±
10% for dt µ 1:5 nm [63].

For an isolated nanotube of any chirality …n;m †, the mode frequency !0RBM has
been shown theoretically to exhibit a simple inverse diameter relationship, i.e.
!0RBM ’ 224 =dt for !0RBM in cm ¡1 and dt in nm, where the proportionality factor is
somewhat sensitive to the details of the calculation [81]. The results of two such
calculations are shown in ® gure 17 and in ® gure 23. The plot in ® gure 23 includes
armchair, zigzag and chiral nanotubes with diameters in the range of current
experimental interest. A strong dependence of the mode frequency on nanotube
diameter is in fact found for all the Raman-active low frequency modes. However,
the actual measurements are usually made on nanotubes which are within a bundle
of nanotubes. As discussed in sections 3.3 and 4.4, this simple relationship for !0RBM
must be corrected for weak inter-tube interactions within a bundle to obtain the
measured mode frequency !RBM . Theoretical calculations predict that these inter-
actions are responsible for a 6 ± 21 cm ¡1 frequency upshift, depending on the details
of the calculation (see sections 3.3 and 4.4) [63, 64, 73, 74].
However, it should be realized that the resonant enhancement of Elaser with
electronic interband transitions (see section 2.2) for the radial breathing mode
(RBM) (and other modes as well) will fall o€ rapidly with increasing nanotube
diameter, and therefore this Raman mode is not as sensitive a probe for the case of
larger diameter nanotubes. For example, Raman studies [28, 82, 83] using several
excitation energies Elaser were carried out on samples containing bundles of
nanotubes with a bimodal distribution of nanotube diameters, having population
maxima in the diameter distribution near that of (10, 10) and (20, 20) nanotubes [82].
The Raman spectra did not show any evidence for the RBM feature of the (20, 20)
740 M. S. Dresselhaus and P. C. Eklund

nanotubes [82], consistent with the expectation that the cross-section for the RBM
falls o€ rapidly with increasing tube diameter. These larger diameter nanotubes were
produced by the coalescence of two smaller nanotubes in a bundle during a high
temperature heat treatment (HTT) ¹1400 8C in 1 atm. of hydrogen. HRTEM images
showed directly that ¹30% of the nanotubes in such a bundle were nearly doubled in
diameter. However, the nanotube perfection of these `coalesced’ nanotubes could not
be veri® ed, and it is possible that defects in the nanotube walls might have broadened
the radial breathing mode line to the point that it could not be observed [82].
State-of-the-ar t Raman samples have a narrow distribution of diameters and
chiralities, which depend sensitively on the catalysts that are used in the synthesis
and the growth conditions, especially the growth temperature T g [9, 84± 86]. For
example, when 1.2 wt% of Ni/Co catalyst is used in the sample synthesis at a
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

temperature of 1150¯ C in 500 torr Ar, the diameter range is 1.0± 1.4 nm, while thinner
nanotubes with diameters of 0.8± 1.0 nm are obtained in the case of a 2.4% Rh/Pd
catalyst with growth at 1100¯ C [87]. Even when the same catalysts are used, a higher
growth temperature gives a larger diameter. At a higher temperature, however,
atomic vibrations prevent the formation of relatively unstable pentagonal rings,
compared with hexagonal rings, so the lower growth temperatures favour the growth
of smaller diameter nanotubes [2].
The approximate !0RBM ¹ 1=dt relationship has been used to estimate the range
of SWNT diameters in nanotube bundles produced by the pulsed laser vaporization
(PLV) process [81], by measuring the e€ ect of the growth conditions and especially of
the average temperature T g in the growth zone on the tube diameter distribution and
on the number of nanotubes per bundle. Results from measurements of the low
frequency Raman spectra collected on SWNT mats grown by pulsed laser vaporiza-
tion (PLV) at various furnace temperatures T g are displayed in ® gure 24. The spectra
(a)± (d ) were recorded using 1064 nm (Elaser ˆ 1:17 eV) excitation, and therefore the
T g -dependence of the diameter distribution of semiconducting nanotubes is explored
in this ® gure. As can be seen, with increasing growth temperature, the sharp peaks
identi® ed with radial breathing modes from nanotubes with various diameters shift
to lower frequency as T g is increased, indicating that the nanotube diameter
distribution shifts toward larger values with increasing temperature T g in the growth
zone. This shift in dt was con® rmed using HRTEM and X-ray di€ raction
characterization measurements [81]. The spectra (e), ( f ) and (g) in ® gure 24, taken
at several values of Elaser (488 nm (2.54 eV) ; 514.5 nm (2.41 eV) ; 647 nm (1.92 eV) ; and
1064 nm (1.17 eV)), provide a characterization of the nanotube diameter distribution
contained in the sample prepared at T g ˆ 1000 8C [81].
More precise use of the radial breathing mode (RBM) frequency to probe the
properties, or diameter distribution, of SWNTs embedded in bundles of SWNTs
requires that the inter-tube interactions be considered [64, 73]. The expression linking
the radial breathing mode !RBM to the nanotube diameter dt has been reported to be
well approximated by the expression

!RBM ˆ D !RBM ‡ !0RBM ˆ D !RBM ‡ !0…10;10†d…10 ;10† =dt ; …20†

where !0RBM is the radial breathing mode frequency for an isolated SWNT, D !RBM is
a frequency upshift which is a constant (see ® gure 19 (a)) for nanotube diameters
near to that of a (10, 10) armchair nanotube d…10 ;10†, and !0…10;10† is the radial
breathing mode frequency of an isolated (10, 10) nanotube. The calculated values of
Phonons in carbon nanotubes 741
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

Figure 24. Room temperature RBM spectra for bundles of SWNTs produced by pulsed
laser vaporization using an Fe/Ni catalyst in a carbon target. Spectra (a)± (d ) are
collected at ® xed laser excitation energy (1.17 eV; Nd:YAG) from samples grown at
T g ˆ 780, 860, 920 and 10008C, respectively. Note that the spectral weight shifts to
smaller RBM frequencies with increasing growth temperature, T g , indicating that dt
increases with increasing T g . The intensities and frequencies of the RBM bands in
spectra (e) ± (g) collected from the same sample (T g ˆ 10008C) but with di€ erent laser
excitation energies (488 nm (2.54 eV) ; 514.5 nm (2.41 eV) ; 647 nm (1.92 eV) ; and
1064 nm (1.17 eV)) are quite di€ erent, demonstrating how di€ erent diameter tubes are
excited as the excitation energy Elaser changes [81].

these parameters vary from one research group to another, but some typical values
are: D !RBM ˆ 14 cm ¡1 [64], 6.5 cm ¡1 [73] and 6 cm ¡1 [65], and !0…10 ;10†d…10;10† ˆ
224 cm ¡1 nm [64], 232 cm ¡1 nm [73] and 214 cm ¡1 nm [65]. Since these parameter
values were obtained by ® ts of the measured !RBM to equation (20) for a limited
range of nanotube diameters (0.7± 1.5 nm), the values of these parameters should be
considered in pairs. Interestingly, both sets of parameters listed above arrive at about
the same RBM frequency (equation (20)) for a (10, 10) nanotube in a bundle, i.e.
176 ± 177 cm ¡1 . Similar results, but with somewhat larger upshifts D !RBM , were also
obtained by Henrard et al. [74]. It has been found [42, 63, 74] that D !RBM decreases
signi® cantly with increasing nanotube diameter dt (see inset to ® gure 19 (a)).
However, as already discussed, tubes with dt > 2 nm are not expected to exhibit a
large (resonantly enhanced) Raman cross-section, and their RBM band is therefore
much more di cult to detect.
742 M. S. Dresselhaus and P. C. Eklund
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

Figure 25. Raman spectra for a SWNT sample prepared from a catalyst mixture Ni± Co± S
using solar energy and using Elaser excitation at 2.54, 2.41 and 1.92 eV. Spectra for the
radial breathing mode band are shown on the left and for the tangential stretch G-
band are shown on the right. The top scale indicates the nanotube diameters
corresponding to the RBM frequencies on the lower scale by use of equation (20)
[73].

Alvarez et al. [73] used the radial breathing mode to evaluate the e€ ect of sulphur
(S) on the synthesis of SWNTs in a solar-powered oven, and the results of that study
are shown in ® gure 25, where we display room temperature RBM spectra of SWNT
bundles synthesized in a solar oven using a Ni± Co± S catalyst. The RBM spectra for
these solar energy-derived nanotubes show a very broad range of RBM frequencies,
indicating that the diameter distribution of these solar oven nanotubes is also
broader than, for example, the nanotube distribution usually obtained from the
PLV or electric arc synthesis methods. The spectra in ® gure 25 were collected at
Elaser ˆ2.54 eV (top), 2.41 eV (middle) and 1.92 eV (bottom), and the scale for dt
shown at the top of the 2.54 eV excitation spectrum is calculated according to
equation (20) using D !RBM ˆ 6:5 cm ¡1 and !0…10;10†d…10 ;10† ˆ 232 cm ¡1 nm [73]. As
Elaser is varied, di€ erent nanotubes within the sample have interband transitions (see
® gure 8) in resonance with the incident or scattered photons, thereby accounting for
Phonons in carbon nanotubes 743
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

Figure 26. Plot of the observed Raman frequencies for the tangential G-band modes taken
with 2.41 eV laser excitation energy for several di€ erent single-wall carbon nanotube
samples, prepared with di€ erent catalysts under di€ erent growth conditions yielding
di€ erent mean nanotube radii r ˆ dt =2. The dashed lines are theoretical LO and TO
energy dispersion curves for 2D graphite along G± M in the 2D Brillouin zone as a
function of q and q =2 for the n ˆ 1 and n ˆ 2, respectively. Solid dots and the symbol
`T’ correspond to the observed frequencies of the peaks and shoulders in the Raman
spectra, respectively [93].

the di€ erent spectra shown in ® gure 25. By carrying out measurements of !RBM for
many Elaser values, the diameter distribution of the nanotube sample can be
estimated [84± 86, 88± 92]. This procedure turns out to be a practical method for
characterizing SWNT samples to be used for the measurement of many physical
properties. In such characterization studies it is important that RBM spectra be
taken with several di€ erent laser excitation energies Elaser .

4.2. T angential stretch modes


The features in the Raman spectra of single-wall carbon nanotubes in the range
1550¡1600 cm ¡1 (see ® gure 20) are identi® ed with the tangential stretch G-band
modes and can be understood by zone-folding of the 2D graphite phonon dispersion
relations [93] as discussed in section 3.1, and therefore show a number of
characteristics that relate to the low dimensionality of the carbon nanotube. The
dashed lines in ® gure 26 are theoretical LO and TO energy dispersion curves in
graphite plotted along G± M in the 2D graphite Brillouin zone as a function of
wavevector in units of 109 cm ¡1 . For armchair nanotubes, the G± M direction
corresponds to the 2D wave vector for the graphene sheet directed along the equator
of the nanotubes for n ˆ 1 and n ˆ 2 (® rst harmonic vibration) , respectively.
Theoretical plots in ® gure 26 are made for the LO …n ˆ 1†, TO …n ˆ 1†, and TO
(n ˆ 2) phonons, which were identi® ed with the A1g , E1g and E2g Raman modes,
respectively [93]. For the TO …n ˆ 2† curve, the plot in ® gure 26 is made against q=2.
Because of the small size of the nanotube diameters of SWNTs, there are only a few
atoms along the circumference of the nanotubes, and consequently there are only a
few allowed wavevectors. The application of Born± von Karman boundary con-
ditions to zone-folded phonons requires an integral number p of wavelengths ¶ to ® t
744 M. S. Dresselhaus and P. C. Eklund

into the circumference of the nanotube (to set up a standing wave pattern). Thus the
boundary conditions require that p¶ ˆ pdt , where ¶ ˆ 2p =q, and p ˆ 0;1;2;. . .. The
allowed wavevectors for nanotubes are therefore q ˆ 2p =dt , so that the magnitude of
the lowest allowed non-zero wavevector will depend on the nanotube diameter dt .
Therefore, the measured frequencies for the tangential phonon modes for di€ erent
diameter nanotubes should provide experimental points on the dispersion relations,
which collectively can be used to plot out the phonon dispersion curves near the G-
point.
In ® gure 26, the observed Raman spectral frequencies are shown for three single-
wall nanotube samples prepared with di€ erent catalysts and under di€ erent growth
conditions, yielding nanotubes with di€ erent mean diameters (radii) and di€ erent
distributions of diameters (radii). Speci® cally the catalysts and resulting average
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

nanotube radii …r ˆ dt =2 are: (a) Fe/Ni (r ˆ 0:55 nm), (b) Co (r ˆ 0:65 nm) and (c)
La (r ˆ 1:0 nm), each with a distribution D r ˆ §0:1 nm. The distribution in the
experimental nanotube radii is inferred from use of the dependence of the radial
breathing mode frequency on nanotube diameter (!RBM / 1=dt ) and measurements
of !RBM for several laser excitation energies Elaser for isolated SWNTs (see
section 4.1). The data in ® gure 26 shows a clear relationship between
…1 =r† ˆ …2 =dt† (or wavevector) and the zone-folded phonon energy dispersion curves,
arising from the small number of allowed wave vectors in the circumferential
direction of the nanotube as discussed above. From ® gure 15, we see A1g and E2g
modes that are in the circumferential direction in the case of armchair nanotubes,
while the E1g mode vibrations are along the nanotube axis. This work [93] suggests
that the upshifted A1g mode is the most intense and the downshifted TO (n ˆ 1) E1g
mode is less intense. It is interesting to observe that the mean frequency between the
TO (n ˆ 1) and LO (n ˆ 1) modes in ® gure 26 should be close to that of graphite
(1582 cm ¡1 ), consistent with experimental results on many nanotube samples.
Most studies of the tangential band for single-wall carbon nanotubes have
focused primarily on the dependence of the spectra on laser excitation energy, along
with the spectral dependence on nanotube diameter. Shown in ® gure 27 are the
Raman spectra between 1400± 1700 cm ¡1 obtained from a SWNT sample with a
diameter distribution dt ˆ 1:37 § 0:20 nm using di€ erent laser excitation energies
(0:94 µ Elaser µ 3:05 eV) [28, 93± 95]. The main features in these spectra are associ-
ated with the tangential C± C stretch modes of the SWNTs (see ® gure 15). We see
that all the spectra obtained in ® gure 27 for Elaser < 1.7 eV or Elaser > 2.2 eV are quite
similar. An example of the ® t of these spectra to a set of Lorentzian oscillators with
peaks at 1563 cm ¡1 , 1591 cm ¡1 and 1601 cm ¡1 is shown in ® gure 28 (a). In fact, all
the spectra for these ranges of Elaser (see ® gure 27) can be ® t by the same set of
Lorentzian oscillators, with essentially the same oscillator strengths and linewidths.
From the plot in ® gure 8, the spectrum in ® gure 28 (a) is identi® ed with a resonant
Raman process involving interband transitions between singularities in the 1D
density of electronic states for semiconducting nanotubes in resonance with the
incident and/or scattered photons.
In contrast, the spectra in ® gure 27 obtained in the narrow range 1:7 < Elaser µ
2.2 eV are qualitatively di€ erent, showing Raman bands that are broader, and the
centre of the band has downshifted by ¹30 cm ¡1 . All of these spectra have been
described in terms of three characteristic Lorentzian components that are not present
in the spectra for the semiconducting nanotubes excited by Elaser outside of this
energy range. The additional characteristic components have peaks at 1515 cm ¡1 ,
Phonons in carbon nanotubes 745
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

Figure 27. Raman spectra of the tangential G-band modes of carbon nanotubes with
diameters in the range dt ˆ 1:37 § 0:20 nm, obtained with several di€ erent laser lines.
The inset shows low resolution Raman spectra between 1300 and 2800 cm ¡1 in the
range of laser energies 2.00± 2.18 eV where the metallic nanotubes are dominant [94].

1540 cm ¡1 and 1581 cm ¡1 , as shown in ® gure 28 (b), in addition to the three


Lorentzian components identi® ed above with semiconducting nanotubes. By refer-
ring to ® gure 8 these three additional broader Lorentzian components (at 1515 cm ¡1 ,
1540 cm ¡1 and 1581 cm ¡1 ) are identi® ed with metallic nanotubes, with Elaser for the
incident and/or scattered photons being in resonance with the interband electronic
transition for metallic nanotubes. The ® tting of the 1540 cm ¡1 feature for the metallic
nanotubes to a Breit± Wigner± Fano lineshape is discussed below [29, 73]. The inset to
® gure 27 shows low resolution Raman spectra between 1300 and 2800 cm ¡1 using
three laser energies in the metallic tube transition region between the two regimes
where the semiconducting nanotubes are resonant [94]. The spectra in the inset show
that the intensity of the second-order band at 2700 cm ¡1 (see section 4.8) is almost
independent of Elaser , whereas the intensity of the tangential band is enhanced when
Elaser approaches the metallic window near 2 eV.
Referring to ® gure 8 and section 4, we see that for Elaser ˆ 1:92 eV, a SWNT
S
sample with dt ˆ 1:49 § 0:20 nm can be in resonance with the E33 …dt† transition for
semiconducting nanotubes at the high end of the diameter distribution and with the
746 M. S. Dresselhaus and P. C. Eklund

1.49 eV 1.92 eV
s i t y ( a .u(a.u.)

1594 1540
.)

In t e n s i t y ( a .u .)
In t e n Signal

1591
1601 1515 1581
Raman

1601
1569

0 0
1400 1500
1500 1550 1600
1600 1650 1700
1700 1400 1500
1500 1550 1600
1600 1650 1700
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

1400 1450 1400 1450 1700


?1 ?1
Raman shift (cm ) -1 Raman shift (cm ) -1
Raman Shift (cm ) Raman Shift (cm )
Figure 28. Lorentzian ® ts to the Stokes tangential bands for Elaser ˆ 1:49 eV and 1.92 eV for
nanotubes with dt ˆ 1:49 § 0:20 [96].

M
E11 …dt† transition for metallic nanotubes at the lower end of the diameter distri-
bution. In ® gure 8 we also can see that by going to SWNT samples with larger
diameter, the resonance of the interband transition for metallic nanotubes E11 M d
… t†
with the incident or scattered photons can more easily be limited to only metallic
nanotubes at lower Elaser , or limited only to semiconducting nanotubes in resonance
S
with the E22 …dt† transition. Figure 8 also shows that if the SWNT sample has a broad
distribution of nanotube diameters as in ® gure 25, the spectra for the tangential band
will show resonant contributions from both semiconducting and metallic nanotubes.
For example, the broad low frequency tails, especially for the traces at
M
Elaser ˆ 2:41 eV and 2.54 eV can be explained in terms of a resonant E11 …dt†
contribution from the small diameter metallic nanotubes contained in the sample.
The strategic selection of Elaser to excite only metallic or only semiconducting
nanotubes is discussed further in section 4.5.
For a metallic nanotube with a given diameter dt , the enhancement of its Raman
peaks will occur every time the incident or scattered photon is in resonance with the
energy separation between the highest valence subband Ev1 …dt † and the lowest
conduction subband Ec1 …dt †, so that E11 …dt † ˆ Ec1 …dt† ¡ Ev1 …dt † [42]. Since the
sample used in obtaining the spectra in ® gure 28 contains metallic nanotubes with
diameters in the range 1.1± 1.6 nm (see inset of ® gure 29), the overall enhancement of
the intensity of a particular Raman mode, such as the most intense feature identi® ed
with metallic nanotubes which is at 1540 cm ¡1 , is given by the sum of the
contributions of each individual nanotube with a given diameter dt , weighted by
the distribution of diameters which is here assumed to be Gaussian. The expected
Raman intensity I …Elaser ;dt † for the dominant Lorentzian oscillator (at 1540 cm ¡1 )
associated with the metallic nanotubes can then be written as [97]:
" #
X ¡…dt ¡ d0†2
I …Elaser ; dt † ˆ A exp
d D dt2 =4 t

M
£ ‰ …E11 …dt† ¡ Elaser †2 ‡ G2e =4Š¡1
M
£ ‰ …E11 …dt† ¡ Elaser ‡ Eph†2 ‡ G2e =4Š¡1 ; …21†
Phonons in carbon nanotubes 747
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

Figure 29. The solid circles represent the intensity ratio of the Raman peaks at 1540 and
1593 cm ¡1 , and the solid curve represents the ® t to the experimental data using
equations (21) and (22) [94]. The inset shows the distribution of diameters measured
by TEM [28] and the Gaussian ® t to the diameter distribution data [94].

where d0 and D dt denote the centre and the half-width of the Gaussian distribution
dt ˆ d0 § D dt of nanotube diameters, obtained by transmission electron microscopy
measurements, Eph is the average energy (0.20 eV) of the tangential phonons and the
damping factor Ge accounts for the width of the singularities in the electronic density
of states (DOS) and the lifetime of the excited state [42]. The factor A is for simplicity
assumed to be a constant, which is equivalent to assuming that all the metallic
nanotubes have equal Raman scattering cross-sections for the dominant oscillator at
1540 cm ¡1 (see ® gure 28). The intensity I …Elaser ;dt† in equation (21) is very sensitive
to Elaser for a given SWNT sample, and a plot of I …Elaser ;dt † versus Elaser , as in
® gure 29, serves to de® ne the metallic window for a particular SWNT sample. The
metallic window is de® ned as the range in Elaser where metallic SWNTs within a
given sample contribute resonantly to the Raman spectra. The expression

M 6aC¡C® 0
E11 …dt† ˆ …22†
dt
M
provides a good approximation to the centre of E11 …dt† for metallic nanotubes of
diameter dt and arbitrary chiral angle (see ® gure 8) [98, 99], where aC¡C is the nearest
neighbour carbon± carbon distance and ® 0 is the electronic overlap integral. In the
absence of a calibration of the absolute intensity of I …Elaser ;dt †, the intensity of the
dominant component in the Raman spectra for semiconducting nanotubes (at
¹1593 cm ¡1 ) is used to normalize the intensity of the dominant component associ-
ated with the metallic nanotubes (at ¹1540 cm ¡1 ), since most spectral traces showing
748 M. S. Dresselhaus and P. C. Eklund

contributions from metallic nanotubes (see ® gure 28 (b)) also show contributions
from semiconducting nanotubes. The resulting experimental intensity ratio
I1540 =I1593 is plotted in ® gure 29 versus Elaser . The best ® t of equations (21) and
(22) to the plot of the intensity ratio I1540 =I1593 versus Elaser measurements in ® gure
29 for a SWNT sample with d0 ˆ 1:37 nm and D dt ˆ 0:18 nm, was achieved
for ® 0 ˆ 2:95 § 0:05 eV, Ge ˆ 0:04 § 0:02 eV, a full width of the distribution
D E11M
…dt† ˆ 0:24 eV, and a mean value for the energy separation hE11 M
…dt†i ˆ
M M
1:84 eV. These values for hE11 d
… t†i and E d
11 … t † are in good agreement with the
M [ ]
direct measurements of E11 d
… t† by scanning tunnelling spectroscopy (STS) 13 and
with electron energy loss spectroscopy (EELS) experiments on individual metallic
nanotubes [100].
At the present time, there is no detailed theory available for the explanation
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

of the broad downshifted tangential bands associated with the metallic


nanotubes, except for the general argument that these phenomena arise from
the interaction between the conduction electrons and the tangential mode phonons
[94], and perhaps involving phonon± surface plasmon interactions [101, 102].
By analysing Raman lineshapes associated only with semiconducting nanotubes
or only with metallic nanotubes, it should be possible to determine whether any of
the lineshape constituents should be ® t to a Breit± Wigner± Fano (BWF) lineshape. It
is presently believed that the resonance Raman lineshapes for semiconducting
nanotubes can best be ® t by Lorentzian oscillators, while the lineshapes for the
low frequency features in the resonance Raman spectra for metallic nanotubes (the
stongest feature near 1540 cm ¡1 ) are better ® t by a BWF function [29, 73, 101, 102].
Such a detailed lineshape study should provide information on the coupling of the
electrons to other excitations of the nanotube [29, 103], since a Breit± Wigner± Fano
lineshape is an indication of a sharp Raman line interacting with a broad Raman
continuum background, presumably associated with electronic excitations. We will
frequently return to the unusual low dimensional properties of the tangential G-band
for carbon nanotubes in the remaining subsections of section 4.

4.3. T emperature dependence of the Raman spectra


The dependence of the Raman spectra on temperature provides valuable
information about anharmonic terms in the lattice potential energy. In general,
the change in phonon frequency with temperature can be attributed to a purely
thermal e€ ect …@! =@T †V and to a volume-related e€ ect …@! =@V †T …@V =@T †P through
lattice expansion phenomena, so that we can write

D !ˆ …@T@!† D
V
T‡ …@V@!† …@V@T † D
T P
T: …23†

Direct measurements on highly oriented pyrolytic graphite (HOPG) [104]


indicate that D !=D T is mainly due to a purely thermal e€ ect and that the
volume-related e€ ect is very small, suggesting that a similar behaviour could be
expected in SWNTs.
There are three studies thus far on the temperature dependence of the Raman
spectra in carbon nanotubes [92, 105, 106]. Two of the measurements are on
MWNTs where the incident laser power level is used to vary the temperature, and
the anti-Stokes/Stokes Raman intensity ratio is then used to determine the nanotube
temperature. One preliminary measurement of the T dependence of the Raman
Phonons in carbon nanotubes 749

spectra on SWNTs is currently available, and this measurement was done on a


sample in a cryostat which allowed the control of the temperature between 5 and
500 K and emphasis was here given to the dramatic increase in linewidth of the radial
breathing mode as the temperature was increased [92]. In experiments carried out in
cryostats, care must be exercised to provide He exchange gas to the sample to give
good thermal contact between the sample and the thermal reservoir to be sure that
the temperature of the sample is properly controlled. Because of the special resonant
Raman e€ ect in carbon nanotubes, di€ erent nanotubes are resonant in the Stokes
and anti-Stokes processes at a given laser excitation energy, and therefore the anti-
Stokes/Stokes intensity ratio does not in general provide an accurate determination
of the nanotube temperature, as discussed further in section 4.5. The reported
temperature-dependen t Raman spectra for MWNTs [105, 106] therefore need
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

further re-examination, and consequently caution should be applied to use of the


current literature on the temperature dependence of the nanotube mode frequencies
and intensities [105, 106].
For all the Raman-active modes, the e€ ect of increasing temperature is to lower
the mode frequency [105, 106], consistent with results for graphite and other sp2
carbons [104, 107]. For example, values of the temperature dependence of the
Raman-active G-band has been reported to be ¡0:011 cm ¡1 K ¡1 for highly oriented
pyrolytic graphite, ¡0:028 cm ¡1 K ¡1 for disordered graphite, ¡0:027 cm ¡1 K ¡1 for
activated carbons [104] and ¡0:026 cm ¡1 K ¡1 for MWNTs [106]. The reported
downshift in the frequency of the radial breathing mode for SWNTs is
¹0.012 cm ¡1 K ¡1 [92]. The various authors ® nd di€ erent relative magnitudes for
the downshifts in mode frequency for the various modes for the MWNTs, with
¡0:019 cm ¡1 K ¡1 given for the 1340 cm ¡1 D-band mode, ¡0:029 cm ¡1 K ¡1 given for
the 1630 cm ¡1 D 0-band feature, and ¡0:034 cm ¡1 K ¡1 for the 2700 cm ¡1 G 0-band.
Although there is disagreement in the literature about the absolute magnitude of the
temperature dependence …@!=@T † for the various modes, there is general agreement
about the relative magnitudes of …@! =@T † for the tangential G-band, the D-band and
the G 0-band [105, 106]. The value of …@! =@T † for the radial breathing mode in
MWNTs is found to be smaller than for the other modes [105].
The larger value of @!tang =@T for the tangential G-band in carbon nanotubes
relative to crystalline graphite can be attributed to the curvature of the nanotubes,
leading to a mixing of a small out-of-plane force constant component into the
tangential normal modes. Since the out-of-plane lattice constants and force
constants of 2D graphite are strongly temperature dependent, the modes with
displacements in the circumferential direction of carbon nanotubes are expected to
exhibit a larger value for @!tang =@T relative to that for a graphene sheet. The larger
value of @!tang =@T for disordered sp2 carbons relative to graphite is related to
defects. Within the Elaser range where only semiconducting nanotubes are in
resonance, it is expected that once reliable experimental data on @!tang =@T are
available for SWNTs, the mode frequencies for the tangential modes can be used as
an indicator of the lattice temperature of the nanotubes.
Better measurements of …@!=@T † are needed for the various Raman-active modes
on puri® ed, well characterized SWNTs where the temperature is measured directly
and the nanotube sample is in good thermal contact with the thermal reservoir. It
should further be determined whether or not …@! =@T † for a given mode is the same
for semiconducting and metallic nanotubes.
750 M. S. Dresselhaus and P. C. Eklund
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

Figure 30. The pressure dependence of the room-temperature Raman spectra of SWNT
bundles, for the low-frequency radial breathing band (a) and the high-frequency
tangential G-bands denoted by T1, T2, T3 (b), for a sample with a mean diameter
between that of a (9, 9) and a (10, 10) nanotube (dt ˆ 1:31 § 0:07 nm) [64].

4.4. High pressure e€ ects on the tangential modes


Pressure-dependent studies are particularly interesting for single-wall carbon
nanotube bundles, because pressure can be used to decrease the separation between
adjacent SWNTs and therefore to increase the inter-tube coupling. Thus pressure
provides a sensitive technique for measurement of the e€ ect of inter-tube interactions
in typical SWNT samples used for the measurement of Raman spectra and the other
phonon-relate d measurements described in this review.
The pressure dependence of the radial breathing band and of the tangential band
of SWNTs (dt 1:31 § 0:07 nm) was measured up to 5.2 GPa using a diamond anvil
cell and a methanol± ethanol (4:1 volume ratio) pressure transmission medium [64].
The laser excitation (5 mW) was at 514.5 nm (Elaser ˆ 2:41 eV) inside the high
pressure cell, corresponding to the regime where the interband transitions for the
S
semiconducting nanotubes (E33 …dt† as shown in ® gure 8) should be in resonance with
the incident and scattered photons [64]. An increase in the nanotube mode frequency
is expected with increasing pressure due to the increase in the nearest neighbour C± C
interactions and consequently in the force constants and also to an increase in the
inter-tube interaction leading to a distortion of the nanotube cross-section. The
experimental results for both the radial breathing (R) band and for the tangential
(T 1 , T 2 and T 3 ) bands (see ® gure 30) all show increases in mode frequency with
increasing pressure [64], in agreement with these intuitive predictions. For both the
low frequency and high frequency Raman bands, the scattering intensity decreases
and the linewidth increases with increasing pressure, and the Raman spectra are
largely (but not completely) recovered when the pressure is released. The intensity
decrease for the radial breathing mode with increasing pressure is quite severe, and
Phonons in carbon nanotubes 751

Table 3. Values of the ® tting parameters aP and bP in


equation (24) obtained by ® tting the measured pressure-
dependent Raman frequencies in ® gure 30 [64].

!…0† aP bP
(cm¡1 ) (cm¡1 GPa ¡1 ) (cm ¡1 GPa ¡2 )

186 § 1 7§ 1 ±
1550 § 1 8§ 1 ±
1564 § 1 10 § 1 ¡0:9 § 0:3
1593:0 § 0:7 7:1 § 0:8 ¡0:4 § 0:2
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

® gure 30 shows that the mode is completely quenched at a pressure of 1.9 GPa [64].
The pressure dependence of the mode frequency !…P† at 300 K was ® t by a relation

!…P† ˆ !…0† ‡ aP P ‡ bP P2 …24†


and the results for the coe cients aP and bP are given in table 3, showing a
predominantly linear dependence of the mode frequencies on pressure. Since the
displacements of the radial breathing mode are more sensitive to the presence of
atoms on adjacent nanotubes, a larger percentage upshift of the radial breathing
mode relative to the tangential mode is expected, in agreement with experiment (see
table 3). Calculations [65] of the expected pressure dependence of the radial
breathing mode, the low frequency E2g squash mode (17 cm ¡1 in ® gure 15), and
the tangential G-band (1550± 1600 cm ¡1 in ® gure 15) have been carried out. The
calculations show that !RBM has an approximate linear increase with pressure up to
¹10 cm ¡1 GPa ¡1 , and the calculated pressure dependence of the tangential G-band
frequency is also small at 15 cm ¡1 GPa ¡1 , but much larger than the 5 cm ¡1 GPa¡1
value observed for graphite [108]. These calculated values for SWNTs [65] are
consistent with the experimental measurements [64, 65].
To explain these results in more detail, model calculations were performed on
(9, 9) SWNTs on the basis of 3 models : Model I in which the entire bundle of
nanotubes, arranged in a triangular lattice, is subjected to an external compression,
Model II in which the individual nanotubes are each compressed symmetrically and
inter-tube coupling is ignored, and Model III in which the pressure medium is
allowed to penetrate into the interstitial channels between the nanotubes, thereby
introducing an angular dependence [64]. Models I and III included van der Waals
coupling and Model II did not. All model calculations are based on a generalized
tight binding molecular dynamics (GTBMD) calculation for an isolated nanotube
[109], to which a Leonard-Jones potential is added to account for the inter-tube
coupling. A comparison between the model calculations and the experimental
measurements for the frequency shifts for each of the modes is shown in ® gure 31,
where a reasonably good ® t to the experimental data is obtained with Model I. From
® gure 31, we see that the radial breathing mode frequency for Model II which
neglects the inter-tube interaction !0RBM is 14 cm ¡1 lower in frequency than !RBM for
Model I which includes this interaction, thereby giving a value for D !RBM in
equation (20). It is thus concluded that for SWNTs with diameters in the range
between a (9, 9) and a (10, 10) nanotube (1:31 § 0:07 nm), the frequency of the radial
752 M. S. Dresselhaus and P. C. Eklund
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

Figure 31. The pressure dependence of the (a) radial and (b) tangential vibrational mode
frequencies of a bundle of (9, 9) SWNTs. The points are experimental and the curves
are ® t to various models (see text). The pressure dependence of the E2g2 mode for
graphite is included [64].

Figure 32. The pressure dependence, as calculated from Model I, of (a) the lattice constants
and (b) the hexagonal distortion of the cross-section of an individual nanotube in the
bundle [64]. The lattice constant a along the nanotube axis is normalized to a0 (at
zero external pressure) which is close to the value 2.46 A Ê for a graphene sheet.
Likewise ¹ and ¹ 0 are normalized to ¹0 (at zero applied pressure) which is close to
1.72 nm as reported for X-ray di€ raction measurements on SWNT bundles [6] (see
text).

breathing mode for SWNTs in a bundle is given by equation (20), and D !RBM can be
determined from calculations such as those shown in ® gure 31.
The model calculations also addressed the nanotube distortions introduced by
the inter-tube interaction. The resulting calculated pressure dependence of the
graphene lattice constant a along the nanotube axis and of the lattice constant ¹
Phonons in carbon nanotubes 753

for the nanotube bundle is shown in ® gure 32. As shown schematically by the inset to
® gure 32 (b), the e€ ect of pressure is to increase the inter-tube interaction within a
nanotube bundle and to cause some faceting and departures from a circular cross-
section denoted by the equilibrium radius r0 . Predictions for the ¯ attening e€ ect,
expressed by r…<†, is found to be about 1% per GPa of pressure as shown in
® gure 32 (b), where the pressure dependence of r…>† =r0 and r…<† =r…>† are also plotted
[64]. Polygonalization e€ ects have been reported for carbon nanotubes not subject to
external pressure using high resolution TEM studies [110 ± 112]. The e€ ect of this
polygonalization on the electronic structure of SWNTs has been shown to create
bandgaps in metallic nanotubes (such as (12, 0) nanotubes) , or to cause semiconduct-
ing nanotubes to become metallic [113].
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

4.5. Anti-Stokes spectra


Study of the anti-Stokes spectra [104, 106, 107] (scattered photon having energy
Elaser ‡ Eph ) in comparison to the Stokes spectra (scattered photon having energy
Elaser ¡ Eph ) provides much new information about the resonant Raman spectra and
the electronic structure of single-wall carbon nanotubes, including new physics about
anti-Stokes and Stokes scattering phenomena [96]. The additional information
provided by study of both the Stokes and anti-Stokes spectra includes: (1)
observation of resonant Raman scattering pro® les to which only metallic nanotubes
contribute, and other scattering pro® les to which only semiconducting nanotubes
contribute, thereby allowing a more meaningful analysis to be made of the lineshapes
of resonance Raman phenomena in SWNTs; (2) observation of very di€ erent Raman
spectra for the Stokes process as compared to that for the anti-Stokes process, which
is a new phenomenon for carbon-based materials, and relates to the unique 1D
electronic structure of SWNTs; (3) evidence that resonant Raman scattering is
stronger for metallic SWNTs than for semiconducting SWNTs, thereby establishing
a stronger electron± phonon coupling process for metallic SWNTs; (4) the ratio of the
anti-Stokes to the Stokes intensities of the tangential band cannot be used to
determine the sample temperature T s , because di€ erent nanotubes (e.g. metallic
SWNTs in one case, and semiconducting SWNTs in the other case) can contribute to
the Stokes and anti-Stokes spectra.
The ® rst report of an anti-Stokes feature in the Raman spectra for carbon
nanotubes was for the radial breathing mode for multi-wall carbon nanotubes where
the same frequency (120 cm ¡1 ) was found for both the Stokes and anti-Stokes
features. The relative intensities between the anti-Stokes and Stokes features IAS =IS
was used by these authors [105] in the traditional way to determine the sample
temperature using the relation

…! ‡ !0 †4 - 0 =k B T † ¡ 1Š¡1 ;
IAS =IS ˆ ‰ exp …h! …25†
…! ¡ !0 †4
in which ! is the laser frequency and !0 is the phonon frequency. No anomalous
behaviour for the anti-Stokes spectra was reported by these authors [105]. The anti-
Stokes to Stokes intensity ratio for the D-band, the tangential G-band, the D 0 band
at ¹1620 cm ¡1 ; and the G 0-band (see section 4.8) were all used to determine the
nanotube temperature (see section 4.3) for MWNTs [106].
A detailed study of the anti-Stokes spectra in comparison to the Stokes spectra of
SWNTs has recently been carried out [96]. Striking di€ erences are observed between
754 M. S. Dresselhaus and P. C. Eklund

1586 1537
632.8 nm 782.0 nm
(1.96 eV) (1.58 eV)

Anti Stokes Anti Stokes


1319 1299 2602
1735
1915
1580
1586
2620 2566
Stokes Stokes
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

1315 1741 1290 1739


1921 1867

1200
1200 1600
1600 2000
2000 2400
2400 2800 1200
28001200 1600
1600 2000
2000 2400
2400 2800
2800
?1-1 ?1 -1
Raman Shift
Raman shift (cm )) Raman Shift (cm ) )
Raman shift (cm
Figure 33. Stokes and anti-Stokes Raman spectra over a wide frequency range for SWNTs
with dt ˆ 1:49 § 0:20 nm at Elaser ˆ 1:58 eV (782 nm) and 1.96 eV (632.8 nm) [96].

the Stokes and anti-Stokes spectra of SWNTS, as shown in ® gure 33, where the
Stokes and anti-Stokes spectra from 1200± 2800 cm ¡1 are presented for two di€ erent
laser excitation energies Elaser (1.58 eV and 1.96 eV). These spectra include features
associated with the ® rst-order spectra (the tangential band in the region 1500±
1600 cm ¡1 and the D-band in the region 1280± 1330 cm ¡1 ), and second-order features
(combination bands at ¹1740 cm ¡1 and in the 1860± 1925 cm ¡1 range, and the G 0-
band at 2540 ± 2620 cm ¡1 which is an overtone of the D-band). We discuss these
di€ erences in behaviour between the Stokes and anti-Stokes spectra in more detail
below.
The large di€ erences in lineshape between the tangential band for the Stokes and
anti-Stokes spectra are shown more clearly in ® gure 34, at four di€ erent Elaser values
between 1.49 eV and 2.19 eV. Here the behaviour of the Stokes and anti-Stokes
spectra for the tangential band changes as a function of Elaser and the spectra also
change relative to one another as Elaser is varied. The di€ erent characteristic
lineshapes for the tangential band discussed below allow us to easily distinguish
between metallic and semiconducting nanotubes. At Elaser ˆ 2:19 eV, the Stokes and
anti-Stokes spectra in ® gure 34 are almost the same, and both are typical of resonant
Raman spectra characteristic of semiconducting nanotubes (strongest feature at
1591 cm ¡1 ), while at Elaser = 1.92 eV, the Stokes and anti-Stokes spectra are very
di€ erent from each other, the Stokes spectrum showing domination by metallic
nanotubes (strongest feature at 1540 cm ¡1 ), and the anti-Stokes spectrum showing
domination by semiconducting nanotubes. At Elaser = 1.58 eV both the Stokes and
anti-Stokes spectra are both characteristic of metallic nanotubes, while at
Elaser = 1.49 eV, the anti-Stokes spectrum is dominated by metallic nanotubes and
the Stokes spectrum is typical of semiconducting nanotubes. Figure 35 shows the
metallic windows for the Stokes and anti-Stokes processes for a SWNT sample with
a diameter distribution dt ˆ 1:49 § 0:20 nm, as discussed below, and the metallic
Phonons in carbon nanotubes 755

1591 1591
Stokes Anti Stokes
2.19 eV 2.19eV
1582
1540 1588

1.92 1.92
1540 1537

1.58 1.58
1592 1540

1.49 1.49
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

0
1300
1300 1400 1500
1500 1600 1700
1700 1800 1900
1900 1300
1300 1400 1500
1500 1600 1700
1700 1800 1900
1900
?1 ?1
Raman
Figure 34. Stokes and shift (cm )Raman spectra for SWNTs
anti-Stokes Ramanofshift
dt (cm ) § 0:20 nm taken
ˆ 1:49
at 4 di€ erent values of Elaser to illustrate di€ erences in spectral lineshape between
the Stokes and anti-Stokes spectra [96]. For Elaser ˆ 2:19 eV, both the Stokes and
anti-Stokes processes are in resonance with semiconducting nanotubes; for
Elaser ˆ 1:92 eV, the Stokes process is in resonance with metallic nanotubes and
the anti-Stokes process is in resonance with semiconducting nanotubes ; for
Elaser ˆ 1:58 eV, both processes are in resonance with metallic nanotubes ; and ® nally
for Elaser ˆ 1:49 eV, the Stokes process is in resonance with semiconducting
nanotubes, and the anti-Stokes process is in resonance with metallic nanotubes. The
nanotube-speci® c resonant Raman process (see ® gure 8) associated with the special
characteristics of the 1D density of electronic states is responsible for the observed
di€ erences between the Stokes and anti-Stokes lineshapes [96].
windows in ® gure 35 account for the observations in ® gure 34. Whereas for 2D and
3D sp2 carbon materials, the Stokes and anti-Stokes tangential bands at a given Elaser
value are essentially identical, the unusual resonant Raman process for 1D carbon
nanotubes gives rise to di€ erences in the Stokes and anti-Stokes tangential G-band
spectra when one spectrum is excited with a laser frequency from within the metallic
window discussed in section 4.2, and the other spectrum is not, in accordance with
® gure 8. These di€ erences between the Stokes and anti-Stokes spectra are unique to
SWNTs in comparison to other sp2 carbon-based materials and arise from the
di€ erences in the one-dimensional density of electronic states for metallic and
semiconducting nanotubes discussed in section 2.2. This observation allows selection
of Elaser (e.g. Elaser ˆ 1:49 eV) to resonantly excite only semiconducting SWNTs in
the Stokes spectra or only metallic nanotubes in the anti-Stokes spectra (see
® gure 35). These di€ erences between the Stokes and anti-Stokes spectra at a given
Elaser is due to the unique resonant enhancement phenomena arising from the one-
dimensional electronic (1D) density of states of carbon nanotubes. Figure 35 clearly
shows that only semiconducting nanotubes are in resonance with Elaser ˆ 2:19 eV,
while at 1.92 eV metallic nanotubes are in resonance for the Stokes process and
semiconducting nanotubes for the anti-Stokes process. Furthermore, ® gure 35 shows
that at Elaser ˆ 1:58 eV, metallic nanotubes are in resonance for both the Stokes and
anti-Stokes processes, and at Elaser ˆ 1:49 eV the semiconducting nanotubes are in
resonance for the Stokes process and the metallic nanotubes for the anti-Stokes
process, in agreement with the experimental observations in ® gure 34.
To compare the Stokes and anti-Stokes spectra for semiconducting and metallic
nanotubes in more detail, we show in ® gure 36 a lineshape analysis for the tangential
756 M. S. Dresselhaus and P. C. Eklund

AS
S

1.69 eV
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

1.25 1.55 1.85 2.15


Laser Energy (eV)
Figure 35. The calculated laser excitation energy dependence of the intensity IM …Elaser ; dt †
using equation (26) for the strongest Lorentzian oscillator (1540 cm¡1 ) for metallic
nanotubes in a SWNT sample with dt ˆ 1:49 § 0:20 nm. The parameters used in
equation (26) were determined from ® ts of experimental Stokes spectra to this
equation. These plots thus de® ne the metallic resonance window for Stokes (solid
curve) and anti-Stokes (dotted curve) scattering processes, where metallic nanotubes
are in resonance with the incident or scattered photons. The arrow at 1.69 eV is at the
centre of the metallic window [96].

band at Elaser ˆ 1:49 eV and 1.92 eV, which for simplicity was carried out only in
terms of Lorentzian components [96]. For semiconducting nanotubes, the Stokes
spectrum at Elaser ˆ 1:49 eV can be interpreted in terms of three characteristic
Lorentzian peaks: at 1569 cm ¡1 , 1594 cm ¡1 and 1601 cm ¡1 [94]. At the same
Elaser ˆ 1:49 eV, the anti-Stokes (metallic) spectrum shows only three broad Lor-
entzian components at 1515 cm ¡1 , 1540 cm ¡1 and 1581 cm ¡1 , none of which are
present in the Stokes (semiconducting) spectrum at the same Elaser value. It is
interesting that features associated with semiconducting nanotubes are absent from
this anti-Stokes spectrum at Elaser ˆ 1:49 eV. In contrast, the Stokes tangential band
at 1.92 eV is mostly dominated by three Lorentzian components characteristic of
metallic nanotubes, but also shows Lorentzian components identi® ed with semi-
S
conducting nanotubes in resonance with E33 …dt† at the large dt end of the diameter
distribution (see ® gure 8). Likewise the anti-Stokes (semiconducting) spectrum at
Elaser ˆ 1:92 eV has a long tail at low ! (below ¹1550 cm ¡1 ), indicating a signi® cant
M
contribution from metallic E11 …dt† nanotubes at the low end of the diameter
distribution (® gure 8). Although the low ! tails often seen in the semiconducting
tangential mode spectra could be identi® ed with a Breit± Wigner± Fano lineshape
[29], the explanation of these tails suggested by ® gure 8 is due to the presence of some
metallic nanotubes in the sample that are resonant with either the incident or
scattered photons. Study of the anti-Stokes spectra accesses greater contributions
from the lower right quadrant of ® gure 8, thereby providing more accurate
information about lineshapes for metallic and semiconducting tangential bands,
because the Eii …dt † bands of points at constant dt in ® gure 8 are narrower and more
separated from one another in this limit [30, 31]. Since the Stokes and anti-Stokes
spectra in ® gure 36 taken at 1.49 eV have a common incident Elaser value, some
explanation is needed for the interpretation of the Stokes spectra arising from
contributions of resonant semiconducting nanotubes and the anti-Stokes spectra
Phonons in carbon nanotubes 757

anti-Stokes Stokes
1.49 eV 1540 1.49 eV
1594
( a .u .)

In t e n s i t y ( a .u .)
t e n s i t y (a.u.)

1581
I nIntensity

1601

1569
1515

0 0
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

14001450 1500
1400 1500 1550 1600
1600 16501700
?1
14001450 1500
1700 1400 1500 1550 1600
1600 16501700
?1
1700
Ramanshift
Raman shift (cm
(cm)-1) Raman shift
Raman shift(cm ) -1 )
(cm

anti-Stokes Stokes
1.92 eV 1.92 eV
1582 1540
(a .u .)

In t e n s i t y (a .u .)
t e n s i t y (a.u.)

15671591
I nIntensity

1601
1557 1588 1515
1581

0 0
1400
1400 1450 1500
1500 1550 1600
1600 16501700
?1
14001450 1500
1700 1400 1500 1550 1600
1600 16501700
?1
1700
Ramanshift
Raman shift (cm
(cm)-1 ) Raman
Raman shift(cm
shift (cm-1))

Figure 36. Lorentzian ® ts to the Stokes and anti-Stokes tangential bands for Elaser ˆ 1:49 eV
and 1.92 eV and dt ˆ 1:49 § 0:20 nm [96]. It is of particular interest that only metallic
nanotubes are in resonance for the anti-Stokes process at Elaser ˆ 1:49 eV, and only
semiconducting nanotubes are in resonance for the Stokes process at Elaser ˆ 1:49 eV,
thus allowing more accurate lineshape analyses to be carried out.

arising from contributions of resonant metallic nanotubes. This interpretation is


based on having the incident photons non-resonant with either the metallic, or the
semiconducting nanotubes in the SWNT sample (see ® gure 8). The special selection
of Elaser so that the incident photon is essentially non-resonant with either metallic or
semiconducting nanotubes, but the scattered photon is strongly resonant with
metallic nanotubes in the anti-Stokes process and with semiconducting nanotubes
for the Stokes process provides a useful method for determining the resonant Raman
lineshapes for semiconducting and metallic nanotubes more accurately. Work to
date suggests that the observed lineshapes are well ® t by Lorentzian oscillators for
the case of semiconducting nanotubes, as was done in the analysis of ® gure 36,
though for metallic nanotubes the strong feature near 1540 cm ¡1 appears to be better
® t by a Breit± Wigner ± Fano lineshape, as was reported by various groups [29, 90,
101,102].
The di€ erences in the metallic window for the Stokes and anti-Stokes processes
for SWNTs can be readily explained in terms of the Raman scattering intensity of
metallic nanotubes:
758 M. S. Dresselhaus and P. C. Eklund
( )
X ¡…dt ¡ d0 †2
IM …Elaser ; dt † ˆ A exp
D d 2 =4
M
£ ‰ …E11 …dt† ¡ Elaser § Eph†2 ‡ G2e =4Š¡1
M
£ ‰ …E11 …dt† ¡ Elaser †2 ‡ G2e =4Š¡1 …26†

in which IM …Elaser ; dt † is the scattering intensity for the Stokes process (‡ sign) and
for the anti-Stokes process (¡ sign) for metallic nanotubes in resonance with the
M
E11 …dt† electronic transition between the highest lying valence band van Hove
singularity and the lowest lying conduction band singularity in their 1D electronic
density of states [1, 94]. The remaining parameters in equation (26) have the same
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

de® nitions as the corresponding parameters in equation (21). The quantity A in


equation (26) is a dimensionless factor proportional to the e€ ective cross-section for
a particular phonon mode for the Stokes process and also includes a multiplicative
Boltzmann factor exp …¡Eph =kT † for the anti-Stokes process [94, 114]. The factor A
will be di€ erent for the various oscillators which were used in the lineshape analysis
of ® gure 36, because of the di€ erences in the normal mode displacements and
symmetry classi® cations associated with the various oscillators.
A plot of IM …Elaser ; dt † for the Stokes spectra, based on a ® t of equation (26) to
measurements on a SWNT sample with dt ˆ 1:49 § 0:20 nm at many laser energies
Elaser is given in ® gure 35 (solid curve) [94, 114], along with the corresponding
predicted IM …Elaser ; dt † curve for the anti-Stokes spectra (dotted curve) using the
same Stokes parameters. The central point denoted by an arrow accurately gives the
centre of the metallic window for nanotubes with diameter dt , which occurs at
1.69 eV for a SWNT sample with dt ˆ 1:49 § 0:20 nm [96] and at 1.80 eV for a
M
SWNT sample with dt ˆ 1:35 § 0:20 nm [115]. Using the formula E11 …dt† ˆ
6aC¡C® 0 =dt , valid for armchair nanotubes [30], yields a value of ® 0 ˆ 2:89 eV
assuming aC¡C ˆ 0:142 nm, in good agreement with ® 0 ˆ 2:91 eV, obtained from
® ts of Stokes spectra to equation (26) for SWNTs with dt ˆ 1:37 § 0:18 nm [94].
Thus the carbon± carbon overlap energy ® 0 for SWNTs is sensitively determined by
the centre of the metallic window and the dt value for the SWNTs, yielding a value of
® 0 ˆ 2:9 § 0:1 eV [30, 32], which is less than 10% lower than ® 0 for graphite [116].
Di€ erences between the Stokes and anti-Stokes spectra are also found (see
® gure 33) for the D-band, G 0-band (see section 4.8), and the !tang ‡ 2!RBM
combination band features with regard to frequency and relative intensity, while
the Stokes and anti-Stokes spectra for the !tang ‡ !RBM band at Elaser ˆ 1:96 eV
excitation are almost the same. No second-order features are observed in the anti-
Stokes spectra above 1700 cm ¡1 for Elaser ˆ 1:96 eV and this observation is explained
by a Boltzmann factor reduction in relative intensity, where it is noted that the anti-
Stokes spectrum at 1.96 eV laser excitation is dominated by semiconducting
nanotubes. The di€ erent frequency shifts of the G 0-band features (see section 4.8)
in ® gure 33 can be explained by the di€ ering energies of the scattered photons which
are involved with the 2D resonance Raman enhancement e€ ects associated with the
K-point in the Brillouin zone [117]. For example, an expected frequency di€ erence of
34 cm ¡1 between the anti-Stokes and Stokes G 0-band frequencies is predicted from
the measured …@!=@Elaser † slope of 106 cm ¡1 /eV for the G 0-band dispersion [118],
which is to be compared with the observed frequency di€ erence of 36 cm ¡1 shown in
® gure 33 (also see section 4.8). The relative intensities of the G 0 features in the Stokes
Phonons in carbon nanotubes 759

and anti-Stokes spectra at Elaser ˆ 1:58 eV can only, in part, be accounted for by the
Boltzmann factor exp…¡Eph =kT †, which a€ ects the anti-Stokes intensity strongly,
but not the Stokes intensity. The experiments [96] show that the G 0-band intensity at
Elaser ˆ 1:58 eV, where the metallic nanotubes dominate the anti-Stokes spectrum, is
much greater than one would expect from the Boltzmann factor argument in relation
to the Stokes intensity at Elaser ˆ 1:58 eV, where the spectrum is mostly dominated
by metallic nanotubes, but some semiconducting nanotubes might also contribute to
the spectra. This additional enhancement of the intensity of the G 0-band for the
metallic nanotubes relative to semiconducting SWNTs implies a stronger electron±
phonon coupling for metallic nanotubes. The second-order features identi® ed with
!tang ‡ 2!RBM (see section 4.9) are observed at 1915 cm ¡1 and at 1921 cm ¡1 in the
anti-Stokes (Stokes) spectrum at Elaser ˆ 1:58 eV (1.96 eV), respectively, and have
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

scattered photons at 1.78 eV and 1.76 eV, thereby accounting for the similar phonon
frequencies that are observed in the two spectra [96, 101].
The observation of anti-Stokes scattering intensity requires population of the ® rst
excited phonon state …n1 †. The ratio n1 =n0 , where n0 is the zero phonon ground state
population for a phonon of energy h! - 0 is

- =kT Š ;
n1 =n0 ˆ exp ‰ ¡h! …27†
0

-
so that the intensity ratio of the anti-Stokes IAS to the Stokes IS signals at Elaser ˆ h!
is expected to be as given in equation (25). Equation (25) is commonly used for
ordinary sp2 carbons to determine the sample lattice temperature from the anti-
Stokes to Stokes Raman intensity. Since di€ erent kinds of nanotubes are resonant in
the anti-Stokes and Stokes spectra at a given Elaser , this equation cannot be used to
determine the sample temperature in SWNTs. However, when both the Stokes and
anti-Stokes spectra at a given Elaser are dominated by semiconducting nanotubes,
equation (25) could be used to provide a reasonable estimate of the sample
temperature.
It is signi® cant that the 1540 cm ¡1 feature has a large intensity in the anti-Stokes
tangential G-band spectrum at Elaser ˆ 1:58 eV, and the presence of the G 0-band and
the second-order features, respectively, at !M M
tang ‡ !RBM and at !tang ‡ 2!RBM in the
anti-Stokes spectrum are all well resolved at Elaser ˆ 1:58 eV where the metallic
nanotube contribution is dominant. It is also signi® cant that these second-order
features are absent in the anti-Stokes spectrum at Elaser ˆ 1:96 eV where the
semiconducting nanotubes are dominant. These observation s indicate that the
resonance Raman process (electron± phonon coupling) occurs more strongly for
metallic nanotubes than for semiconducting nanotubes.

4.6. Surface enhanced Raman spectra in carbon nanotubes


It is well established that the Raman signal from molecules can be enhanced by
many orders of magnitude when these molecules are adsorbed on metallic nano-
structures [119, 120] through the surface-enhanced Raman scattering (SERS) e€ ect
[121, 122]. In fact, the SERS e€ ect can provide enough sensitivity (signal enhance-
ment factors of up to 10 14 ) to measure the Raman spectrum of single molecules when
these molecules are attached to silver nanostructured particles, due to the very
inhomogeneous ® eld distribution near these metal nanoparticles , and to the
extremely large electromagnetic ® elds associated with the excited surface plasmons
[123± 128].
760 M. S. Dresselhaus and P. C. Eklund

This sensitivity of the SERS technique provided initial stimulus to the use of the
SERS technique to reduce the SWNT sample size necessary for obtaining a Raman
spectrum, thereby reducing the inhomogeneous line broadening associated with the
distribution of nanotube diameters and chiral angles within a typical SWNT sample
used for Raman spectroscopy studies. The SERS e€ ect has thus been applied to
study the Raman e€ ect in SWNTs and large enhancement factors in the Raman
signal have been reported [115, 129, 130]. Both Fourier transform SERS [129] and
frequency scanned SERS [115, 130] have been applied to study SWNTs, and both
gold and silver surfaces have been used for the metal substrates. The reason for using
di€ erent metal substrates is to provide a means for distinguishing between the two
main SERS enhancement mechanisms: the `electromagnetic’ enhancement mechan-
ism which is not sensitive to the metal substrate species, and the charge transfer or
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

`chemical’ mechanism which is sensitive to the choice of the metal substrate. For
some Ag and Au nanostructured metal substrates that have been used in SERS
nanotube experiments, there is a large overlap in the broad surface plasmon
absorption bands so that the SERS enhancement e€ ects can be observed over a
broad range of Elaser values. The sensitivity of the SERS probe can be especially high
in the case of SWNTs, since the enhancement e€ ects bene® t both from the strong
resonance Raman e€ ect associated with the van Hove singularities in the 1D
electronic density of states, and from SERS enhancement e€ ects associated with
strong electrical ® elds at the nanotube± metal interface. These two e€ ects can be
combined multiplicatively for laser excitation wavelengths in the visible/near
infrared spectral region. In this review of the SERS studies on SWNTs, a comparison
is made between the spectral features in the SERS spectra with the normal resonant
Raman spectra taken at the same laser excitation energy Elaser . These comparisons
between the SERS and normal Raman spectra reveal qualitatively di€ erent
responses between semiconducting and metallic nanotubes. For example, the
semiconducting nanotubes show basically the same Raman spectral shape in the
SERS and resonant Raman spectra (RRS), except for a scaling factor in signal, while
the metallic nanotubes show notable di€ erences in spectral shape between the SERS
and normal Raman spectra. These topics are further elaborated in this section.
In this review, the comparison between the SERS and normal Raman spectra is
® rst made over a broad frequency shift range to show the big picture, followed by a
more detailed comparison of the behaviour of the tangential G-band in the 1500±
1650 cm ¡1 range. Then the enhancement of the SERS signal with laser excitation
power is discussed and spectra taken under very high enhancement conditions are
presented. The comparison between the SERS and normal Raman spectra in
nanotubes reveals behaviours not present in conventional crystalline samples.
Figure 37 shows a comparison, over a broad frequency range from 500±
3500 cm ¡1 , between the spectrum for normal resonance Raman scattering (RRS)
at Elaser ˆ 1:96 eV (® gure 37 (a)) and the surface-enhanced resonant Raman spectra
(SERRS) taken at the same Elaser value. These spectra were reported for bundles of
single-wall carbon nanotubes on a rough silver ® lm (® gure 37 (b)) and on a rough
gold ® lm (® gure 37 (c)). The dominant feature in the spectra in ® gure 37 [115] is
associated with the ® rst-order tangential G-band occurring in the phonon frequency
range 1500± 1650 cm ¡1 and the lineshape for this band implies a strong resonant
contribution from metallic nanotubes. The features of lower intensity near
1310 cm ¡1 are identi® ed with ® rst-order Raman-active phonon modes derived from
zone-edge phonons near the K-point in the 2D graphene Brillouin zone [117] (see
Phonons in carbon nanotubes 761

1591
1.96 eV
Raman Intensity (a.u.) 2600
1307
1730
(a) 1900
1588
1541
2624

1311
1741
(b ) 1931
1588

2624
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

1315
1739
(c) 1931

500 1000 1500 2000 2500 3000 3500


Raman shift (cm-1 )
Figure 37. Normal resonant Raman scattering (RR) spectrum (a), and surface enhanced
resonant Raman scattering (SERS) spectra of single-wall nanotubes (SWNTs)
adsorbed on (b) gold and (c) silver island ® lms in the 500± 3500 cm¡1 range using laser
excitation at Elaser ˆ 632:8 nm (1.96 eV) [115].

section 4.8). Similar to assignments made for normal second-order resonant Raman
scattering, the feature in the 1730± 1740 cm ¡1 range is tentatively associated with a
combination mode [114] involving the tangential and radial breathing mode phonons
(!tang + !RBM ). The feature in the 1900± 1930 cm ¡1 range is also tentatively
attributed to a combination mode [114] (!tang + 2!RBM ), as discussed further in
section 4.9. A weak feature near 2440 cm ¡1 (only observed at higher magni® cation) is
identi® ed with a non-resonant overtone in the second-order spectrum of the K-point
vibration at about 1220 cm ¡1 [114], as discussed further in section 4.8, and ® nally the
strong feature in the 2600± 2624 cm ¡1 range is due to the G 0-band which occurs as an
overtone of the D-band feature at twice the D-band frequency (see section 4.8).
Figure 37 shows that qualitatively similar spectral features are observed for the
Stokes process both for normal resonant Raman scattering (RRS) and for surface-
enhanced resonant Raman scattering (SERRS). The di€ erences in the detailed
lineshape, peak frequencies, and relative intensities between the RRS and SERRS
spectra are further discussed below, because these di€ erences provide important
information about the electron± phonon coupling in SWNTs. It is also signi® cant
that the SERRS spectra for the Ag and Au ® lms in ® gure 37 are almost identical, as
discussed further below. The features in ® gure 37 which show only a small shift in
phonon frequency as a function of Elaser in the RRS spectra also show very small
frequency shifts between features in the RRS spectra and in the corresponding
SERRS spectra.
For laser excitation energies where only semiconducting nanotubes contribute to
the resonant Stokes spectra, the tangential phonon mode region yields Lorentzian
components at 1563 cm ¡1 , 1591 cm ¡1 and 1601 cm ¡1 for both the normal resonant
Raman spectra [94] and for the corresponding SERRS spectra, showing the same
peak frequencies and relative intensities for the three constituent Lorentzian com-
762 M. S. Dresselhaus and P. C. Eklund

1.96 eV (a)
(a )

1591

RRS 1540
1515

0
11400
4 00 14 5 0 1500
1 5 00 1550 1600
1 60 0 1650 1700
17 0 0

(b)
(b )

1540
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

1591
SERRS 1515
(Au)

0
1400 1500 1600 1700

(c)
(c)

1540
1591
SERRS 1515
(Ag)

0
11400
4 00 14 5 0 1500
1 5 00 1550 1600
1 60 0 1 6 50 1700
17 0 0

Raman shift (cm-1)


Figure 38. Deconvolved spectra of the tangential vibrational band obtained with laser
excitation at Elaser ˆ 1:96 eV for (a) normal resonant Raman spectroscopy, (b) SERS
on an Au substrate and (c) SERS on an Ag substrate. For all three traces, a
Lorentzian lineshape analysis was made in terms of the same Lorentzian oscillators at
1515 cm ¡1 , 1540 cm ¡1 and 1580 cm ¡1 for the metallic nanotubes and at 1563 cm ¡1 ,
1591 cm ¡1 and 1601 cm¡1 for the semiconducting nanotubes. The two dominant
components in the metallic nanotube regime (at 1515 and 1540 cm¡1 ) and the
strongest component in the semiconducting nanotube regime (at 1591 cm¡1 ) are
explicitly labelled and the sums of the six Lorentzian components are also explicitly
shown by the dashed curves [115].

ponents. The same conclusions are reached for other Elaser excitation energies for
which only semiconducting nanotubes contribute resonantly to the anti-Stokes RRS
and SERRS spectra.
For Elaser within the 1:8 < Elaser < 2:0 eV range, the RRS and the SERRS spectra
for a SWNT sample with dt ˆ 1:35 § 0:20 nm both can be simply described by the
appearance of Lorentzian components at 1515 cm ¡1 , 1540 cm ¡1 and 1581 cm ¡1 ,
which are identi® ed (see ® gure 8) with resonantly enhanced features from metallic
M
nanotubes in resonance with the E11 …dt† transition [94]. In this case, signi® cant
changes can be seen when comparing the RRS spectra to the corresponding SERRS
spectra, as shown in ® gure 38 at Elaser ˆ 1:96 eV. First we note the similarity between
the SERS spectra for nanotubes on two di€ erent metal substrates, Au and Ag,
indicating that the dominant SERS enhancement factor is associated with the
electromagnetic mechanism which is not sensitive to the metal substrate species.
Similar results are obtained for other Elaser values within the metallic window (see
Phonons in carbon nanotubes 763

section 4.2). Figure 38 further shows that the contributions of certain Lorentzian
components for the metallic nanotubes (1540 and 1515 cm ¡1 components) are much
more pronounced for the SERRS spectra than for the RRS spectrum at
Elaser ˆ 1:96 eV (and also at other values of Elaser within the metallic window),
though the peak frequencies for the three metallic Lorentzian components appear to
be the same. The same conclusions are reached if we compare the RRS and SERRS
anti-Stokes spectra when the incident and/or scattered photons are in resonance with
metallic nanotubes in the anti-Stokes process [115]. It is interesting that the relative
intensity of the Lorentzian component at 1581 cm ¡1 , which is also associated with
metallic nanotubes, is more similar between the RRS and SERRS spectra than are
the 1540 cm ¡1 and 1515 cm ¡1 features. As discussed below, the di€ erent relative
intensities between the RRS and SERRS spectra for the various Lorentzian
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

components (1515 cm ¡1 , 1540 cm ¡1 and 1581 cm ¡1 ) associated with metallic nano-


tubes are attributed to two factors: (1) the much larger magnitude of the SERRS
charge transfer enhancement factor in the metallic nanotubes, and (2) the stronger
electron ± phonon coupling that occurs in metallic nanotubes, and both of these
processes are sensitive to the phonon mode symmetries and normal mode displace-
ments [115].
We now discuss the very large SERS enhancement factors that occur in SWNTs
and the unique aspects of the SERS phenomena in SWNTs relative to other systems
in which the SERS phenomenon is observed. Figure 39 shows a comparison between
the surface-enhanced (SERRS) and normal resonance Raman scattering (RRS) for
both the Stokes (S) and anti-Stokes (AS) processes in SWNTs (with dt ˆ 1:35 §
0:20 nm) [130]. The spectra are taken at Elaser ˆ 1:49 eV excitation and in the 1500±
1650 cm ¡1 phonon frequency range (tangential C± C stretching mode) [130]. In
® gure 39 we see that the Stokes spectra at Elaser ˆ 1:49 eV are dominated by
contributions from semiconducting nanotubes, whereas the anti-Stokes spectra are
dominated by contributions from metallic nanotubes [130], as discussed in
section 4.5. Also shown in ® gure 39 are the peak heights PS and PAS for the Stokes
and anti-Stokes signals which are plotted as a function of incident laser intensity IL
for Elaser ˆ 1:49 eV.
Figure 39 shows that the RRS and SERRS Stokes spectra at Elaser ˆ 1:49 eV are
quite similar to one another, consistent with the above discussion of the RRS and
SERRS spectra for semiconducting nanotubes. On the other hand, the RRS and
SERRS anti-Stokes spectra which are, in accordance with ® gure 8, dominated by
contributions from metallic nanotubes, have some similarity to one another, but
certainly di€ er in detail, and are very di€ erent from their Stokes counterparts at the
same Elaser value for this SWNT sample (dt ˆ 1:35 § 0:20 nm), consistent with the
® ndings in ® gure 38 and with the discussion in section 4.5.
In normal resonance Raman scattering experiments (RRS), the observation of
relatively strong anti-Stokes signals are related to sample heating through the
thermal excitation of a signi® cant population of the ® rst vibrational phonon level,
whereas in the SERRS experiment for nanotubes, the heating of the nanotubes is
presumably very small due to the very high thermal conductivity of SWNTs and the
e cient heat transfer to the silver substrate. This can be concluded from ® gure 39
(upper right), which displays a plot of the surface-enhanced Stokes signal versus IL
the excitation laser intensity at Elaser ˆ 1:49 eV. For IL between 2 and ¹5 MW cm ¡2 ,
the Stokes SERRS signal increases linearly with IL and no frequency shift
(1592 cm ¡1 ) is observed for the Stokes line of the semiconducting nanotubes up to
764 M. S. Dresselhaus and P. C. Eklund

Anti- Stokes
Stokes
PAS (arb. units) 1588 cm-1 1585 cm-1

PS (arb. units)
1592 cm-1

4 8 12 4 8 12
I L(MW/cm 2 ) I L(MW/cm 2)
1541 1592
1586
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

Raman Signal (arb. units)

Anti- Stokes
Stokes

1580
SERRS 1595
SERRS
1540

RRS RRS

1750 1600 1450 14001450 1500 1600


1600 17001750 1800
1800 1700 1600 1500 1400
Raman Shift (cm-1)
Figure 39. Lower plots show SERRS and RRS anti-Stokes and Stokes spectra of SWNTs
measured with Elaser ˆ 1:49 eV excitation. On the top of the ® gure are surface-
enhanced anti-Stokes (left) and Stokes (right) signal heights PAS and PS plotted as a
function of excitation laser intensity IL for Elaser ˆ 1:49 eV, and using nanostructured
Ag particles for the metal substrate. The lines display linear (right) and quadratic
(left) ® ts to the experimental data for PS and PAS , respectively (see equations (28) and
(29)). The centre of the Stokes SERRS band shifts abruptly from 1592 cm ¡1 to
1588 cm ¡1 to 1585 cm ¡1 with increasing IL (indicated by the vertical dashed lines), as
the PS data changes from one slope to another. At high laser excitation intensity
(IL ¹ 10 MW cm ¡2 ), irreversible destruction of the carbon nanotubes starts [130].

about 5 MW cm ¡2 , indicating very little increase in temperature [106], too little to


result in a measurable frequency shift, as is observed for example in the dependence
of the normal resonance Raman spectra (RRS) of SWNTs on laser intensity due to
heating [106]. The Stokes shift in ® gure 39 (top) changes abruptly from 1592 cm ¡1 to
1588 cm ¡1 at IL ¹5 MW cm ¡2 and then to 1585 cm ¡1 at IL ¹8 MW cm ¡2 . At
IL ¹10 MW cm ¡2 , destruction of the nanotubes starts. The abrupt shifts in the
Raman frequency have been interpreted in terms of a mode switching between the
various phonon modes (E2g , A1g and E1g ) that constitute the tangential band, and is
further discussed in section 4.7 [130]. This mode switching can appear through a
phonon ± phonon coupling mechanism that becomes important at large values of IL
[2]. The plot for PS versus IL in ® gure 39 shows a linear dependence of the Stokes
Raman signal PS on laser intensity IL , for each of the three regions of IL delineated
in ® gure 39, but the slopes of the PS versus IL curves in ® gure 39 are found to be
Phonons in carbon nanotubes 765

di€ erent for each of the three regions characterized by the constant mode frequencies
of 1592 cm ¡1 , 1588 cm ¡1 and 1585 cm ¡1 [130].
Complementary information is obtained by comparison of the anti-Stokes RRS
and SERRS spectra. The anti-Stokes SERRS signal for Elaser ˆ 1:49 eV in ® gure 39
has been identi® ed with metallic nanotubes (see ® gure 8) and appears as a broad
band from 1540± 1580 cm ¡1 , and its signal PAS is nonlinearly dependent on laser
intensity IL between 2 and ¹10 MW cm ¡2 (see ® gure 39 top left). This nonlinearity
can be due to two e€ ects: increasing temperature with increasing laser intensity, and/
or vibrational pumping. Through the second mechanism, a very strong Raman e€ ect
populates excited vibrational levels in excess of the normal Boltzmann population
[123]. Vibrational pumping can be observed only at extremely large e€ ective Raman
cross-sections and is an indication of a very high SERRS enhancement level
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

[123, 126]. The carbon nanotube system is an ideal system to observe such phenom-
ena because of the very high thermal conductivity of carbon nanotubes, which
suppresses heating e€ ects, and their robustness at high temperatures which allows
very high laser excitation intensities IL to be used before the nanotubes are
destroyed.
A simple theoretical estimate for the Stokes and anti-Stokes power levels PS and
PAS can be derived from a rate equation model describing the population/
depopulation of the ® rst excited vibrational level by SERRS Stokes and anti-Stokes
transitions [123]. In the steady state, PAS and PS can be expressed as
- tang
¡h!
PAS ’ exp … kT …nL † †‡ ½1 nL ¼SERS
M NM nL ¼SERRS
M ; …28†

PS ’ NSC nL ¼SERRS
SC ; …29†
where nL is the photon ¯ ux density of the laser, IL ˆ nL Elaser =c is the intensity of the
laser beam, c is the velocity of light, NM and NSC are the numbers of metallic (M)
and semiconducting (SC) nanotubes in the vibrational ground state that interact
with the laser beam, ¼SERS
M is the e€ ective non-resonant SERS cross-section for the
metallic nanotubes, the superscript SERRS refers to the resonant surface-enhanced
Raman process, !tang is the phonon frequency, ½1 is the phonon lifetime, and T is the
sample temperature. The ® rst term in the prefactor in equation (28) in brackets
corresponds to the thermal population of the ® rst vibrational level for the tangential
phonon band. Whereas the Stokes power PS remains linearly dependent on IL up to
5 MW cm ¡2 , the experimental anti-Stokes signal PAS was found to have a quadratic
dependence on laser excitation intensity because of the nL dependence of the second
term in the prefactor and the nL dependence of the common factor on the right of
equation (28).
Figure 39 (top left) shows a ® t of the experimental data to PAS in equation (28).
The results of this ® t indicate that the large anti-Stokes signal in the SERRS
spectrum cannot be explained by laser heating of the sample (the ® rst term in the
prefactor), but rather vibrational pumping (the second term in the prefactor) was
found to make the dominant contribution to the observed nonlinear dependence
of the anti-Stokes signal PAS on IL . This nonlinear vibrational pumping mechanism
for converting photon energy into phonons provides a mechanism for achieving
large intensities for the anti-Stokes process for metallic SWNTs [130]. From the
® tting parameters, the product of the cross-section and the lifetime for the non-
resonant Stokes Raman process for metallic nanotubes is estimated to be
766 M. S. Dresselhaus and P. C. Eklund

Raman Signal (arb. units) 1596 1591

100 nm
D w = 9.5 D w = 17
1596 1589

D w = 14 D w = 27
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

1400 1450
1450 1500 1550 1600
1600 1650 17001750
1750 18001400 1450
1450 1500 1550 1600
1600 1650 17001750
1750 1800
-1
Raman Shift (cm )
Figure 40. Typical fractal colloidal silver clusters (see inset) and selected SERRS Stokes
spectra of the tangential band, with a linewidth D ! (in cm¡1 ) as small as 9.5 cm ¡1 ,
collected from a very few nanotubes adsorbed at such a cluster area. The spectra were
taken at Elaser ˆ 1:49 eV laser excitation with a 1 mm spot size [130].

½1 ¼SERS
M ¹ 5 £ 10 ¡28 cm 2 s. Assuming a vibrational lifetime on the order of 10 ps for
the phonons of the metallic nanotubes (in good agreement with the measured
linewidth in ® gure 40), an e€ ective Raman cross-section on the order of about
10¡16 cm2 is inferred. Starting from an estimated non-resonant Raman cross-section
in the range 10 ¡30 ± 10 ¡28 cm2 for the metallic nanotubes [131], an e€ ective cross-
section of 10 ¡16 cm2 could be reached by SERRS enhancement factors on the order
of 1012 ± 1014 , as are also observed in single molecule studies [123 ± 127]. Enhancement
factors of this magnitude provide hope for observing Raman spectra from very small
numbers of nanotubes, perhaps down to a single nanotube. The large SERS cross-
section ¼SERS
M is associated with extremely large ® eld enhancement e€ ects that can,
for example, occur on fractal Ag nanostructures, especially at longer wavelengths
[128, 132].
The inset to ® gure 40 shows a typical collection of colloidal silver clusters of
nanoscale particles used for obtaining very large enhancement SERRS spectra,
where an extremely low concentration of SWNTs (probably only a few bundles of
SWNTS) were deposited on Ag colloidal particles. SERRS spectra could be
measured in only a few places on the entire sample, only when a SWNT happened
to be attached to a small colloidal Ag particle [130]. In these cases, weak SERRS
Stokes spectra were measured using 1 s collection times. Shown in ® gure 40 are a few
selected SERRS spectra taken at di€ erent spots over the entire sample, which
consisted of many colloidal silver clusters which had no nanotubes attached to them
and a very few that had attached nanotubes. Of particular interest are the small
linewidths observed in some of the spectra in ® gure 40 for the tangential band of the
Stokes spectra. The smallest linewidths are much smaller than those typically
measured in RRS or in SERRS from nanotube bundles ( D ! ¹25 cm ¡1 ). The
narrowest measured linewidth for the tangential band in ® gure 40 is 9.5 cm ¡1 , which
is very close to the expected natural linewidth of the tangential C± C stretching mode
in semiconducting SWNTs deduced from a natural full width at half maximum
Phonons in carbon nanotubes 767

(FWHM) linewidth of 4 cm ¡1 and from the theoretical prediction that the three
unresolved E2g , A1g and E1g mode frequencies [2], that contribute to the tangential
band at ¹1590 cm ¡1 , di€ er from one another by only 6 cm ¡1 (see ® gure 15). These
linewidth comparisons indicate that nanotubes of a unique diameter (or possibly a
single nanotube bundle) contribute to the Raman signal in ® gure 40. The strongly
con® ned electromagnetic ® eld on a silver cluster may selectively probe small numbers
of nanotubes adjacent to the interface. Even stronger SERRS enhancement is
expected at low Raman frequencies [128, 132]. Thus it may be possible for SERRS
to reveal the radial breathing mode spectra for individual SWNTs, free from the
inhomogeneous broadening e€ ects observed in the radial breathing mode spectra in
normal resonant Raman scattering [133].
As mentioned above, there are two major contributions to the surface enhance-
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

ment of the resonant Raman scattering of carbon nanotubes adsorbed on metallic


surfaces: the `electromagnetic’ mechanism and the `chemical’ mechanism, based on a
charge transfer between the metallic surface and the nanotubes. The `electromag-
netic’ mechanism is due to the enhanced electromagnetic ® elds associated with
surface plasmons at or near nanometre size metallic silver and gold particles.
Particularly strong ® eld enhancement (up to 10 to 12 orders of magnitude) can exist
for metallic substrates showing fractal cluster structures [128, 132, 134]. Theoretical
calculations predict that these SERRS-active substrates should exhibit a particularly
strong enhancement at longer wavelength (near infrared) excitation [132, 135]. In
general, the electromagnetic enhancement mechanism is independent of the chemical
nature of the species and is expected to be of similar magnitude for semiconducting
and metallic nanotubes, and also to be the same for all vibrational modes. The
existence of such large enhancement levels has been shown for both semiconducting
and metallic nanotubes [130]. The second enhancement mechanism, the charge-
transfer or chemical e€ ect, strongly depends on the electronic structure of the
nanotubes and the electronic interaction between the nanotubes and the metal
substrate. Therefore, we expect the charge transfer mechanism to sensitively
di€ erentiate between the very di€ erent 1D density of states of metallic and
semiconducting nanotubes. Fitting equations (28) and (29) to the observed depen-
dence of the signal intensities PS and PAS on laser power IL indicates that at the
highest IL values used, the charge transfer or chemical e€ ect contributed about two
orders of magnitude enhancement to the signal for the metallic nanotubes and very
much less for the semiconducting nanotubes [130]. The di€ erent SERRS enhance-
ment factors observed for the 1540 cm ¡1 , the 1515 cm ¡1 and the 1581 cm ¡1
Lorentzian components of the SERRS spectra for metallic nanotubes relative to
the RRS spectra are attributed to the charge transfer enhancement mechanism,
which is operative for the metallic nanotubes, in addition to the strong electro-
magnetic ® eld e€ ect discussed above. Through the charge transfer or chemical e€ ect
we can account for the di€ erent behaviour between the SERRS spectra of metallic
and semiconducting nanotubes, which is attributed to major di€ erences in their
electronic structures and in their 1D density of states near the Fermi level, and to the
stronger electron± phonon coupling in metallic nanotubes compared to semiconduct-
ing nanotubes.

4.7. Polarization studies in Raman scattering


The fundamental structural anisotropy of carbon nanotubes, due to their high
aspect ratio (length/diameter), suggests the importance of polarized Raman spectro-
768 M. S. Dresselhaus and P. C. Eklund

scopy studies. Only limited experimental data [136, 137] and theoretical predictions
[2, 42] are presently available regarding polarization phenomena in the resonant
Raman scattering from carbon nanotubes. The success of the initial reports suggest
that this subject will be an active research area for future activity.
Lattice dynamics calculations based on a bond polarization model have been
carried out under non-resonant scattering conditions for both the relative intensities
of various Raman-active modes in isolated (10, 10) armchair nanotubes for di€ erent
polarizations and polarization geometries and for their angular dependence [2, 43,
138]. Calculations for nanotubes with other …n;m † values, such as for zigzag and
chiral nanotubes, are needed as well as calculations for general …n; m† nanotubes
under resonant conditions, because most experimental polarization studies are likely
to be carried out under resonance conditions, and for samples containing nanotubes
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

with di€ erent chiralities.


The calculated sample orientation dependence of the Raman intensity shows
that, not only the symmetry, but also the direction of the normal mode displace-
ments, are important in determining the angular dependence of the polarization
intensities. As shown below, this angular dependence is special for carbon
nanotubes, and can in fact be used to identify the symmetry assignments for some
of the Raman-active modes.
The Raman intensities for the various Raman-active modes in carbon nanotubes
have been calculated at a phonon temperature T using a non-resonant bond
polarization theory where the temperature dependence of each normal mode is
given by a Bose± Einstein distribution function for the phonons [2]. The eigenfunc-
tions for the various vibrational modes were calculated numerically at the G-point
(k ˆ 0). The mode intensities were speci® cally calculated for two possible geometries
for the polarization of the light: the VV and VH con® gurations. In the VV
con® guration, the incident and the scattered polarizations are parallel to each other,
while for the VH con® guration, the incident and scattered polarizations are
perpendicular to each other. Generally, the cross-section for Raman scattering is a
function of the photon scattering angle. Since the formulae for the bond polarization
theory consider only s-scattered waves [2], the calculated results cannot distinguish
between forward and backward photon scattering. The orientation of the nanotube
axis is speci® ed relative to the polarization direction V.
In ® gure 41, the calculated Raman intensities for the most important Raman-
allowed modes are shown for (10, 10) armchair, (17, 0) zigzag and (11, 8) chiral
Ê , 6.66 A
nanotubes, all having similar radii, which are, respectively, 6.78 A Ê and 6.47 A
Ê
[2]. Results are shown in each case for the VV and VH con® gurations discussed
above. The calculated Raman intensities for nanotube bundles containing aligned
SWNTs but oriented along random directions are found by averaging over the
sample orientation of the nanotube axis relative to the Poynting vector of the light,
in which the average is calculated by summing over many ( ¹50) sample orientations
using the angular dependence results discussed below. As mentioned above, the
presently available polarization calculations (such as in ® gure 41) neglect corrections
due to the inter-tube interactions within a nanotube bundle which lower the
symmetry in-plane normal to the nanotube axis, and therefore these predictions
should be used with caution when interpreting Raman spectra for nanotube bundles.
The calculations in ® gure 41 show that the most intense mode is the low frequency
radial breathing mode with A1g symmetry in the VV geometry. The two low
frequency E2g modes have lower intensities and ® nally the low frequency E1g mode
Phonons in carbon nanotubes 769

VV
VV VH
VH

(a) A1g (10,10)


165
E1g
E2g

17
1585
A1g
E2g

17
r = 6.78 A
E2g E1g 1585
E1g 1587 E1g E2g
368 E2g 118 A1g A1g1587
118 368
1591 E2g 1591
165

0 400 800 1200 1600 0 400 800 1200 1600


units. ]

(10,10)
units)

(b) A1g
(17,0)
[ arb
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

168 E1g
Intensity(arb.

E2g
1587

r = 6.66 A
18 E2g
A1g
Raman intensity

18
E2g 1590
E1g E1g
E1g 1587
E2g E2g
376 A1g A1g1590
119 119
1571 376 E2g1571
168
Raman

0 400 800 1200 1600 0 400 800 1200 1600

(c) A1g
(11,8)
E1g
173
E2g 1585
E2g
19 A1g

1586
19
r = 6.47 A
E2g E1g E2g E1g1585
E1g
E2g A1g1586
386 123 A1g 386
123 1591 E2g1591
173

0 400 800 1200 1600 0 400 800 1200 1600


(11,8)
-1
Raman Shift
Raman cm -1) ]
shift[ (cm

Figure 41. The polarization dependence of the Raman scattering intensity of selected low
and high frequency Raman-allowed modes for (a) (10, 10) armchair, (b) (17, 0) zigzag
and (c) (11, 8) chiral nanotubes for which the nanotube radii are given on the right.
The left column is for the VV scattering con® guration and the right column is for the
VH con® guration [2, 43].

is predicted to have even lower intensity in the VV con® guration. However, for the
VH geometry, the E1g and E2g modes are all predicted to have a modest reduction in
intensity relative to the VV con® guration, but the radial breathing A1g mode is
expected to have a very low intensity in the VH con® guration.
In contrast to the low frequency A1g radial breathing mode, the high frequency
A1g mode, at 1587 cm ¡1 which is associated with the tangential band, does not show
such a large suppression between the VV and VH geometries, because of the
connection of this A1g mode through zone folding to the Raman-active E2g mode
of graphite which has intensity only in the VH geometry. It is also reasonable that
the tangential motions of the higher frequency Raman modes of a nanotube should
have a large Raman intensity, because of the relation of these modes to the E2g mode
of graphite. However, the directions of the carbon atom motions of the tangential
A1g mode are di€ erent for armchair and zigzag nanotubes, since the CˆC bond-
stretching motions can be seen in the horizontally and the vertically vibrating Cˆ C
bonds for armchair and zigzag nanotubes, respectively. Thus the curvature of the
770 M. S. Dresselhaus and P. C. Eklund

nanotube a€ ects the frequency of these modes di€ erently. Although these higher
frequency modes (! > 1500 cm ¡1 ) are di cult to distinguished from one another
because of their very similar frequencies, it should be possible to identify the di€ erent
modes experimentally, once puri® ed aligned samples become available, through
studies of the angular dependence of the Raman intensities, which are discussed
below. The calculated results show almost no intensity for the intermediate
frequency Raman modes around 1200± 1500 cm ¡1 [42]. The Raman experiments on
single-wall nanotubes show some weak features [28], which might arise from a
lowering of the symmetry of the nanotube due to nanotube deformations or due to
inter-tube interactions, or may be associated with ® nite nanotube lengths [79].
As discussed below, angular dependent polarization studies of SWNTs are
expected to provide information about the identity of speci® c features in the
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

experimental Raman spectra with mode symmetries, thereby allowing interpretation


of the experimental spectra in terms of theoretical calculations of mode intensities.
Theoretical angular dependent calculations of the intensities of the various Raman-
active modes have been carried out only for the (10, 10) armchair nanotubes and the
results are plotted in ® gure 42 as a function of nanotube orientation with respect to
the polarization direction. These three rotations of the nanotube are denoted by
³i …i ˆ 1;2;3), as shown on the right hand side of ® gure 42. Here ³1 is the angle that
the nanotube axis makes with the z axis as the nanotube axis rotates from the z axis
to the x axis in the xz plane normal to the y axis. The polarization vectors of the light
lie along the z and x axes, respectively, for the V and H polarizations. Likewise ³2 is
the angle that the nanotube axis makes with respect to the z axis as the nanotube axis
rotates in the yz plane from the z axis to the y axis. The angle ³3 denotes the rotation
angle of the nanotube around its own axis. The x;y;z axes are de® ned so that the two
inequivalent carbon atoms for the armchair nanotubes are both placed along the x
axis when ³3 ˆ 08 (see ® gure 1). In an actual `aligned’ sample of SWNTs, the axes of
the nanotubes are aligned but the tube orientations with respect to ³3 are random.
Figure 42 shows that the angular dependence of the Raman intensities as a function
of ³1 and ³2 are somewhat di€ erent from each other for the VV con® guration and
very di€ erent from each other for the VH con® guration. We further note that the
ten-fold symmetry axis (C10 ) of the (10,10) nanotubes which is not a symmetry
operation that is compatible with the Cartesian axes.
First we show how ® gure 42 can be used to identify mode symmetries in a Raman
polarization experiment using aligned nanotubes. In ® gure 42 we see that the Raman
intensity for the A1g mode at 1587 cm ¡1 as a function of ³1 for the VV con® guration
has a maximum at ³1 ˆ 0, and ³2 ˆ 0, and is the only tangential mode with
appreciable intensity for this polarization geometry. The tangential A1g mode is
predicted to have no intensity in the VH con® guration for ³1 ˆ 0, ³2 ˆ 0. On the
other hand, the E1g mode at 1585 cm ¡1 is predicted to have a maximum intensity in
the VV con® guration at ³1 ˆ 458, but in the VH con® guration, the E1g mode at
1585 cm ¡1 is predicted to have a maximum intensity at ³1 ˆ 0, ³2 ˆ 0 and a
minimum intensity at ³1 ˆ 458 or ³2 ˆ 908. These polarization properties should
allow us to distinguish these two modes with close-lying frequencies from each other
experimentally, if we have an axially aligned nanotube sample. Also the E2g mode at
1591 cm ¡1 should be distinguishable from the A1g and E1g modes, since the E2g mode
is expected to have a maximum intensity at ³1 ˆ 908 and ³2 ˆ 908 in the VV
geometry; the E2g mode is the only tangential mode that has appreciable intensity
at ³2 ˆ 908 in the VH con® guration.
Phonons in carbon nanotubes 771

VV VH

A1g1587
z

E1g A1g1587 q 1
E1g1585 1585 y
A1g165
x
E1g118
E2g1591 E1g118
A1g165
E2g368 E2g1591
0 30 60 90 0 30 60 90
E2g368
q 1 [ deg. ] q 1 [ deg. ]

A1g1587
z
q 2
R am an intensity (arb.units)
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

E1g1585 y
A1g165
E2g 17
x
E1g118 E2g368
E1g118 E2g1591
E2g 368 E2g1591
0 30 60 90 0 30 60 90
q 2 [ deg. ] E1g1585 q 2 [ deg. ]
A1g1587 z

E1g 1585
y
x q 3
A1g165 E1g118

0 30 60 90 0 30 60 90
q 3 [ deg. ] q 3 [ deg. ]
Figure 42. The polarization dependence of the Raman intensities as a function of the
orientation of the nanotube axis for a (10, 10) armchair nanotube in three principal
planes (see text). The left and right hand ® gures give the angular dependence of the
mode intensities (labelled by symmetry type and frequency) corresponding to the VV
and VH polarizations (see text) [43].

The lower frequency Raman-active modes should also be distinguishable in


accordance with their polarization behaviour. In ® gure 42, the two E2g modes at 368
and 1591 cm ¡1 have almost the same relative intensities in the VV con® guration.
However, the ³2 dependence of the VH con® guration clearly distinguishes the
di€ erent polarization behaviours between these two E2g modes [43]. It should be
noted that the polarization characteristics presented in ® gure 42 are valid only for
armchair nanotubes, and these characteristics are expected to di€ er somewhat for
di€ erent nanotube chiralities.
Even the nanotube modes belonging to the same irreducible representation do
not always have the same angular dependence with regard to the polarization of the
light. For example, the intensity of the A1g mode at 165 cm ¡1 is expected, on the basis
of ® gure 42, to show a di€ erent angular dependence from that of the A1g mode at
1587 cm ¡1 [43]. Also the symmetry analysis for the lower symmetry E1g and E2g
modes may be di cult, even if we can have an aligned nanotube sample for which
the direction of the carbon atoms is ordered, since the 10-fold symmetry of the
(10,10) nanotube does not satisfy the symmetry of the triangular nanotube lattice so
772 M. S. Dresselhaus and P. C. Eklund

that averaging over ³3 is necessary to describe a nanotube bundle. Even for an


aligned SWNT bundle where the SWNTs make an arbitrary angle with the x and z
axes, since the A1g mode at 1587 cm ¡1 is independent of ³3 , the signal for the A1g
tangential mode will be clearly seen in the VV con® guration. The (9,9) armchair
nanotube is of special interest, since it is one of a few examples where the n-fold
symmetry of the nanotube matches the triangular lattice, so that detailed angle-
dependent selection rules can be expected in this case [2].
The only experimental polarized Raman spectra reported to date on SWNTs are
for nanotubes synthesized in the porous channels of aluminophosphat e single
crystals [137]. From the frequency of the Raman feature that was identi® ed with a
radial breathing vibration (¹530 cm ¡1 ), nanotube diameters of < 5 A Ê were inferred
[81], smaller than previously reported in the literature for any SWNT [1, 2], thereby
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

indicating the need for further characterization studies on these samples. The
intensity of the tangential Raman G-band was found to be sensitive to the
polarization geometry, with the tangential mode intensity being the strongest for
the incident and scattered optical E ® elds along the nanotube axis, in agreement with
the polarization e€ ects expected for aligned dipole antennas on the basis of
electromagnetic theory. Using a lineshape analysis for the tangential band based
on three Lorentzian components (see section 4.2), the following peak frequencies,
mode symmetries, and FWHM linewidths were reported: 1585 (E2g ; 36), 1599 (E1g ;
56) and 1615 (A1g ; 30) cm ¡1 . While using the same symmetry assignment arguments
as Kasuya et al. [93], the relative frequency di€ erences obtained by Sun et al. [137]
were much smaller than would be implied by the results in ® gure 26, for nanotubes
with such small nanotube diameters [137]. Con® rmation of these results on well
characterized, aligned SWNTs is needed.
The most detailed experimental polarization results for Raman spectra presently
available are for high purity aligned arrays of multi-wall carbon nanotubes
(MWNTs) [136]. These aligned MWNTs were prepared on silica substrates from
the thermal decomposition of a ferrocene± xylene mixture and had an average
nanotube diameter dt ¹25 nm, with the MWNTs aligned normal to bare silica
substrates [139]. The nanotube diameters in this MWNT sample are probably large
enough so that the resonance Raman scattering e€ ect is substantially weaker than is
typical for SWNTs (1 < dt < 2 nm). This is one reason why we might expect the non-
resonant bond polarizability calculation discussed above to be more applicable to
describe the polarized Raman spectrum for a MWNT sample than for typical SWNT
samples. The thickness of the aligned nanotube bundle was intentionally kept below
1 mm so that the focused laser excitation beam (beam diameter d ¹ 1 mm) over® lled
the aligned MWNTs during the polarized Raman scattering measurements. This
study measured polarization e€ ects associated with …k; k† and …k;?† scattering for
the tangential G-band (1500 ± 1650 cm ¡1 discussed in section 4.2) and for the so-called
D-band (1300 ± 1400 cm ¡1 discussed in section 4.8), as well as providing con® rmation
for a predicted angular dependence of the scattered intensity as a function of the
angle ³m between the polarization direction and the nanotube axis [136]. The angle
³m is the angle that was measured in the polarization experiment.
The polarized spectra in ® gure 43 for four polarization geometries, labelled …XY †,
…Y X†, …Y Y † and …XX†, are shown for the Raman intensities of the D-band and the
tangential G-band for these MWNTs in the backscattering con® guration, with the
greatest intensities observed for the …XX† polarization geometry. The geometry and
notation used to describe these polarization studies on MWNTs are shown in the
Phonons in carbon nanotubes 773
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

Figure 43. Polarized Raman spectra for MWNTs taken at 514.5 nm (2.41 eV) for 4
scattering geometries. The peak frequencies for the various features are in cm ¡1 and
the FWHM linewidths in cm¡1 are given in parentheses. The inset de® nes the
polarization geometries for the aligned nanotubes that lie parallel to the x axis but
there is no preferred angular alignment of the nanotubes in the plane normal to the
nanotube axis [136].

inset to ® gure 43, where x;y;z refer to the coordinates of the MWNTs and X;Y ;Z
refer to the laboratory frame of the light. In this notation, a backscatterin g Raman
-
experiment is described by Z …XY †Z to indicate light incident along the Z axis with
the electric vector in the X direction for the incident light beam, while the Poynting
vector for the scattered light is along the ¡Z direction, with the electric vector for the
scattered light beam in the Y direction. For the polarization results on the MWNTs
shown in ® gures 43 and 44, the backscattering geometry was used with light normal
to the nanotube axis, so that the abbreviated notation …XY †, giving the directions of
the incident and scattered electric ® eld polarizations, is used to label the polarization
geometries shown in these ® gures.
A Lorentzian lineshape analysis for the MWNTs spectra in ® gure 43 shows four
peaks at 1354, 1576, 1583 and 1624 cm ¡1 in almost all the polarization geometries,
but with very di€ erent polarization-dependent relative intensities [136]. From
® gure 43, the experimental polarized integrated intensity ratios for the tangential
band (at 1584 cm ¡1 for the XX spectrum) are IXX =IXX :IY Y =IXX :IXY =IXX :IY X =IXX =
1.00:0.29:0.19:0.39 for this MWNT sample. These polarization results can in part be
774 M. S. Dresselhaus and P. C. Eklund
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

Figure 44. Dependence of the measured Raman intensity for VV scattering as a function of
the angle ³m between the polarization direction and the nanotube axis (see inset to
® gure 43). The inset to this ® gure shows the dependence of the experimental
tangential band intensity ratio R ˆ IVV …³m †=IVV …0† (open and ® lled circles) as a
function of ³m in comparison to theory (® gure 42) for SWNTs [79, 136].

explained by the nanotube geometry which gives rise to slightly di€ erent force
constants along the nanotube axis relative to the circumferential direction, where the
nanotube curvature reduces the force constant. This anisotropy in the force constant
accounts physically for the frequency di€ erence between the components at
1575 cm ¡1 (attributed to vibrations in the circumferential direction) and at
1584 cm ¡1 (attributed to vibrations along the nanotube axis) for the …XX† polariza-
tion (see ® gure 15). Similar frequency di€ erences are observed for the other
polarization geometries, and all the upshifted and downshifted modes in ® gure 43
average to 1580:8 § 0:8 cm ¡1 , in good agreement with the E2g2 graphite mode
frequency. Because of the small number of allowed k vectors in the circumferential
direction, we expect the frequency di€ erence between the lowest and middle
frequency components to increase with decreasing nanotube diameter [93]. For
example, SWNTs with diameters of dt ¹1.4 nm show these two spectral features [28]
to be at 1567 cm ¡1 and 1593 cm ¡1 (see section 4.2), and these two mode frequencies
also average to ¹1580 cm ¡1 . The feature near 1620 cm ¡1 which is quite pronounced
in the spectra for MWNTs, and not present in SWNTs, is associated with the
maximum in the graphene 2D phonon density of states and is called the D 0 band
[116].
No theory has yet been developed to describe polarization e€ ects for MWNTs.
Therefore the experimental polarization results for the tangential band in ® gures 43
Phonons in carbon nanotubes 775

and 44 have been interpreted in terms of the theory developed for (10, 10) SWNTs
[43] as a ® rst approximation. The theory of polarization e€ ects in single-wall
nanotubes was ® rst used to distinguish between the contributions to the tangential
band from modes with A1g , E1g and E2g symmetry (see ® gure 15) [43], which
according to bond polarization model calculations [43, 79] for a (10,10) SWNT
predict Raman-active frequencies at 1587, 1585 and 1591 cm ¡1 , respectively, all
occurring within a range of 6 cm ¡1 , which is much smaller than the observed
linewidth of the tangential G-band for Raman scattering, even for semiconducting
nanotubes [43, 137] (see section 4.2).
The experimental intensities for the …XX† and …Y Y † geometries in ® gure 43 and
in the inset to ® gure 44 correspond, respectively, to ³m ˆ 0 and ³m ˆ 908 for the VV
polarization con® guration, while the …XY † and …Y X† geometries, respectively,
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

correspond to ³m ˆ 0 and ³m ˆ 908 for the VH con® guration, where, V and H


denote vertical and horizontal, respectively (see ® gure 42). Calculations show that
the A1g mode is dominant for the tangential band in the …XX† con® guration, while in
the …Y Y † con® guration, both the A1g and E2g modes should contribute to the
scattering intensity. The intensity ratios predicted for the A1g mode at ³m ˆ 08, the
A1g mode at ³m ˆ 908, and the E2g mode at ³m ˆ 908 are, respectively, 2.91, 0.72 and
0.33. Thus the theoretical value for the intensity ratio for IXX =IY Y is 2.91/1.05 =
1.00/0.36, which is to be compared to the experimental values of IXX =IY Y = 1.00/
0.29. This observation would be consistent with the peak at 1584 cm ¡1 containing the
unresolved E2g and A1g modes of the tangential band.
Calculations for the expected Raman intensity for VH scattering [79] requires an
average to be done over ³3 (see ® gure 42). After carrying out the appropriate angular
averages, a theoretical value for the intensity ratio for the …XY † to …Y X† geometries
of IXY =IY X = 1.00/0.27 was obtained, which is to be compared with the experimental
value of 0.49/1.00. In summary, the theoretical predictions for the relative polariza-
tion intensities for the tangential band for SWNTs are IXX =IXX :IY Y =IXX :IXY =
IXX :IY X =IXX ˆ 1:00:0.36 :0.13:0.03 for the …XX†:…Y Y †:…XY †:…Y X† polarization geo-
metries [43], which are in rough agreement with the corresponding experimental
values for aligned MWNTs IXX =IXX :IY Y =IXX :IXY =IXX :IY X =IXX ˆ 1:00:0.29:0.19:0.39
[136]. The …XX† intensity depends only on the A1g mode, while the …Y Y † intensity
has contributions from both the A1g and E2g modes. Comparison between the
experimental …XX† and …Y Y † polarization data for MWNTs does not seem
consistent with the mode identi® cations made in ® gure 26 for SWNTs. Further
work is necessary to understand the polarization e€ ects and to make de® nitive
symmetry assignments of phonon modes in carbon nanotubes.
Based on theoretical considerations, the intensity of the tangential A1g mode as a
function of ³m is expected to exhibit a minimum at ³m ˆ cos¡1 …1=31=2 † ˆ 54:78 [43].
Figure 44 shows a collection of polarized Raman spectra that were obtained on
aligned MWNTs in the VV con® guration as a function of ³m . Clearly the experi-
mental polarized tangential band intensity ratio R ˆ IVV …³m †=IVV …0† exhibits a
minimum near ³m ˆ 558, in good agreement with the theoretical curve for the
intensity of the VV signal versus ³m shown in the inset to ® gure 44. It should be
noted that the intensity ratio R denoted by the open circle in the inset to ® gure 44
was obtained from the XX and Y Y data depicted in ® gure 43. The data set in
® gure 44 also suggests that the tangential band intensity is dominated by the
intensity of the A1g symmetry mode at 1584 cm ¡1 . It is interesting that the experi-
776 M. S. Dresselhaus and P. C. Eklund

mental D-band intensity is also a minimum near ³m ˆ 558. No theory is yet available
for the angular dependence of the D-band intensity.
It would also be interesting to measure the polarization e€ ects associated with the
A1g radial breathing mode in SWNTs, which was already considered in a preliminary
way [137]. Since the A and B carbon atoms within the unit cell of the graphene 2D
honeycomb lattice move similarly under breathing mode displacements, there should
be no intensity for the …XY † or …Y X† polarization geometries for isolated single-wall
nanotubes (see ® gure 42). Theory for (10, 10) SWNTs further predicts that the
…Y Y † intensity should be stronger than that for the …XX† intensity, with a
polarization intensity ratio …XX†:…Y Y † = 0.20:1.00 for the radial breathing mode.
Since the intensity of the radial breathing mode drops rapidly with increasing tube
diameter (where we note that the experimental intensity of this mode for a (20,20)
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

nanotube was observed to be vanishingly small) [82], polarization studies of the


radial breathing mode should be done on SWNTs with dt < 2 nm. In the actual
experiments, inter-tube coupling within a nanotube bundle (see section 3.3) and the
resulting lowering of the symmetry of the nanotube cross-sections may result in some
relaxation of the polarization e€ ects predicted for isolated SWNTs.
Returning to ® gure 43 we note that the optical absorption spectra of a single-wall
carbon nanotube, calculated for polarized light by Ajiki and Ando [140, 141], show
that the selection rules for optical absorption between the valence and conduction p
subbands with subband index n are D n ˆ 0 and D n ˆ 1 for polarization parallel and
perpendicular to the nanotube axis, respectively. The optical absorption for the
polarization perpendicular to the nanotube axis is predicted to be suppressed almost
completely when the depolarization e€ ect is taken into account. Thus the optical
absorption occurs only when the polarization is parallel to the nanotube axis for
inter-subban d transitions, with D n ˆ 0 for single-wall carbon nanotubes. The
depolarization e€ ect should, however, be relaxed in MWNTs, especially as the
diameter increases. On the basis of the depolarization e€ ect, we expect more e€ ective
excitation of resonant Raman scattering for the …XX† scattering geometry, and for
metallic nanotubes, for which the nanotube acts as a radiation pipe for the laser
excitation. The polarized VV Raman spectra for aligned MWNTs in ® gure 43 appear
to be consistent with the optical anisotropy present in the MWNTs. The discrepancy
between the experimental polarized Raman intensity ratios (regarding ® gure 43) and
theoretical predictions (® gure 42) may be due to a depolarization (antenna) e€ ect
operative also on the scattered light which tends to enhance the experimental Y X:XY
intensity ratio relative to the calculations.

4.8. D-band and G 0-band spectra


One of the interesting features of the Raman spectra in sp2 carbon materials is
the laser energy dependence of the frequency of the disorder-induced D-band which
is observed between 1250 and 1450 cm ¡1 . The D-band is activated in the ® rst-order
scattering process by the presence of in-plane substitutional hetero-atoms, vacancies,
grain boundaries or other defects and by ® nite size e€ ects, all of which lower the
crystalline symmetry of the quasi-in® nite lattice. The association of the D-band with
symmetry-breakin g phenomena results in a D-band intensity that is proportional to
the phonon density of states shown in ® gure 12 (c) for graphite, which should also be
applicable for (10, 10) SWNTs. A dominant feature of the second-order Raman
spectrum is the G 0-band which is the overtone of the D-band and appears in the
range 2500 ± 2900 cm ¡1 when the laser excitation energy Elaser is varied from 1 to
Phonons in carbon nanotubes 777

4.5 eV. Furthermore, the second-order G 0-band is observed even in the case of
crystalline graphite, where the disorder-induced D-band is absent, so that the G 0-
band is an intrinsic property of the 2D graphene lattice. It has been known for about
two decades [142], that there is a strong dependence of the peak frequency for the D-
band and for the G 0-band on Elaser , and this phenomenon occurs in a similar way in
all kinds of sp2 carbon materials, such as, graphon carbon black [143], hydrogenated
amorphous carbon [144], glassy carbon and crystalline graphite [145 ± 147], and
multi-component carbon ® lms [148]. It is believed that the D-band feature observed
in bundles of SWNTs has contributions from the SWNTs themselves, perhaps due to
a ® nite length e€ ect [79], as well as from other carbonaceous materials (e.g.
amorphous sp2 carbon coatings) present in the imperfectly puri® ed SWNT samples
that are now available.
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

Figure 45 illustrates the basic mechanism responsible for the laser excitation
dependence of the D-band and the G 0-band frequencies in sp2 carbons [117, 146,
147]. In the upper part of ® gure 45, we see that electronic transitions between the p
and p ¤ electronic states of 2D graphite with energies corresponding to visible
photons only occur in the vicinity of the K point in the Brillouin zone (BZ) through
which the Fermi level EF passes [2, 41]. The phonons associated with the D-band and
the G 0-band have the same wavevectors D q as the electronic states D k which are in
resonance with the laser. Moreover, it is argued that these phonons belong to the
optic branch that contains the zone centre E2g2 graphitic mode, represented by a
heavy curve in ® gure 45 [2, 41]. The reason why the phonons belonging to this
particular optic branch exhibit an especially large Raman cross-section compared to
other phonons with the same wavevector D q is attributed to their breathing-mode
displacements (see ® gure 46) which would be expected to show strong deformation
potential coupling to the electronic states [117]. Referring to ® gure 46, we see that for
sp2 carbons all the carbon atoms about the points labelled by £ vibrate through
breathing mode atom displacements with respect to point £, while exhibiting typical
optical mode displacements with respect to the centres of the other two hexagons in
the honeycomb lattice [117].
The second-order G 0-band in the Raman spectra of various sp2 carbon materials
is generally much more intense than the disorder-induced D-band. This is due to the
fact that the second-order G 0-band is symmetry-allowed by momentum conservation
requirements, whereas the disorder-induced D-band only appears when there is a
breakdown in the in-plane translational symmetry [117]. For a given sp2 sample, the
intensity of the D-band for sp2 carbons increases smoothly as Elaser decreases due to
structural defects or to ® nite size e€ ects [145, 149]. In contrast, the G 0-band is
symmetry-allowed and is an intrinsic feature of all sp2 carbons. Therefore, a Raman
investigation of the second-order G 0-band by varying Elaser for the incident photon
provides an experimental way to probe the particular optical phonon branch
represented by the heavy curve in ® gure 45.
Raman spectra showing the D-band and the G 0-band in SWNTs are shown in
® gure 47 at laser excitation 632.8 nm (1.96 eV) and 514.5 nm (2.41 eV). These spectra
show features for SWNTs with almost the same peak frequencies as the correspond-
ing D-band and G 0-band in sp2 carbons, and these features in the SWNT spectra
show a qualitatively similar dependence of their peak frequencies on Elaser as is found
in other sp2 carbons. The measurements are made on a SWNT sample with a broad
nanotube diameter distribution, as determined by transmission electron microscopy
(TEM) measurements and shown in the lower right inset to ® gure 47. The most
778 M. S. Dresselhaus and P. C. Eklund
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

Figure 45. Electronic energy bands E …k† (top) and phonon dispersion curves !…q† of 2D
graphite (bottom) [2, 41]. Both the phonon branch that is strongly coupled to
electronic bands in the optical excitation, and the electronic bands near the Fermi
level (E ˆ 0) that have a dispersion relation that is linear in k are indicated by heavy
lines. The initial slope for the low frequency TA phonon branch (which is initially the
same along GM and along GK) is also indicated by heavy lines. The strong coupling
between the electrons of wave vector D k, measured from the K-point in the Brillouin
zone, to phonons of wave vector D q ˆ D k is responsible for the frequency
dependence of the D-band and the G 0-band features in the Raman spectra of sp2
carbons and carbon nanotubes [117].

prominent feature in the second-order spectra in ® gure 47 is the G 0-band feature in


the 2600 ± 2700 cm ¡1 range, and we see that the peak frequency of the G 0-band feature
is strongly dependent on Elaser with !G 0 band ˆ 2633 cm ¡1 and 2682 cm ¡1 at 1.96 eV
and 2.41 eV, respectively. The spectrum in ® gure 47 shows a well-de® ned feature at
1323 cm ¡1 for the Elaser ˆ 1:96 eV trace and at 1345 cm ¡1 for the Elaser ˆ 2:41 eV
trace, which can be identi® ed with the so-called `D-band’, as discussed above.
Figure 47 furthermore shows higher-order features for the D-band and the G 0-band
over a very broad range of phonon frequencies [80]. Whereas the frequencies of the
D-band (!D band ) and the G 0-band (!G 0 band ) depend strongly on the laser energy, the
intensities of the D-band and G 0-band features for a given nanotube sample exhibit
Phonons in carbon nanotubes 779
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

Figure 46. Schematic diagram of the atomic displacements (arrows) in the graphene plane
for the E2g2 mode at the G-point, which can be viewed with respect to the unmarked
centres of the hexagons. For the atomic motions of the six atoms about the hexagon
centres denoted by £, breathing-type displacements are obtained corresponding to
normal modes associated with the K-point in the Brillouin zone [117].

only a weak dependence on Elaser , and the data in ® gure 47 are consistent with
!G 0 band ’ 2!Dband for SWNTs.
Figure 48 compares the second-order spectra in the 2400± 3500 cm ¡1 range for
SWNTs and graphite. Whereas the G 0-band for SWNTs and turbostratic 2D
graphene layers is well described by a Lorentzian lineshape, the G 0-band for graphite
shows a doublet structure (see ® gure 48 and the upper inset to ® gure 47) which
physically is related to the two crystallographically inequivalent graphene layers in
3D graphite [116, 151]. Whereas the wavevector for graphite is essentially a
continuous variable, the wavevector in the circumferential direction for nanotubes
has only a few ( ¹40 for a (10, 10) nanotube) allowed values. Furthermore, in
contrast to the radial breathing mode (section 4.1) and the tangential modes
(section 4.2) for which only a few nanotubes of typical SWNT samples are in
resonance with Elaser and therefore contribute resonantly to a given Raman
spectrum, the mechanism described by ® gure 45 indicates that all nanotubes in the
sample contribute to the D-band and G 0-band spectrum at each Elaser value. In
addition, the zone folding of the K-point phonons in the large 2D Brillouin zone
(BZ) will give rise to Raman-allowed k ˆ 0 phonons in the small 1D BZ of the
nanotubes and these activated phonons will be subject to the same nanotube-
selective resonant Raman scattering process discussed in sections 4.1 and 4.2 for
the radial breathing mode and the tangential modes, respectively.
The strong upshift in the frequency of the G 0-band as Elaser increases is
demonstrated in the Raman spectra shown in ® gure 49 (b) and in the plot of the
peak frequencies of the G 0-band versus Elaser given in ® gure 49 (a). We explain the
strong linear dependence of the G 0-band peak frequency on Elaser in single-wall
carbon nanotubes by the resonance of Elaser with electronic interband transitions
near the K-point in the 2D Brillouin zone as described in ® gure 45 for sp2 carbons.
The argument for the linear dependence of !Dband on Elaser is as follows (see
® gure 45). From the linear k dependence of E …k† near the K-point in the Brillouin
zone, the energy for an interband transition at a point D k from the K-point is
D E ˆ ¬D k ; …30†
780 M. S. Dresselhaus and P. C. Eklund
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

Figure 47. Raman spectra of SWNTs over a very broad phonon frequency range excited
with 514.5 nm (2.41 eV) and 632.8 nm (1.96 eV) excitation. The upper inset illustrates
in more detail the C± C tangential stretching G-band modes and the G 0-band of
SWNTs (solid lines) and a comparison is made to these features in highly oriented
pyrolytic graphite (HOPG) (dotted lines). The lower inset shows the diameter
distribution of isolated SWNTs obtained from HRTEM images [80].

which at resonance with Elaser yields D E ˆ Elaser . Strong electron± phonon coupling
occurs when the phonon and electron wave vectors are equal D q ˆ D k. The D-band
frequency !Dband relative to !K at the K-point is
!Dband ˆ !K ‡ ­ D q; …31†
yielding
Elaser !Dband ¡ !K
ˆ ; …32†
¬ ­
so that
¬
D Elaser ˆ D !Dband : …33†
­
Since the G 0-band is the overtone of the D-band, we can write !G 0 band ’ 2!Dband for
SWNTs. Because of the higher intensity and much better data quality for the G 0-
band relative to the D-band, we discuss the G 0-band ® rst.
From ® gure 49 (a), we ® nd a value for the slope @!G 0 band =@Elaser ˆ 106 cm ¡1/ eV
for SWNTs as compared to @!G 0 band =@Elaser ˆ 101 cm ¡1 =eV for 3D graphite [117,
118]. For all the Raman spectra shown in ® gure 49 (b), the lineshape for the G 0-band
features is ® t by a single Lorentzian component, with a linewidth that has a very
weak dependence on Elaser . The data in ® gure 49 (a) for SWNTs extrapolate to
Phonons in carbon nanotubes 781
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

Figure 48. Second-order Raman spectra of single-wall carbon nanotubes and of 3D graphite
for three di€ erent laser excitation energies [150].

1.83 eV
Raman Intensity (a.u.)

2700 1.96

2680
2.19
Raman shift (cm-1)

2660
2.41
2640

2620 2.71

2600
1.6 2.0 2.4 2.8 2400 2500 2600 2700
Energy (eV)
2400 2500 2600
Raman shift (cm -1 )
2700
(a) (b)

Figure 49. The Raman spectra for the intense G 0-band for ® ve values of Elaser are shown in
(b), while (a) shows a plot of the peak Raman frequency for the G 0-band features
versus laser excitation energy Elaser [114].
782 M. S. Dresselhaus and P. C. Eklund

1.83 eV

( a.u . )
1.92

(a.u.)
In te n s i ty
1.96
a m a n Intensity

2.41
Raman

2.54
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

2.71

0
2360
2360 2400
2400 2440
2440 2480
2480 2520
2520
?1 -1 )
Raman shift (cm
Ramanshift(cm )
Figure 50. The weak non-resonant Raman feature associated with the second harmonic
(overtone) of the K-point phonon in the 2D Brillouin zone for six values of Elaser
taken on a nanotube sample with dt ˆ 1:49 § 0:20 nm [114].

2429 cm ¡1 at Elaser ˆ 0, and this phonon frequency is approximately twice the K-


point phonon frequency in the 2D Brillouin zone of sp2 carbons. This extrapolation
also agrees quite well with the direct measurement of this same phonon frequency
(2440 ± 2445 cm ¡1 ) for SWNTs, as shown in ® gure 50. The low intensity of the K-
point second-order feature is attributed to its non-resonant origin, since the K-point
feature corresponds to the Elaser ˆ 0 limit in ® gure 45. The extrapolation of the
G 0-band peak frequency to Elaser ˆ 0 yields 2429 cm ¡1 which is quite consistent with
the corresponding extrapolation of the D-band peak frequency to 1415 cm ¡1 at
E laser ˆ 0 as discussed below.
A series of spectra showing the D-band in single-wall carbon nanotubes
are presented in ® gure 51 for several values of E laser . The strong dependence
of the D-band peak frequency as a function of E laser is apparent from the
spectra in ® gure 51, and a summary of the collected D-band frequencies as a
function of E laser is shown in ® gure 52 from which we obtain a value of
@ !Dband =@E laser ˆ 51:2 cm ¡1 = eV and an intercept at Elaser ˆ 0 of 1215 cm ¡1 . Meas-
urements on a number of SWNT samples give an average value of
@ !Dband =@E laser ˆ 52 § 1 cm ¡1 / eV which is to be compared with an average value
for graphite of 48 § 3 cm ¡1 / eV [117]. As mentioned above, the extrapolated value of
1215 cm ¡1 for the D-band frequency at the K-point is in good agreement with half
the extrapolated value of !G 0 band ˆ 2429 cm ¡1 to E laser ˆ 0.
We now summarize a number of characteristics for the D-band and G 0-band for
carbon nanotubes. The ® rst relates to polarization e€ ects, which have been reported
only for MWNTs thus far, and only for the D-band (not for the G 0-band). In this
connection, ® gure 43 shows that the D-band spectra for multi-wall nanotubes
( ¹25 nm diameter) have a strong polarization dependence that is di€ erent from that
of the Lorentzian components of the tangential G-band, and also very di€ erent from
the weak polarization dependence observed for disordered carbons [136] (see
Phonons in carbon nanotubes 783

1.49 eV

1.58
Raman Intensity (a.u.)
1.83

1.92

1.96
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

2.19

2.41

2.54

2.71
1190 1240 1290 1340 1390
Raman Shift (cm-1)
Figure 51. Raman spectra in the range 1190± 1390 cm¡1 for SWNTs showing the upshift of
the D-band peak frequency with increasing Elaser [101, 102].

Figure 52. Frequency of the `D-band’ for single-wall carbon nanotubes as a function of
laser excitation energy. The line is a least squares ® t to the data points for one SWNT
sample. The ® t to the data points yields a slope of 51.2 cm¡1/ eV and an intercept of
1215 cm ¡1 at Elaser ˆ 0 [150].
784 M. S. Dresselhaus and P. C. Eklund

section 4.7). Further study of the polarization e€ ects associated with the D-band in
carbon nanotubes both experimentally and theoretically is needed.
Di€ erences between the Stokes and anti-Stokes spectra are also found in ® gure 33
at Elaser ˆ 1:58 eV and 1.96 eV for the D-band and G 0-band with regard to their peak
frequencies and the relative intensities. No second-order features are observed in the
anti-Stokes spectra for Elaser ˆ 1:96 eV, which is consistent with a Boltzmann factor
reduction in relative intensity. However, the observation of Raman intensity in the
G 0-band in the anti-Stokes spectrum at 1.58 eV cannot simply be explained by a
Boltzmann factor exp…¡Eph =k B T † e€ ect, but rather may be due to an enhanced
electron ± phonon coupling mechanism for metallic nanotubes. Additional enhance-
ment of the anti-Stokes process for metallic SWNTs relative to semiconducting
SWNTs is observed experimentally, implying a stronger electron± phonon coupling
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

for metallic nanotubes.


The di€ erent frequency shifts of the G 0-band features in ® gure 33 can be
explained by the di€ ering energies of the scattered photons which are involved with
the 2D resonance enhancement e€ ects associated with the K-point in the Brillouin
zone [117]. For example, an expected di€ erence of 34 cm ¡1 between the anti-Stokes
and Stokes G 0-band frequencies is predicted from the measured …@!G 0 band =@Elaser †
slope of 106 cm ¡1/ eV for the G 0-band dispersion [118], which is in good agreement
with the observed frequency di€ erence of 36 cm ¡1 shown in ® gure 33.
Next we review present knowledge about the D-band and G 0-band features as
they are revealed in the surface-enhanced Raman spectra SERS discussed in
section 4.6. In ® gure 37, we see that the D-band feature (1307± 1315 cm ¡1 ) in the
® rst-order Stokes spectrum at 1.96 eV which lies within the metallic window for this
sample is upshifted by 4± 8 cm ¡1 in the surface enhanced resonant Raman spectrum
(SERRS) relative to the normal resonant Raman spectrum (RRS), while the
corresponding second-order G 0-band is upshifted by 24 cm ¡1 upon interaction with
the metal substrate. It is interesting to note that the upshift of the G 0-band associated
with semiconducting nanotubes (observed at Elaser ˆ 2:41 eV) is only 13 cm ¡1 ,
showing that the SERRS-induced shift in the G 0-band is much larger for Elaser
values where metallic nanotubes contribute strongly to the Raman spectra than for
Elaser in the semiconducting regime. Furthermore at Elaser ˆ 1:96 eV (within the
metallic window), the intensity of the G 0-band relative to that of the tangential mode,
denoted simply by …IG 0 =Itang †, has a value of 0.4 for the normal resonant Raman
spectra (RRS) as compared to a signi® cantly larger value of 0.6 for the SERRS
spectra. In contrast, for Elaser ˆ 2:41 eV, where only the semiconducting nanotubes
are in resonance with Elaser , the ratio …IG 0 =Itang † has the same smaller value of 0.2 for
both the RRS and SERRS spectra.
An even larger upshift of 36 cm ¡1 is observed in the frequency of the second-
order G 0-band for Elaser ˆ 1:49 eV for the nanotubes adsorbed to a Ag nano-
structural substrate (from 2600 cm ¡1 for the normal RRS spectrum to 2636 cm ¡1
for the SERRS spectrum) at the anti-Stokes side of the Raman spectra. This larger
upshift of the G 0-band is attributed to the higher charge transfer that is produced by
the interaction of the metallic nanotubes with the nanoscale Ag particles in the
colloidal Ag substrate relative to the ® lm Ag substrate where the upshift was only
¹24 cm ¡1 . Since the anti-Stokes spectrum at Elaser ˆ 1:49 eV is dominated by
resonance with metallic nanotubes, we can explain the large intensity of the G 0-
band relative to the tangential band, which seems to be a factor of ¹2 larger for the
RRS spectrum relative to the SERRS spectrum. This observation is explained by the
Phonons in carbon nanotubes 785

lower frequency of the G 0-band in the RRS spectrum relative to the SERRS
spectrum. Heating e€ ects are less important for SERRS than for RRS because of
the good thermal contact between the metallic nanotubes and the metal substrates in
SERRS experiments. It would be expected that a phonon with a high vibrational
frequency ( ¹2600 cm ¡1 ) would be very weak in the room temperature anti-Stokes
spectrum because of the highly unfavourable Boltzmann factor. The fact that the G 0 -
band can be observed in the SERRS anti-Stokes spectrum at 1.49 eV is attributed to
an optical pumping e€ ect which is operative for metallic nanotubes (see section 4.6)
[130].
The large upshifts in the G 0-band frequencies in the SERRS spectra for
all nanotubes relative to the corresponding RRS spectra, and the relative increase
in the intensities of the G 0 Raman band in the SERRS spectra for the metallic
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

nanotubes indicates that for all k values the electronic structure E …k† of the SWNTs
is signi® cantly perturbed by the metal substrates to which the SWNTs are adsorbed.
Following this line of reasoning, it can be argued that the interaction between
the nanotubes and the metal substrate also perturbs the electronic structure near
the K-point of the 2D Brillouin zone, so that the D-band and the G 0-band
vibrational spectra, which are resonantly excited by a given laser energy Elaser ,
now correspond to a slightly larger D k relative to the K-point in the 2D Brillouin
zone. But since the D-band (and the G 0-band) phonon that is resonantly enhanced
has the same wave vector as that of the resonant electronic transition D q ˆ D k [117],
the upshift in the SERRS spectrum relative to the RRS spectrum is a measure of the
strength of the electronic interaction between the nanotube and the metal substrate.
Because of the much larger experimentally observed upshift of the G 0-band
frequency for the metallic nanotubes, we conclude that the interaction of the metallic
nanotubes with the metal substrate is much stronger than for the semiconducting
nanotubes. This conclusion is also consistent with the increased intensity of certain
components associated with the tangential band of metallic nanotubes in the SERRS
spectra relative to the intensity of the same components in the RRS spectra. These
arguments also support the conclusion that electron ± phonon interaction e€ ects are
stronger for the metallic nanotubes than for the semiconducting nanotubes,
consistent with the large carrier density in metallic single-wall nanotubes.

4.9. Overtone and combination modes


In molecular (0D) Raman spectroscopy, the combination and overtone modes
are generally more important for the interpretation of the phonon data than is the
case for 3D solids. This is related to the fact that in 3D solids the inclusion of two or
more phonon modes in a scattering process relaxes the wave vector selection rule,
and thereby gives relatively broad structures in the Raman spectra. However, in 1D
materials the k-space integration is only in one direction and thus the higher-order
Raman features remain relatively sharp.
We discuss below the various features in the second-order spectrum associated
with the harmonics (overtones) and combination modes of the two dominant
features of the ® rst-order spectrum, which are the low frequency radial breathing
mode !RBM and the high frequency tangential band !tang . We also discuss harmonics
and combination modes for the D-band and the G 0-band discussed in section 4.8.
In the case of fullerenes, a more complete analysis of the combination and
overtone modes has been made [152, 153] and a detailed tentative identi® cation of all
786 M. S. Dresselhaus and P. C. Eklund
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

Figure 53. The Raman spectra for the radial breathing mode band (! RBM ) and its second
harmonic at 2!RBM for two laser excitation energies 1.58 eV (785 nm) and 2.54 eV
(488 nm) for an SWNT sample with dt ˆ 1:49 § 0:20 nm [114].

the phonons has been given. At the present time the available data for carbon
nanotubes does not support such a de® nitive identi® cation.

4.9.1. Overtones
In ® gure 53, we see (on a scale amjli® cation of 20£) spectral features identi® ed
with the second harmonic (overtone) of the radial breathing mode 2!RBM (see
section 4.1) at two di€ erent laser excitation energies. The spectra at Elaser ˆ 1:58 eV
(785 nm) show features in the ® rst-order spectrum at !RBM ˆ 150 cm ¡1 and 162 cm ¡1
as well as second-order lines at approximately 2!RBM ˆ 301 cm ¡1 and 330 cm ¡1 . A
change in Elaser excites di€ erent nanotubes, so that for Elaser ˆ 2:54 eV (488 nm), the
® rst-order Raman spectrum shows a strong feature at !RBM ˆ 159 cm ¡1 and a much
weaker second harmonic at 2!RBM ˆ 320 cm ¡1 (see ® gure 53). Similar trends are
observed at other values of Elaser [114]. The intensity of the overtone of the radial
breathing mode reaches a maximum for Elaser values where contributions from
metallic nanotubes dominate the spectra for the tangential band (see section 4.2),
suggesting that the 2!RBM feature also is stronger for metallic nanotubes relative to
semiconducting nanotubes.
Phonons in carbon nanotubes 787
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

Figure 54. Raman spectra for a SWNT sample with dt ˆ 1:49 § 0:20 nm for the second
harmonic of the tangential G-band mode, at ® ve laser excitation energies. The ® rst-
order spectra for the tangential G-band mode are displayed in ® gure 27 [114].

In contrast, the second harmonic of the tangential band 2!tang which occurs in
the range 3100± 3250 cm ¡1 (see ® gure 54) has very di€ erent characteristics from the
second harmonic of the radial breathing mode 2!RBM shown in ® gure 53. We see in
® gure 54 the evolution of the second harmonic of the tangential band at 2!tang for
® ve laser energies in the range 1.96 ± 2.71 eV. The central frequency and linewidth of
this second-order band exhibits a relatively weak dependence on Elaser for
2:19 < Elaser µ 2:71 eV, where the semiconducting nanotubes dominate the ® rst-
order spectra for this sample. However, for Elaser ˆ 1:96 eV, where the dominant
contribution to the ® rst-order spectrum comes from metallic nanotubes, the second-
order spectrum is downshifted and is much broader than for the higher Elaser values,
consistent with the behaviour of the ® rst-order tangential band discussed in
section 4.2.
The feature in the second-order Raman spectrum for 3D graphite near 3240 cm ¡1
(see ® gure 48) is strongly a€ ected by the shape of the uppermost optical branch of
the !…q† phonon dispersion curves for graphite [116, 154], which exhibits a peak in
the phonon density of states near 1620 cm ¡1 , associated with intermediate wave
vector (non-zone centre) phonons. This peak in the phonon density of states is
responsible for the feature in the second-order spectrum of graphite near 3240 cm ¡1 ,
and the frequency is upshifted by 76 cm ¡1 from twice the zone-centre phonon mode
in graphite at 2 £ 1582 cm ¡1 ˆ 3164 cm ¡1 . Calculations of the phonon dispersion
788 M. S. Dresselhaus and P. C. Eklund

1.96 eV 1.96 eV

Raman Intensity (a.u.)

1540 3153
1581
3122
3178
1590 3203
1515
3082
1601

1400 1500 1600 1700 3000 3100 3200 3300


Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

2.71 eV 2.71 eV
3181
1592
Raman Intensity (a.u.)

1567 3216
1599
3153

1400 1500 1600 1700 3000 3100 3200 3300


Raman shift (cm -1 ) Raman shift (cm -1 )
Figure 55. A line shape analysis of the spectral features in the ® rst-order spectra (left) and
in the second-order spectra (right) for the tangential bands taken for Elaser ˆ 1:96 eV
(632.8 nm) (top) and 2.71 eV (457.9 nm) (bottom) for a SWNT sample with
dt ˆ 1:49 § 0:20 nm [114].

curves in SWNTs [2, 60] show no corresponding peak in the density of states away
from the Brillouin zone centre, and therefore we only expect a Raman feature close
to twice that for the ® rst-order tangential band, consistent with experiments for
SWNTs [114].
A Lorentzian lineshape analysis of the second-order spectrum associated with the
second harmonic of the tangential band shows qualitatively di€ erent behaviour for
semiconducting nanotubes in comparison to metallic nanotubes (see ® gure 55). For
Elaser ˆ 2:71 eV where semiconducting nanotubes mainly contribute resonantly to
the ® rst and second order spectra [94], the dominant feature in the second-order
spectrum is at 3181 cm ¡1 , which is close to twice the frequency of the dominant mode
in the ® rst-order spectrum 2…1592† ˆ 3184 cm ¡1 , and the second-order feature is only
slightly broader than twice the FWHM linewidth of the ® rst-order feature. The
frequencies of the two weaker features in the second-order spectrum at 2.71 eV
correspond approximatel y to twice the frequencies of the ® rst-order features. As
Elaser decreases from 2.71 eV, the peak frequency of the entire second-order band (see
® gure 54) downshifts, especially for the lowest value of Elaser , because new tangential
peaks associated with metallic nanotubes are resonantly enhanced. In contrast, the
second-order spectrum at Elaser ˆ 1:96 eV in ® gure 54 shows a broad, asymmetric
band with more scattering intensity at low phonon frequencies. Analysis of the
lineshape of this Raman band in ® gure 55 shows a feature at 3082 cm ¡1 , which is
Phonons in carbon nanotubes 789
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

Figure 56. Spectral features tentatively associated with combination bands for !tang ‡ !RBM
and !tang ‡ 2!RBM in the second-order Raman spectra of carbon nanotubes at
Elaser ˆ 1:58 eV (785 nm) and 1.96 eV (632.8 nm) [114].

close to 2 £ 1540 cm ¡1 , thereby providing support for the interpretation that metallic
nanotubes are contributing to this second-order Raman band at Elaser ˆ 1:96 eV.
The second-order tangential band at Elaser ˆ 1:96 eV in ® gure 55 is so broad and
featureless, that it is not meaningful to attach signi® cance to the Lorentzian
components obtained from a Lorentzian lineshape analysis. The spectra in
® gure 47 show that the overtones for the tangential band are strong enough to
observe well-resolved spectral features to third order (3G) (see section 4.9.3) [80].
The largest linewidths are observed when both the incident and scattered photons
are in resonance with metallic nanotubes for both the ® rst and second order
scattering processes. Intermediate linewidths are observed when both semiconduct-
ing and metallic nanotubes contribute to the scattering process.

4.9.2. Combination modes


Resonant Raman e€ ects associated with the 1D electron density of states
singularities also give rise to resonant Raman e€ ects in the combination modes.
One example of a combination mode occurs at the sum frequency between a
tangential and one or two radial breathing mode phonons !tang ‡ !RBM and
!tang ‡ 2!RBM , as shown in the spectra in ® gure 56 taken at Elaser ˆ 1:58 eV and
1.96 eV. Taking the radial breathing mode at !RBM ˆ 150 cm ¡1 for Elaser ˆ 1:58 eV
(see ® gure 53), and the most intense line at 1591 cm ¡1 for the tangential band !tang
yields a sum of !tang ‡ !RBM ˆ 1741 cm ¡1 , in good agreement with the combination
mode at 1742 cm ¡1 [114].
The other higher frequency feature in ® gure 56 is tentatively assigned to a second
combination band associated with !tang ‡ 2!RBM . Taking !tang ˆ 1540 cm ¡1 and
!RBM ˆ 160 cm ¡1 at Elaser ˆ 1:58 eV (see ® gure 56 and [114]), we obtain
(!tang ‡ 2!RBM† ˆ 1860 cm ¡1 for the metallic regime, in good agreement with the
experimental peak at 1871 cm ¡1 . For the spectrum at Elaser ˆ 1:96 eV, we take
!tang ˆ 1590 cm ¡1 appropriate for semiconducting nanotubes and !RBM ˆ 165 cm ¡1 ,
then we get !tang ‡ 2!RBM ˆ 1920 cm ¡1 , which accounts for the observed frequency
in ® gure 56 at 1925 cm ¡1 [114].
790 M. S. Dresselhaus and P. C. Eklund

Table 4. Experimental (expt) peak positions ! (cm ¡1 ) and the theoretical assignments (th) of
selected ® rst-order and higher-order Raman features of SWNTs excited by 488 nm
(2.54 eV) excitation [80, 101].

!expt !th …@!=@E †expt …@!=@E †th


¡1 ¡1 ¡1
(cm ) Assignment (cm ) (cm /eV) (cm ¡1 =eV)

1352 D 1348 50 53
1582 G 1582 0 0
2696 G0 2696 109 106
2946 D+ G 2930 50 53
3184 2G 3164 5 0
4273 G0‡ G 4278 129 106
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

4800 3G 4746 6 0
5330 2G 0 5392 260 212
5848 G 0 ‡ 2G 5860 133 106
6885 2G 0 ‡ G 6974 194 212

Figure 37 shows that the SERS e€ ect upshifts the !tang ‡ !RBM combination
mode by ¹10 cm ¡1 , but upshifts the !tang ‡ 2!RBM combination mode by ¹31 cm ¡1 .
The observed combination mode frequencies suggest that the (!tang ‡ !RBM )
combination mode might be more enhanced by the SERS process for metallic
nanotubes. The larger upshift for the (!tang ‡ 2!RBM ) combination mode is
correlated with a larger variation of this mode frequency with Elaser in the normal
resonant Raman spectra [114].

4.9.3. Overtones and combination modes for the D-band and G 0-band
Since the D-band and G 0-band involve a di€ erent resonance process (see
section 4.8) than the radial breathing mode or the tangential band (also called the
G-band), the harmonics and overtones of the D-band and G 0-band show di€ erent
properties, as discussed below. The overtones and combination modes for the D-
band and G 0-band at Elaser ˆ 1:92 eV (632.8 nm) and 2.41 eV (514.5 nm) are shown in
® gure 47. A listing of the frequencies and dispersion of the D-band, the G 0-band, the
tangential G-band and of their overtones and combination modes is given in table 4
for Elaser ˆ 2:54 eV (488 nm) excitation [80]. The table includes their assignments,
which are made on the basis of their frequencies and their dispersion …@! =@Elaser †.
Their observed dispersion …@! =@Elaser †expt is determined from the spectral informa-
tion provided for Elaser ˆ 1:92 eV, 2.41 eV and 2.54 eV [80] and the characteristic
dispersion of the D-band and G 0-band discussed in section 4.8. The theoretical
estimates, !th , given in table 47 are obtained from the assignments listed for each
spectral feature and the values of the G 0-band and G-band (tangential band) mode
frequencies. Also listed in the table is the predicted dispersion …@!=@Elaser †th for each
entry based on …@!G 0 band =@Elaser † ˆ 106 cm ¡1 / eV and …@!G band =@Elaser † ˆ 0. Good
agreement is obtained between the observed mode frequencies and their dispersions
and the values predicted from the mode assignments and from the well-established
frequencies and dispersion for the D-band, G-band and G 0-band. The large
dispersion of the D-band and the G 0-band and the absence of dispersion for the
G-band are key factors in the identi® cation of the overtones and combination modes
in table 4. The strong electron ± phonon coupling in the SWNTs makes it possible to
Phonons in carbon nanotubes 791
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

Figure 57. Raman spectra of pristine, moderately I-doped and saturation I-doped single-
wall nanotube samples (T ˆ 300 K, 514.5 nm laser excitation) where molten I2 is the
dopant. The insert shows the photoluminescence spectrum due to the intercalated
polyiodide chains in the moderately doped sample where sharp Raman lines are
superimposed on the broad PL spectrum [155].

observe overtones beyond the fourth harmonic. It should be noted that the
experimental dispersion of the G 0-band appears to be greater than 106 cm ¡1 / eV
for the higher-order modes.

4.10. Raman studies of doped carbon nanotubes


Raman spectra have also been taken on doped single-wall carbon nanotube ropes
[39, 40, 155 ± 159]. These spectra show e€ ects similar to the e€ ect of alkali metal and
halogen intercalation into graphite [160], which exhibits characteristic upshifts in the
E2g2 mode frequency associated with the donation of electrons from graphite to the
halogens in the case of acceptors, and characteristic downshifts in mode frequencies
associated with donor charge transfer to graphite in the case of alkali metal
intercalation (see ® gure 57 for the upshift in the tangential G-band frequency for
iodine intercalated single-wall carbon nanotubes) [155]. To date, Raman scattering
studies have been carried out mainly on as-prepared (and unpuri® ed) material taken
directly from the synthesis chamber and exposed to the following reactants: (donors)
Li [158], K and Rb [161], and Cs [39, 40]; (acceptors) sulphuric acid (H2 SO4 ) [159],
Br2 [39, 40, 161], iodine (I2 ) vapour [161], and molten iodine [155]. It is of interest to
mention that iodine does not form a graphite intercalation compound (GIC) but
does e€ ectively dope SWNTs. Many authors have assumed that for charge transfer
processes in single-wall carbon nanotubes the dopant resides as ions (and also
possibly as neutral atoms) in the interstitial channels between the nanotubes in the
triangular nanotube lattice. While this presumption has yet to be con® rmed by X-ray
di€ raction, or other structural probes, the presumption implies that isolated SWNTs
which have no interstices may behave di€ erently from SWNT bundles which do have
interstices regarding the stability of the adsorbed dopants in charge transfer
processes.
792 M. S. Dresselhaus and P. C. Eklund
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

Figure 58 (a). Resonance Raman spectra at many values of Elaser for a fully doped bromine
SWNT sample prepared using a NiY catalyst (1:24 < dt < 1:58 nm). An additional
peak around 240 cm ¡1 can be seen for laser excitation energies Elaser ¶ 1:96 eV. After
taking these spectra, the sample chamber was evacuated at room temperature. (b)
(left scale) The dependence on photon energy of the optical density (absorption
spectra) for a pristine (undoped) SWNT sample, and (right scale) the Raman
intensity ratio of the features at ¹240 cm¡1 appearing only in the doped sample to
the RBM feature at ¹180 cm ¡1 . The additional Raman peak near 240 cm ¡1 only
appears when the metallic window is satis® ed by the laser excitation energy [39, 40].

Measurements of the RBM spectra for Br2 doped SWNTs with diameters in the
range 1:24 < dt < 1:58 nm show that in addition to the RBM features near 190 cm ¡1 ,
a new feature appears near 240 cm ¡1 for Elaser > 1:8 eV, as shown in ® gure 58 (a) [39,
40]. The Elaser dependence of the ratio of the intensity of the 240 cm ¡1 feature to that
of the 180 cm ¡1 is shown in more detail using the right hand scale of ® gure 58 (b),
and these data are to be compared with the optical absorption spectra of the same
SWNT sample, shown on the left scale of the same ® gure, before Br2 doping. Figure
58(b) shows absorption peaks for the E11 S d , ES d M d
… t† 22 … t† and E11 … t† interband
transitions between van Hove singularities, as discussed in section 2.2 [29]. Doping
with bromine to the saturation limit completely suppresses the three optical
absorption features [37], as EF is lowered by about 1 eV through the charge transfer
process, giving rise to a suppression of interband transitions by 2D EF ¹ 2 eV, as
shown in ® gure 58 (b). The authors [39, 40] argue that upon evacuation of the sample
chamber, the bromine dopant is completely removed from the isolated SWNTs
present in the sample, but not from the bundles which retain some of the adsorbed
bromine even after evacuation of the chamber at 300 K. The authors [39, 40] relate
these arguments to the observation of the same tangential mode spectrum before
bromine addition and after evacuation of the sample chamber, as shown in ® gure 59.
Because of the lowering of the Fermi level for the Br-doped nanotubes, no optical
M
absorption and therefore no resonant Raman scattering occurs within the E11 …dt†
Phonons in carbon nanotubes 793
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

Figure 59. The Raman spectra for the tangential band for a pristine sample (before doping
the SWNT sample with bromine) (upper ® gure) and for the same SWNT sample
which was evacuated after Br2 doping to saturation at room temperature (lower
® gure). The spectral feature at 1540 cm ¡1 observed in the sample before Br2 doping
(upper ® gure) is missing in the spectrum for the evacuated sample (lower ® gure). The
laser excitation energy Elaser ˆ 1:78 eV is within the metallic window for this SWNT
sample before doping, but the downshift in Fermi level by ¹1 eV due to Br 2 doping
[37] prevents optical absorption at Elaser ˆ 1:78 eV associated with the E11 M
…dt † inter-
subband transition for the Br2 -doped sample (see text) [39, 40].

metallic window for the Br-doped SWNTs. The authors therefore argue [39, 40] that
only the isolated SWNTs, from which the adsorbed Br is readily removed by
evacuation of the sample chamber, can be resonantly excited by the Raman process
at Elaser ˆ 1:78 eV for the evacuated sample. In this connection ® gure 59 shows that
the Raman peak at 1540 cm ¡1 identi® ed with metallic nanotubes is not present in the
Raman spectrum at 1.78 eV for the evacuated sample. Their interpretation of this
observation [39, 40] is that the Br-doped nanotubes in the nanotube bundles do not
M
contribute to the 1540 cm ¡1 feature because light in the E11 …dt† metallic window is
not absorbed when bromine is adsorbed to the SWNT bundles, and that the isolated
SWNTs also do not contribute to the 1540 cm ¡1 feature, suggesting that the
excitation of the 1540 cm ¡1 structure requires the nanotubes to be within a SWNT
bundle [39, 40]. This conclusion needs to be con® rmed by other independent
experiments and understood in terms of a physical mechanism for the observation
of the 1540 cm ¡1 feature associated with metallic nanotubes in the Raman spectra. It
can also be argued that only the metallic nanotubes which have mobile charge retain
794 M. S. Dresselhaus and P. C. Eklund
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

Figure 60. Z-contrast image of iodine atoms inside a single-wall carbon nanotube (upper
left). The inset in this image is a maximum-entropy processed image to reduce the
noise, and the schematic diagram shows the proposed structure for the iodine atoms
as a double helix structure. The schematics (right and below) show how the proposed
iodine chains are located within a (10,10) armchair nanotube (from Fan et al. [38]).

adsorbed bromine after evacuation of the sample chamber, so that only semicon-
ducting nanotubes can contribute resonantly to the tangential band Raman spectra
at Elaser ˆ 1:78 eV for the evacuated sample chamber, since the metallic nanotubes in
this case do not absorb light at 1.78 eV [37].
Recently, the high resolution Z-contrast STEM technique, which is a sub-
nanoscale probe capable of imaging individual atoms, was used to produce images
of individual SWNTs that had been doped through contact with molten iodine [38].
The images were taken on nanotubes which protruded beyond the ends of the iodine
doped bundles, showing chains of iodine atoms located inside the nanotube, but not
in the interstitial channels. Supporting theoretical calculations con® rmed that charge
transfer actually occurs between iodine and the nanotube wall in the case of linear I5¡
and I3¡ intercalated molecules. From these theoretical results, a strong interaction
between the iodine molecule and the nanotube substrate was inferred. Several such
Z-contrast images on di€ erent SWNTs were obtained and the images were
interpreted to indicate that the iodine enters within the nanotube core region and
that charged polyiodide chains are intertwined inside an individual nanotube in the
form of a double helix. Figure 60 [38] shows a high resolution Z-contrast image of a
SWNT containing strands of iodine. The image shows a specially processed image to
reduce the noise. The period of the helix is ¹5 nm, and the maximum separation of
the strands is ¹0.65 nm, consistent with the inside diameter of a (10, 10) armchair
nanotube. Other iodine helicities were also observed in other images. It should be
noted that no post-synthesis treatment to open the nanotube ends was carried out, so
it is not known if some nanotube ends were naturally open or whether the iodine
itself opened the nanotube ends to receive the dopant, or if the dopant entered
through defects in the nanotube wall. Because most of the nanotube ends were
presumed to be closed, at least at the start of the iodine doping reaction, the bulk of
the iodine dopant was viewed as residing in the interstitial channels [38]. Calculations
Phonons in carbon nanotubes 795

of the Raman mode frequencies for a charged helical chain of iodine atoms have yet
to be carried out.
Charge transfer reactions of as-prepared bundles of single-wall carbon nanotubes
with molten iodine (see ® gure 57) produced a much more dramatic e€ ect on the
Raman spectrum than was observed when the reaction was carried out in iodine
vapour [155]. The Raman spectra clearly showed the e€ ects of a charge transfer
reaction between the iodine and the single-wall nanotube bundles to be reversible
and uniform, based on the upshift of the tangential band frequencies with increasing
exposure of the nanotube bundles to iodine, and the downshift of !tang when the
iodine was de-intercalated. This reaction produces an air stable compound, which by
weight uptake measurements is found to have an approximate stoichiometry C12 I.
The Raman spectra in ® gure 57 for undoped (pristine), moderately I-doped and
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

saturation I-doped single-wall nanotube bundles [155, 157] show a low frequency
region which is rich in structure, identi® ed with the presence of intercalated charged
polyiodide chains (I3¡ and I5¡ ). No Raman evidence for neutral I2 (215 cm ¡1 ) was
found in the nanotube samples. Saturation doping was found to convert the I3¡
observed in the moderately doped material into I5¡ . Analysis of these data suggested
that intercalation downshif ts the radial breathing mode band from 186 to 175 cm ¡1 ,
while the tangential band is upshif ted by 8 cm ¡1 from 1593 to 1601 cm ¡1 . The radial
mode downshift may be due to a coupling of the nanotube wall to the heavy iodine
chains, while the small (8 cm ¡1 ) tangential mode upshift (compared to the 24 cm ¡1
upshift of the tangential mode upon Br2 intercalation) is attributed to the iodine
chains being only singly ionized. If all the iodine were in the form of I5¡ in the C12 I
compound, we might write the formula as C‡ 60 (I5 ) , which is equivalent to one hole
¡

per 60 C-atoms, and we therefore ® nd an upshift of 8 cm ¡1 per additional hole per 60


C-atoms in the single-wall nanotubes (or 480 cm ¡1 per hole per C-atom). This shift
can be compared to that of 6 cm ¡1 per added electron in M-doped C60 (or 360 cm ¡1
per electron per C-atom) [1].
Both the radial and tangential Raman-active nanotube modes have been
observed to upshift or downshift signi® cantly with doping. Since the sign of the
mode frequency shift in nanotubes is consistent with earlier studies of intercalation
in GICs and C60 , the shift has therefore been interpreted [161] in terms of C± C bond
expansion or contraction. For example, Br2 intercalation makes the tangential
modes upshif t by 24 cm ¡1 . In the case of the donor dopants (K, Rb), the tangential
vibrational bands downshif t, and the spectra are remarkably similar, suggesting that
under the experimental conditions used, both reactions proceed to the same endpoint
stoichiometry. The highest frequency tangential modes in the K and Rb intercalated
nanotube Raman spectra were ® t to a Breit± Wigner± Fano lineshape, similar to that
found for the ® rst stage MC 8 GICs (M ˆ K, Rb, Cs), although the coupling constant
…1 =qBWF† is a factor of three lower than the …1 =qBWF† observed in GICs [162],
thereby leading to a more symmetric (Lorentzian-like) lineshape for the doped
nanotubes. It should be noted that the spectra in ® gure 57 were taken for Elaser
outside the metallic window, while the spectra in ® gure 59 were taken for Elaser within
the metallic window for their respective SWNT samples. While there is no
inconsistency between the reported Raman results in the two cases, further work
as a function of Elaser is needed to gain a detailed picture of the mechanism
responsible for the Raman feature at 1540 cm ¡1 identi® ed with metallic nanotubes
with dt ¹ 1:4 nm. A variety of other Raman features in the region 900± 1400 cm ¡1
were also observed for the donor intercalated nanotubes, but no speci® c assignments
796 M. S. Dresselhaus and P. C. Eklund

of the mode features were made, since the corresponding modes in the undoped
SWNT sample were not observed.
In situ Raman scattering studies, performed during the electrochemical anodic
oxidation of single-wall carbon nanotube bundles in sulphuric acid [159], are of
special interest, and can be directly compared with similar studies on H 2 SO4 GICs
[163]. In the case of the nanotubes, a rapid spontaneous shift of ¹15 cm ¡1 in the
tangential Raman modes was observed under open circuit conditions which was not
observed in the GIC system. In the H2 SO4 single-wall nanotube study, a direct
measure was obtained for the charge transfer e€ ect on the tangential mode
frequency, i.e. 320 cm ¡1 per hole per C-atom, in reasonable agreement with values
discussed above for M-doped C60 [1].
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

5. Thermal properties
The thermal properties of carbon nanotubes, including their speci® c heat,
thermal conductivity and thermopower, strongly depend on the phonon dispersion
relations and the phonon density of states. Although our present knowledge of these
thermal properties is very limited [54], we expect these properties to be unique to
carbon nanotubes and to display features characteristic of the low dimensionality of
the carbon nanotubes.

5.1. Speci® c heat


Since almost all experiments on thermal properties are done on collections of
carbon nanotubes, which contain both metallic and semiconducting nanotubes, the
heat capacity (and the speci® c heat which is the heat capacity per gram of sample), in
principle, consists of contributions from both phonons and electrons. For 3D
crystalline graphite and 2D graphene layers, from which carbon nanotubes are
derived, the dominant contribution to the heat capacity comes from the phonons,
while the electronic contribution is so small that it can be essentially neglected, even
at low temperatures (i.e. below 4 K). Thus, we expect the heat capacity for SWNTs
to be dominated by phonons. The phonon contribution to the heat capacity can in
general be written as:
- 2
h!

Cph ˆ
… † »…!† exp …h!- =k T † d! ;
… 1 kB
kB T
B
…34†
0
- =k B T † ¡ 1Š2
‰ exp …h!
where »…!† is the phonon density of states determined by the phonon dispersion
relations !…q† for the various phonon modes, which have been calculated for SWNTs
(see section 3.1). In the low T regime, Cph for isolated SWNTs is expected to be
dominated by the four acoustic phonon modes discussed in section 3.1, including the
non-degenerate LA mode, the doubly-degenerat e TA modes, and the non-degenerate
twist mode (TW) of the nanotube. All of these modes, which are populated at low T ,
exhibit a linear ! ˆ vk dependence, where v and ! are the velocity of sound and the
frequency for each of the acoustic modes. From equation (34) we see that the low-
temperature speci® c heat contains information about the dimensionality of the
system, primarily through the density of phonon states »…!†, which is sensitive to
dimensionality. Because the low frequency phonons for an isolated single-wall
Phonons in carbon nanotubes 797

carbon nanotube have di€ erent characteristics from 2D graphite (see section 3.1) and
especially di€ erences in the low frequency phonon density of states (see ® gure 12 (c)),
we expect the speci® c heat for the SWNTs to have a di€ erent temperature
dependence than that for a 2D graphene sheet.
For an isolated graphene sheet, two of the (! ˆ vk) acoustic modes have a very
high sound velocity with vLA ˆ 24 km s¡1 for the LA mode and vTA ˆ 15 km s¡1 for
the TA mode, while the third out-of-plane transverse (ZAP) mode is described by a
parabolic dispersion relation, ! ˆ ¬k 2 , with ¬ ¹ 6 £ 10 ¡7 m2 s¡1 [151]. Thus the low-
temperature speci® c heat of a graphene sheet has a contribution from the in-plane
modes that is proportional to T 2 , and a smaller contribution from the ZAP mode
that goes as T 3 .
The e€ ects of combining weakly interacting graphene sheets in a correlated
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

stacking arrangement to form crystalline 3D graphite introduces dispersion along


the c axis, which raises the dimensionality of the system, and the speci® c heat
consequently exhibits a temperature dependence Cph ¹ T 2:2 at low temperatures.
Since the c-axis phonons have very low frequencies, thermal energies of ¹50 K are
su cient to occupy all the ZAP phonon states, so that for T > 50 K, the speci® c heat
of 3D graphite shows a T 2 dependence and basically exhibits 2D behaviour. This
crossover between 2D and quasi-3D behaviour, below which the interplanar coupled
behaviour becomes important, is denoted by a maximum in a plot of Cph versus T 2
[164± 166].
Now let us consider the speci® c heat for isolated single-wall carbon nanotubes
(i.e. non-interacting nanotubes) . Our discussion of !…k† for SWNTs (see section 3.1)
shows that SWNTs have 4 acoustic branches, which for small k have a linear
dispersion relation ! ˆ vk. In the low T regime, only the acoustic branches of the
SWNTs will be populated. The LA mode in the SWNTs is exactly analogous to the
LA mode in a graphene sheet. The TA modes in a SWNT, on the other hand, are a
combination of the in-plane and out-of-plane TA modes in a graphene sheet, while
the twist mode is special to carbon nanotubes and is related to the in-plane TA mode
(see section 3.1). These modes all show a linear ! ˆ vk relation at low k, but as
® gure 14 shows, the various acoustic branches ¯ atten at rather di€ erent k vectors.
Furthermore there is no mode for the isolated single-wall nanotube that is analogous
to the ZAP mode of a graphene layer which has a quadratic dependence on phonon
wavevector. All four acoustic phonon modes for the nanotube have high phonon
velocities: vLA ˆ 20 km s¡1 , vTA ˆ 9 km s¡1 , and vTW ˆ 15 km s¡1 for a (10, 10)
nanotube. The lowest lying optic mode for a (10, 10) nanotube (® gure 14) is the
squash mode, which is calculated to peak at 17 cm ¡1 for an isolated nanotube, which
corresponds roughly to a thermal energy of 25 K.
Since the phonon dispersion relations for all four acoustic branches are of the
form ! ˆ vk, at least at low wave vectors, and the phonon density of states is
expected to be constant, independent of !, we can then expect Cph to be linear in
temperature at low T [59]
k2 T -
hv
Cph ˆ const -B ; T ½ ; …35†
hv k B dt
where v is an appropriately averaged velocity of sound and dt is the nanotube
diameter. As the temperature is raised, the lower lying `subbands’ will become
occupied (see ® gure 14), and the power-law dependence T p of the speci® c heat will
show p increasing from 1 toward 2 approaching the power-law dependence p ˆ 2
798 M. S. Dresselhaus and P. C. Eklund

which is valid for an isolated graphene layer. Preliminary inelastic neutron scattering
studies suggest that the phonon density of states of SWNTs above ¹400 cm ¡1 are
very close to that of a graphene layer, suggesting that the transition to T 2 behaviour
for the speci® c heat is not complete until well above room temperature [167].
Benedict et al. [59] have shown theoretically that Cph / T , provided that the
nanotube diameter dt and the temperature T are su ciently small, i.e.
T ½ 2hv - =…k B dt †, and we note that the temperature of validity of this relationship
depends on the nanotube diameter. The crossover from a linear to a quadratic T
dependence is estimated to occur at T ’ 2hv - =…k B dt † or at ¹ 100 K for a (10,10)
armchair nanotube.
The electron contribution to the speci® c heat is also expected to be linear in T for
metallic nanotubes, especially at low T , while the contribution of the semiconducting
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

nanotubes is much smaller at low T and is expected to exhibit an exponential


dependence since the density of states at EF in this case vanishes. Therefore, the ratio
Cel =Cph is expected to be independent of T at low T where Cel has the best
opportunity to contribute signi® cantly to the speci® c heat. But since v =vF ’ 10 ¡2 ,
where v is the velocity of sound and vF is the Fermi velocity, and the carrier density
for SWNTs is lower than in graphite, we expect phonons to dominate the speci® c
heat of isolated SWNTs over the entire temperature range of measurement. We will
see below that dimensional crossover e€ ects should also be present in bundles of
SWNTs because of the weak force constants between carbon atoms on adjacent
nanotubes.
Figure 14 shows the low-energy phonon dispersion relations for a single (10, 10)
nanotube. Rolling a graphene sheet into a nanotube has two major e€ ects on the
phonon dispersion relations as discussed in section 3.1. First, the two-dimensional
phonon dispersion relations of the sheet is collapsed onto one dimension because of
the periodic boundary conditions in the circumferential direction of the nanotube,
leading to the development of discrete phonon `subbands’ . At the G-point, the
splitting between the subbands is of order [59]
-
2hv
D Eˆ d : …36†
t

The second e€ ect of rolling the graphene sheet to form a nanotube is to create a new
acoustic mode (the twist mode) and to create low-energy subbands by zone folding
the acoustic phonon branches, as discussed in section 3.1.
Experimental measurements by Yi et al. [168] of the speci® c heat for MWNTs
(20 < dt < 40 nm) in the temperature range 10 < T < 300 K show a linear depen-
dence of Cp on T (see ® gure 61). The inset to ® gure 61 compares Cp …T † for MWNTs
and graphite, emphasizing the di€ erence in behaviour in the low temperature speci® c
heat. The extraordinarily large range of this linear T dependence (10 < T < 300 K)
is somewhat unexpected considering the large diameters of these MWNTs. The
authors attributed this linear T behaviour out to relatively high T values to the
essentially isolated nanotube shell constituents of these MWNTs [168]. Thus
the authors conclude that the interaction between carbon atoms on adjacent walls
is very weak because of the turbostrati c stacking of sequential layers.
The speci® c heat measurements of Mizel et al. [164] on MWNTs and on ropes of
SWNTs (see ® gure 62) give similar results to Yi et al. [168] in the temperature range
100 < T < 200 K, regarding both the absolute magnitude of the speci® c heat and the
linear temperature dependence of the speci® c heat. The roughly quadratic depen-
Phonons in carbon nanotubes 799
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

Figure 61. Measured speci® c heat of MWNTs [164, 168], compared to the calculated
phonon speci® c heat of graphene, graphite and isolated nanotubes [54].

Figure 62. Measured speci® c heat of MWNTs [164, 168], compared to the calculated
phonon speci® c heat of graphite and ropes of SWNTs.

dence of the speci® c heat at low temperature in ® gure 62 for ropes of SWNTs,
MWNTs and graphite is clearly di€ erent from the low temperature behaviour in
® gure 61. More detailed plots of the speci® c heat data in ® gure 62 as Cp =T versus T
(not shown) [164] reveal some di€ erences in behaviour between the SWNT ropes, the
800 M. S. Dresselhaus and P. C. Eklund

MWNTs and graphite below 50 K, and especially below 5 K. Much smaller


di€ erences are found between the behaviour of the MWNTs and graphite at low
temperature [164]. Further work on better characterized samples is necessary to
determine the speci® c heat for nanotubes at low temperature and to determine the
di€ erences in Cp …T † between SWNTs, SWNT bundles, MWNTs and graphite.
Calculations have been carried out to consider the e€ ect of inter-tube interactions
in a SWNT rope on the phonon modes at low frequencies [59]. These calculations
show that inter-tube interactions introduce a weak dispersion in the transverse
direction, just as occurs in 3D graphite relative to a graphene sheet. In addition,
some calculations indicate that the twist mode becomes an optical mode because of
the presence of a non-zero shear modulus between neighbouring nanotubes [59].
More detailed experiments on highly puri® ed and well characterized samples are
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

needed to evaluate the importance of these inter-tube interactions on the low


temperature speci® c heat.
Also, the magnetic ® eld dependence of the speci® c heat might be measurable
because of the large e€ ect that the magnetic ® eld could have on the density of
electronic states and therefore also on the interband transitions. Such e€ ects have
been considered theoretically [169], but no experimental observations of such e€ ects
have been reported thus far.

5.2. T hermal conductivity


Diamond and graphite (in-plane) display the highest measured thermal con-
ductivity of any known material, and, of course, the thermal conduction is entirely
by phonons in the case of diamond, and almost totally by phonons in the case of the
basal plane of graphite. The apparent long-range crystallinity of nanotubes suggests
that the thermal conductivity of nanotubes along the nanotube axis should also be
high. The thermal conductivity therefore provides an important tool for probing the
interesting low-energy phonon structure of nanotubes.
In the one-dimensional limit, the phonon thermal conductivity can be written as:
X
µph ˆ Cv2 ½ ; …37†
where C, v and ½ are the heat capacity per unit volume, phonon group velocity along
the nanotube axis and the relaxation time of a given phonon mode, and the sum in
equation (37) is over all phonon modes. While the phonon thermal conductivity
cannot be measured directly, the electronic contribution µel can generally be
determined from the electrical conductivity ¼ by the Wiedemann± Franz law.
µel
º L 0; …38†
¼T
where the Lorenz number for a free electron gas is L 0 = 2:45 £ 10 ¡8 …V =K†2 and the
total thermal conductivity µ is given by µ ˆ µph ‡ µel . Thermal conductivity meas-
urements to date [54, 168, 170] indicate that the phonon contribution µph is
dominant µph ¾ µel in MWNTs and SWNTs at all temperatures. Therefore the
study of µ…T † is important for describing phonon properties of carbon nanotubes,
the subject of this review article.
In general, at high temperatures, the dominant contribution to the inelastic
phonon scattering processes is phonon ± phonon Umklapp scattering, while at low
temperatures inelastic phonon scattering is generally by boundary or defect
scattering. Thus at low temperature (T ½ YD ), the relaxation time ½ is expected
Phonons in carbon nanotubes 801
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

Figure 63. Thermal conductivity of a SWNT mat sample. The linear T dependence of µ…T †
at low T is emphasized in the inset [54, 170].

to be independent of T so that the temperature dependence of the phonon thermal


conductivity is the same as that of the speci® c heat discussed in section 5.1. Since the
Debye temperature YD is large (>2000 K) for graphite and presumably also for the
1D nanotubes, it is easy to satisfy T ½ YD even at room temperature. However, in
an anisotropic material like graphite or 1D carbon nanotubes, the weighting of each
phonon mode by the factor v2 ½ becomes especially important, since for an isotropic
material the thermal conductivity is most sensitive to those phonon modes with the
highest group velocities and the longest scattering times. In analogy to graphite,
where the in-plane thermal conductivity can be closely approximate d by neglecting
the inter-planar coupling [151], it is expected that for SWNTs, µ…T † along the
nanotube axis should be insensitive to inter-tube phonon modes, which have
signi® cantly smaller v than for on-tube phonons, and, in a ® nite bundle, a
signi® cantly reduced ½ is expected for inter-tube modes as compared to the on-tube
modes. In this respect, µ…T † in SWNTs is expected to display 1D properties more
clearly than the speci® c heat.
The expected linear T dependence of µ…T † at low temperature 10 < T < 25 K is
shown in ® gure 63 for a SWNT mat sample (dt ¹ 1:4 nm), and for higher
temperatures a more rapid increase in µ…T † was found [54, 170]. Above 40 K,
µ…T † in ® gure 63 can be approximated by a T 2 dependence until about 100 K, above
which the slope of µ…T † begins to decrease, indicative of the onset of phonon±
phonon Umklapp scattering. From an estimation of the volume density of
nanotubes in this mat sample, the room-temperatur e magnitude of the thermal
conductivity (µmat ) was estimated to be ¹35 W m ¡1 K ¡1 . Simultaneous measurement
of the electrical and thermal conductance of a given sample yields a ratio µ =¼T
which is more than a factor of 102 at all temperatures, con® rming the dominance of
phonons in the thermal conductivity of SWNTs. The linear T dependence of µ…T †
for T < 25 K in ® gure 63 implies a linear T low-temperature dependence for the
speci® c heat of SWNTs, which was not seen experimentally in ® gure 62 by this group
(see section 5.1) [164].
802 M. S. Dresselhaus and P. C. Eklund

Figure 64. Thermal conductivity of a bulk sample of aligned MWNTs and the data for
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

µ…T † is ® t by a T 2 dependence (curve) up to about 120 K [54, 168].

Measurements of the thermal conductivity for aligned multi-wall nanotubes have


also been reported [168] using the 3! measurement technique. These results for µ…T †
on multi-wall nanotubes (see ® gure 64) over the temperature range 10 < T < 300 K
look qualitatively similar to those in ® gure 63 for SWNTs, except perhaps at low
temperature, where there is no obvious region of linear T dependence for the
MWNTs, and near room temperature, where the multi-wall µ…T † in ® gure 64 shows
less saturation than the µ…T † for the SWNT in ® gure 63. It is curious that the µ…T †
for the MWNT in ® gure 64 show a T 2 dependence at low T while a linear C …T † ¹ T
dependence is clearly seen for the same sample in ® gure 61. One possible explanation
for these di€ erences in behaviour may be that all the walls of the MWNTs contribute
to the speci® c heat, but that thermal contact in the thermal conductivity meas-
urements is only made to the outermost wall of the MWNTs. The inner walls, having
smaller diameters, would be expected to behave more like SWNTs, but the outer-
most wall with an average diameter of 30 nm would be expected to be the only part
of the MWNT that contributes signi® cantly to the thermal conductivity. Further-
more, the outermost wall with a diameter of ¹30 nm is expected to be more like 2D
graphite than like a SWNT. Since the MWNTs of this sample had 10± 30 walls [168],
this interpretation would imply that the measured µ…T † is perhaps one order of
magnitude lower than the ideal µ…T † because thermal contact is only made to one of
the walls of each nanotube. If this interpretation is correct, the magnitude of µ…T †
for SWNTs could be more comparable to that of graphite ( ¹2000 W m ¡1 K ¡1 at
300 K) than would be implied by the value ( ¹27 W m ¡1 K ¡1 at 300 K) given in
® gure 64. Measurements of µ…T † above 200 K are subject to radiation losses which
must be carefully taken into account.
In contrast, measurements of µ…T † for carbon ® bres [116] show µ…T † to follow a
T 2:3 temperature dependence from low T to ¹100 K and which then increases more
slowly until a peak is reached at ¹150 K, above which the thermal conductivity
begins to decrease with increasing temperature, from ¹6000 W m ¡1 K ¡1 at the peak
to ¹2000 W m ¡1 K ¡1 at room temperature for the most crystalline vapour grown
carbon ® bres [116]. This decrease in thermal conductivity is due to the onset of
strong phonon± phonon Umklapp scattering, which becomes more e€ ective with
increasing temperature as higher-energy phonons are thermally populated. For less
structurally ordered carbon ® bres of various types (see ® gure 65), µ…T † can be ® t to a
Phonons in carbon nanotubes 803
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

Figure 65. Log± log plot of the temperature dependence of the thermal conductivity of
various carbon ® bres for a variety of precursor materials and heat treatment
temperatures. The ® bre with the highest thermal conductivity is a vapour grown
carbon ® bre heat treated to ¹2800 8C and shows a temperature dependence of µ…T †
close to that of single crystal graphite [116, 171, 172].

T 2 dependence, indicative of the two-dimensional nature of these disordered carbons


(called turbostratic carbons because they lack interplanar stacking order).
We now discuss the experimental low-temperature thermal conductivity of the
SWNTs in more detail [170, 173]. At low temperature, some carbon nanotube
samples have been found to have a linear temperature dependence for µ…T † and
deviations from this linear behaviour begin to appear between 25± 35 K, as shown in
® gures 63 and 66 [170]. At low temperature, only the four acoustic modes are
804 M. S. Dresselhaus and P. C. Eklund
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

Figure 66. Points denote the measured low-temperature thermal conductivity µ…T † of
SWNTs, compared to a model based on ! ˆ vk for only acoustic branches (long
dashed curve) and to a model (solid curve) that includes contributions from both the
2
acoustic branch and the lowest optical phonon !2 ˆ !20 ‡ …vk† branch which is
indicated by the dotted curve (see text) [54].

populated (see ® gure 14), while at slightly higher temperatures the lowest zone-
folded phonon `subband’ (at 17 cm ¡1 or an equivalent thermal temperature of 25 K)
should begin to contribute. This behaviour can be accounted for by using a
simpli® ed model (see ® gure 66), considering acoustic phonon contributions as well
as optical excitations. In a simple zone-folding picture, the acoustic band has a
dispersion ! ˆ vk, where k is the phonon wave vector, and the (doubly degenerate)
lowest energy phonon subband has been modelled by a dispersion relation

!2 ˆ v2 k 2 ‡ !20 ; …39†
where the subband energy, as approximate d by equation (36), was written as
- 0 ˆ 2hv
h! - =dt and for simplicity the same velocity of sound was used as for the
acoustic branch [54]. This approximation is equivalent to taking ! ˆ vk in the limit
!0 ! 0. The thermal conductivity from each branch in ® gure 66 can then be
estimated using equation (37) and assuming a constant scattering time ½, and
estimating the frequency !0 for the lowest subband edge from the calculated
dispersion relations, or from the measured Raman spectra.
Figure 66 shows a comparison between the measured µ…T † of SWNTs, compared
to calculations based on the model discussed above. In making the ® ts, v was chosen
to be 2 km s¡1 , which is a very low value for sp2 carbon-based materials. The top
dashed line represents µ…T † of the acoustic band which is linear in T , as expected for
a 1D linear ! ˆ vk dispersion relation and a constant ½ . The lower dashed line in
® gure 66 represents the contribution from the lowest optical phonon subband, which
is zero at low temperatures, and begins to contribute near 35 K (24 cm ¡1 ). The solid
curve, which is the sum of the two contributions , ® ts the experimental data quite well
with this choice of parameters. The measured linear slope can be used to calculate
the scattering time, or, equivalently, a scattering length [54].
This model o€ ers some support to the interpretation that the observed linear T
dependence of the low-T thermal conductivity is due to one-dimensional behaviour,
Phonons in carbon nanotubes 805

making nanotubes the only material which is of small enough size for this type of 1D
phonon quantization to be observable. The restricted geometry of the SWNTs may
also a€ ect the thermal conductivity at high temperature where Umklapp scattering
in general is the dominant scattering mechanism, but Umklapp scattering is expected
to be suppressed in one dimension because of the low availability of appropriate
phonons for conservation of energy and wave vector [174]. In this connection it is
interesting to note that the Umklapp process above 100 K seems to be stronger for
the MWNTs (® gure 64) than for the SWNTs (® gure 63). Measurements on well
characterized and puri® ed nanotubes with di€ erent diameters, as well as additional
theoretical modelling, should prove interesting [169, 175].
Calculations of the magnetic ® eld dependence of the thermal conductivity have
been carried out [175], indicating step structures associated with the Zeeman
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

splitting. These e€ ects may be observable below 1 K, and could a€ ect thermal
properties through a magnetic ® eld dependent electron± phonon interaction.

5.3. T hermopower
The thermoelectric power (TEP) or Seebeck coe cient S gives the voltage
developed across a sample exposed to a temperature gradient. The dominant
contribution to the TEP is usually purely electronic (the `drift’ contribution), though
in some materials, such as graphite, an electron± phonon scattering contribution (the
`phonon drag’ e€ ect) also becomes signi® cant in certain temperature ranges,
especially at low temperature. Thus even though the thermal properties of nanotubes
are dominated by phonons, we expect that the TEP will be primarily governed by the
electronic drift contribution, the electronic band-structur e of the nanotubes, and
electron scattering mechanisms.
We ® rst consider the expected TEP of a metallic SWNT by considering
calculations for the nanotube electronic band-structur e near the Fermi level (see
® gure 5). For a conventional metal, the electronic contribution to the thermopower
S …T † is linear in T at low temperature and the magnitude of S is given by the Mott
expression [176], which reduces in one dimension to

p2 k 2B T v0 ½ 0
S 1¡D ˆ ¡ ‡ ; …40†
3e v ½

where v is the electronic band velocity, ½ is the electronic relaxation time, the prime
superscript indicates that the derivatives of v and ½ are with respect to energy, and
the expression in brackets is evaluated at the Fermi level. In general, S < 0 for
electron-like systems, where v 0 > 0, while S > 0 for hole-like materials. The simple
models thus far developed for metallic SWNTs have assumed that the overlap
integral in the tight binding approximation for a graphene sheet vanishes, yielding
symmetric electron and hole bands (mirror bands). For graphite itself, the C± C
nearest neighbour band overlap integral is s º 0:13 leading to substantial asymmetry
in the electronic structure of the valence and conduction bands [2]. The electronic
energy bands for SWNTs near EF (see ® gure 5) are highly linear at EF . Therefore, the
® rst term in equation (40) should be close to zero under the assumption of mirror
bands. The second term should also be small, because ½ should be approximately
constant when the electronic density of states is constant (as occurs for SWNTs in
accordance with ® gure 6 (b)). Thus an isolated metallic nanotube should have a TEP
which is close to zero. This is a general result based on the linearity and the electron±
806 M. S. Dresselhaus and P. C. Eklund

hole symmetry of the simple band-structur e shown in ® gure 5. Other mechanisms


giving rise to thermopower (such as phonon drag, or carrier localization) also are
expected to yield S ˆ 0 for systems with electron± hole symmetry. Away from the
Fermi level, higher subbands display signi® cant curvature. The subbands above EF
are electron-like, while the subbands below EF are hole-like. Therefore, a hole-doped
metallic nanotube should display a positive TEP, while an electron-doped nanotube
should display a negative TEP. In view of the positive contributions to S from holes
and negative contributions to S from electrons, it can be expected that measurements
of S …T † for well characterized electron and hole concentrations could be used to
study the departures from electron± hole symmetry and to give information of the
appropriate value for the overlap integral that should be used to model SWNTs.
We next consider the expected S …T † of a semiconducting nanotube, based on a
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

typical nanotube band structure (see ® gure 5). When the Fermi level is in the gap, the
contribution from the conduction or valence band is given by
kB ¯ ln ½ 5
S semi …T † ˆ … ±
‡ †
‡ ;
e k B T ¯ ln " 2
…41†

where ± is the electrochemical potential and represents the energy di€ erence between
the band edge and the Fermi level. Thus, when there is a zero density of states at the
Fermi level, the TEP varies as (1 =T ) for a semiconductor, rather than as T for a
metal. The total TEP will be the sum of the contributions from both the conduction
and valence bands, weighted by their respective conductivities. In an intrinsic
semiconducting nanotube, the Fermi level lies exactly between the conduction and
valence bands where the density of states vanishes. Assuming mirror bands, the
contributions from the conduction and valence bands should be equal in magnitude
and opposite in sign, and the total thermopower from an intrinsic semiconducting
nanotube should be zero, just as in the case of metal nanotubes under the
assumption of mirror bands. Movement of the Fermi level toward one of the bands
will cause that band to dominate the TEP, producing a semiconducting-like TEP
which is positive for hole-doping and negative for electron-doping. Higher doping
levels will move the Fermi level into the conduction or valence band, producing a
degenerate semiconductor with a metallic (i.e. linear in T ) electron-like or hole-like
thermopower, respectively.
Figure 67 shows the temperature dependence of the thermopower reported for a
bulk `mat’ sample of SWNT ropes [173]. A linear T dependence of S …T † is observed
at low T suggesting that metallic nanotubes make the dominant contribution in this
limit. In light of the expected small TEP of intrinsic metallic and semiconducting
nanotubes, the observed magnitude ( ¹70 mV K ¡1 at 300 K) is surprisingly large. The
measured thermopower (® gure 67) decreases with decreasing T and at low T
approaches zero with a linear slope, consistent with a non-zero density of states at
the Fermi level.
Other measurements of S …T † on as-grown mat samples prepared by either the
pulsed laser deposition (PLD) or arc method yield similar results for S …T †, which are
also similar to the results in ® gure 67 [155]. However, S …T † was strongly reduced (by
as much as a factor of 4 by doping with iodine) but S …T † was still found to be
positive after iodine doping and S …T † still showed a linear T dependence at low T .
Temperature dependent resistance R…T † measurements on the same samples showed
a decrease in R…T † by about a factor of 30, indicating that the e€ ect of saturation
iodine doping is to introduce a large concentration of holes into the p-electron bands
Phonons in carbon nanotubes 807
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

Figure 67. The temperature dependence of the thermopower of a bulk mat sample of as-
grown SWNTs from 4.2 K to 300 K [54, 173].

of the SWNTs [155]. Adsorbed gases are also found to sensitively a€ ect the
magnitude of S [177].
Attempts have been made to model the observed S …T † of SWNTs in ® gure 67 in
terms of metallic and semiconducting nanotubes in parallel, and S …T † for these two
types of nanotubes in parallel has been calculated by weighting S for each nanotube
type by its conductance and dividing by the sum of the conductances [173]. This
model, however, fails to ® t the experimental data for reasonable values of the ® tting
parameters. Another approach to ® tting the TEP measurements for SWNTs was
made by assuming that residual amounts of magnetic impurities from the catalyst
used to promote the SWNT growth process remains in the nanotube sample after
growth is completed [178]. The preparation of a SWNT sample free from transition
metal catalyst would be interesting for S …T † measurements to determine asymmetries
between the electronic structure and occupation of the conduction and valence
bands. At present, the connection between thermopower measurements and the
phonons in SWNTs remains to be clari® ed.

6. Concluding remarks
The ® eld of carbon nanotube research is remarkable in terms of the unique
physical properties of the carbon nanotubes, some of which are reviewed in this
article. In most sub® elds of condensed matter physics, experimental results have led
the way and theoretical explanations have followed to give the sub® eld a ® rm
foundation. However, for the case of research on carbon nanotubes, theoretical
predictions have often led experimental investigations. This situation is directly
related both to the di culty in synthesis of su cient quantities of pure and well-
characterized materials for experimental investigations and to the fundamental
nature of SWNTs as an attractive prototype system for the theoretical investigation
of 1D phenomena.
Although extensive experimental study of the phonons in carbon nanotubes is
quite recent [28], progress in the experimental aspects of this ® eld since 1997 has been
808 M. S. Dresselhaus and P. C. Eklund

rapid, encouraging theorists to develop more sophisticated models, especially in


connection with the various resonant Raman phenomena. Despite this rapid
progress, the ® eld is still at an early stage of development, from both an experimental
and theoretical standpoint. Thus, in writing this review article, the authors have
attempted to focus on those models and measurements that they feel will stand the
test of time and will help to guide future developments in this ® eld.
At this point it is not clear whether the applications of carbon nanotubes will be
su cient to insure long term interest in carbon nanotube-base d materials. However,
because of the unique properties of single-wall carbon nanotubes and their use as a
prototype 1D materials system, it is likely that scienti® c interest in carbon nanotubes
will continue to be a focus of the condensed matter research community for a few
more years to come.
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

Acknowledgements
The authors gratefully acknowledge the helpful and informative discussions with
Drs G. Dresselhaus, J.-C. Charlier, A. Jorio, A. M. Rao, Professors M. Endo,
R. Saito, H. Kataura, M. A. Pimenta, Mr K. A. Williams, and Ms Sandra Brown.
They are also thankful to many other colleagues for their assistance with the
preparation of this article. The research of MD is supported by NSF grant
DMR 98-04734, and the work of PCE by NSF grants OSR-9452895 and DMR 98-
09686 and by the University of Kentucky Center for Applied Energy Research.

References
[1] D RESSELHAUS, M . S., D RESSELHAUS, G ., and E KLUND, P . C ., 1996, Science of
Fullerenes and Carbon Nanotubes (New York: Academic Press).
[2] SAITO, R ., D RESSELHAUS, G ., and D RESSELHAUS, M . S., 1998, Physical Properties of
Carbon Nanotubes (London: Imperial College Press).
[3] D RESSELHAUS, M . S., D RESSELHAUS, G ., and SAITO, R ., 1992, Phys. Rev. B, 45, 6234.
[4] SAITO, R ., F UJITA, M ., D RESSELHAUS, G ., and D RESSELHAUS, M . S., 1992, Appl. Phys.
L ett., 60, 2204.
[5] T ANS, S. J ., D EVORET, M . H ., D AI , H ., T HESS, A ., SMALLEY, R . E ., G EERLIGS, L . J .,
and D EKKER, C ., 1997, Nature (London), 386, 474.
[6] T HESS, A ., L EE, R ., N IKOLAEV, P ., D AI, H ., P ETIT, P ., R OBERT , J ., X U, C ., L EE,
Y . H ., K IM, S. G ., R INZLER, A . G ., C OLBERT , D . T ., SCUSERIA, G . E ., T OMAíNEK, D .,
F ISCHER, J . E ., and SMALLEY, R . E ., 1996, Science, 273, 483.
[7] H ENRARD, L ., L OISEAU, A ., J OURNET, C ., and BERNIER, P ., 2000, Eur. J. Phys., to be
published.
[8] L OISEAU, A ., private communication.
[9] J OURNET, C ., M ASER , W . K ., BERNIER, P ., L OISEAU, A ., L AMY DE LA C HAPELLE, M .,
L EFRANT, S., D ENIARD, P ., L EE, R ., and F ISCHER, J . E ., 1997, Nature (London), 388,
756.
[10] C HARLIER, J .-C ., and M ICHENAUD, J . P ., 1993, Phys. Rev. L ett., 70, 1858.
[11] C HARLIER, J .-C ., E BBESEN , T . W ., and L AMBIN, P H ., 1996, Phys. Rev. B, 53, 11108.
[12] I IJIMA, S., 1991, Nature (London), 354, 56.
[13] W ILDOï ER , J . W . G ., VENEMA, L . C ., R INZLER, A . G ., SMALLEY, R . E ., and D EKKER,
C ., 1998, Nature (London), 391, 59.
[14] O DOM, T . W ., H UANG, J . L ., K IM, P ., and L IEBER, C . M ., 1998, Nature (London), 391,
62.
[15] I IJIMA, S., 1993, Mater. Sci. Eng., B19, 172.
[16] SAITO, R ., D RESSELHAUS, G ., and D RESSELHAUS, M . S., 1992, Chem. Phys. L ett., 195,
537.
[17] M INTMIRE, J . W ., D UNLAP , B. I ., and W HITE, C . T ., 1992, Phys. Rev. L ett., 68, 631.
Phonons in carbon nanotubes 809

[18] H AMADA, T ., F URUYAMA, M ., T OMIOKA, T ., and E NDO, M ., 1992, J. mater. Res., 7,


1178 ; 1992, ibid., 7, 2612.
[19] H ARIGAYA, K ., 1992, Chem. Phys. L ett., 189, 79.
[20] T ANAKA, K ., O KADA, M ., O KAHARA, K ., and Y AMABE, T ., 1992, Chem. Phys. L ett.,
191, 469.
[21] SAITO, R ., F UJITA, M ., D RESSELHAUS, G ., and D RESSELHAUS, M . S., 1992, Electrical,
Optical and Magnetic Properties of Organic Solid State Materials, MRS Symposia
Proceedings, Boston, edited by L. Y. Chiang, A. F. Garito and D. J. Sandman
(Pittsburgh, PA : Materials Research Society Press), p. 333.
[22] SAITO, R ., F UJITA, M ., D RESSELHAUS, G ., and D RESSELHAUS, M . S., 1992, Phys. Rev.
B, 46, 1804.
[23] L OUIE, S. G ., 2000, Physics and Chemistry of Materials with L ow-Dimensional
Structures : Fullerene-Based Materials, Vol. 23, edited by W. Andreoni (Kluwer
Academic), p. 381.
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

[24] P AINTER, G . S., and E LLIS, D . E ., 1970, Phys. Rev. B, 1, 4747.


[25] SAITO, R ., D RESSELHAUS, G ., and D RESSELHAUS, M . S., 1993, J. appl. Phys., 73, 494.
[26] D RESSELHAUS, M . S., 1998, Nature (London), 391, 19.
[27] C HARLIER, J .-C ., private communication.
[28] R AO, A . M ., R ICHTER, E ., BANDOW, S., C HASE, B., E KLUND, P . C ., W ILLIAMS, K . W .,
M ENON, M ., SUBBASWAMY, K . R ., T HESS, A ., SMALLEY, R . E ., D RESSELHAUS, G ., and
D RESSELHAUS, M . S., 1997, Science, 275, 187.
[29] K ATAURA, H ., K UMAZAWA, Y ., M ANIWA, Y ., U MEZU, I ., SUZUKI, S., O HTSUKA, Y .,
and A CHIBA, Y ., 1999, Synthetic Met., 103, 2555.
[30] D RESSELHAUS, G ., P IMENTA, M . A ., SAITO, R ., C HARLIER, J .-C ., BROWN, S. D . M .,
C ORIO, P ., M ARUCCI, A ., and D RESSELHAUS, M . S., 2000, Science and Applications of
Nanotubes, edited by D. Toma nek, and R. J. Enbody (New York: Kluwer Academic) ;
1999, Proceedings of the International W orkshop on the Science and Applications of
Nanotubes, Michigan State University, East Lansing, MI, USA, 24± 27 July, p. 275.
[31] Y AKOBSON, B. L ., and A VOURIS, P ., 2000, Carbon Nanotubes, edited by M. S.
Drasselhaus, G. Drasselhaus and P. Avouris (Berlin: Springer-Verlag), in press.
[32] SAITO, R ., D RESSELHAUS, G ., and D RESSELHAUS, M . S., 2000, Phys. Rev. B, 61, 2981.
[33] SATTLER, K ., 1995, Carbon, 33, 915.
[34] K IM, P ., O DOM, T ., H UANG, J .-L ., and L IEBER , C . M ., 1999, Phys. Rev. L ett., 82,
1225.
[35] K ATAURA, H ., K UMAZAWA, Y ., K OJIMA, N ., M ANIWA, Y ., U MEZU , I ., M ASUBUCHI,
S., K AZAMA, S., Z HAO, X ., A NDO, Y ., O HTSUKA, Y ., SUZUKI, S., and A CHIBA, Y .,
1999, Proceedings of the International W inter School on Electronic Properties of Novel
Materials (IW EPNM’ 99), edited by H. Kuzmany, M. Mehring and J. Fink
(Woodbury, NY : American Institute of Physics) ; 2000, AIP Conference Proceedings (in
press).
[36] K ATAURA, H ., K IMURA, A ., O HTSUKA, Y ., SUZUKI, S., M ANIWA, Y ., H ANYU, T ., and
A CHIBA, Y ., 1998, Jpn. J. appl. Phys., 37, L616.
[37] K AZAOUI, S., M INAMI, N ., J ACQUEMIN , R ., K ATAURA, H ., and A CHIBA, Y ., 1999,
Phys. Rev. B, 60, 13339.
[38] F AN, X ., D ICKEY, E . C ., E KLUND, P . C ., W ILLIAMS, K . A ., G RIGORIAN, L ., BUCZKO,
R ., P ANTELIDES, S. T ., and P ENNYCOOK, S. J ., 2000, Phys. Rev. L ett., 84, 4621.
[39] K ATAURA, H ., K UMAZAWA, Y ., K OJIMA, N ., M ANIWA, Y ., U MEZU , I ., M ASUBUCHI,
S., K AZAMA, S., O HTSUKA, Y ., SUZUKI, S., and A CHIBA, Y ., 2000, Molec. Cryst. liq.
Cryst, to be published.
[40] SAITO, R ., and K ATAURA, H ., 2000, Carbon Nanotubes, edited by M. S. Dresselhaus,
G. Dresselhaus and P. Avouris (Berlin: Springer-Verlag), in press.
[41] J ISHI, R . A ., VENKATARAMAN, L ., D RESSELHAUS, M . S., and D RESSELHAUS, G ., 1993,
Chem. Phys. L ett., 209, 77.
[42] R ICHTER, E ., and SUBBASWAMY, K . R ., 1997, Phys. Rev. L ett., 79, 2738.
[43] SAITO, R ., T AKEYA, T ., K IMURA, T ., D RESSELHAUS, G ., and D RESSELHAUS, M . S.,
1998, Phys. Rev. B, 57, 4145.
[44] J ISHI, R . A ., I NOMATA, D ., N AKAO, K ., D RESSELHAUS, M . S., and D RESSELHAUS, G .,
1994, J. phys. Soc. Jpn., 63, 2252.
810 M. S. Dresselhaus and P. C. Eklund

[45] D ING, Q ., J IANG, Q ., J IN, Z ., and T IAN, D ., 1996, Fullerene Sci. Technol., 4, 31.
[46] C HARLIER, A ., M CR AE, E ., C HARLIER, M .-F ., SPIRE, A ., and F ORSTER, S., 1998,
Phys. Rev. B, 57, 6689.
[47] M ENON, M ., R ICHTER, E ., and SUBBASWAMY, K . R ., 1996, J. chem. Phys., 104, 5875.
[48] K HOKHRYAKOV, N ., SAVINSKIuI, S., and M OLINA, J ., 1995, JET P L ett. (Pis’ma Zh.
Eksp. T eor.), 62, 617.
[49] Y U, J ., K ALIA, K ., and VASHISHTA, P ., 1995, J. chem. Phys., 103, 6697.
[50] K Uï RTI , J ., K RESSE, G ., and K UZMANY, H ., 1998, Phys. Rev. B, 58, R8869.
[51] SANCHEZ-P ORTAL, D ., A RTACHO, E ., SOLER , J . M ., R UBIO, A ., and O RDEJOíN, P .,
1999, Phys. Rev. B, 59, 12678.
[52] A IZAWA, T ., SOUDA, R ., O TANI, S., I SHIZAWA, Y ., and O SHIMA, C ., 1990, Phys. Rev.
B, 42, 11 469.
[53] O SHIMA, C., A IZAWA, T ., SOUDA, R ., I SHIZAWA, Y ., and SUMIYOSHI, Y ., 1988, Solid
State Commun. , 65, 1601.
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

[54] H ONE, J ., 2000, Carbon Nanotubes, edited by M. S. Dresselhaus, G. Dresselhaus and


P. Avouris (Berlin: Springer-Verlag), in press.
[55] P OPOV, V. N ., VAN D OREN, V. E ., and BALKANSKI, M ., 2000, Phys. Rev. B, 61, 3078.
[56] M AEDA, T ., and H ORIE, C ., 1999, Physica B, 263± 264, 479.
[57] C HARLIER, J . C ., G ONZE, X ., and M ICHENAUD, J . P ., 1994, Carbon, 32, 289.
[58] L AMBIN, P H., and M EUNIER, V., 1998, Proceedings of the W inter School on Electronic
Properties Novel Materials, edited by H. Kuzmany, J. Fink, M. Mehring and S. Roth,
Kirchberg Winter School, AIP Conference Proceedings 442 (Woodbury: AIP), p. 504.
[59] BENEDICT, L . X ., L OUIE, S. G ., C OHEN, and M . L ., 1996, Solid State Commun. , 100,
177.
[60] E KLUND, P . C ., H OLDEN, J . M ., and J ISHI, R . A ., 1995, Carbon, 33, 959.
[61] R OBERTSON, D . H ., BRENNER, D . W ., and M INTMIRE, J . W ., 1992, Phys. Rev. B, 45,
12 592.
[62] M INTMIRE, J . W ., and W HITE, C . T ., 1995, Carbon, 33, 893.
[63] R ICHTER, E ., and E KLUND, P . C ., 1999, unpublished.
[64] VENKATESWARAN , U . D ., R AO, A . M ., R ICHTER, E ., M ENON, M ., R INZLER , A .,
SMALLEY, R . E ., and E KLUND, P . C ., 1999, Phys. Rev. B, 59, 10928.
[65] K AHN, D ., and L U , J . P ., 1999, Phys. Rev. B, 60, 6535.
[66] K WON, Y . K ., SAITO, S., and T OMAí NEK , D ., 1998, Phys. Rev. B, 58, R13 314.
[67] K WON, Y .-K ., T OMAíNEK, D ., L EE, Y . H ., L EE, K . H ., and SAITO, S., 1998, J. mater.
Res., 13, 2363.
[68] T OMAíNEK, D ., 1999, private communication.
[69] D ELANEY, P ., J OON-C HOI, H ., I HM, J ., L OUIE, S., and C OHEN, M . L ., 1999, Phys. Rev.
B, 60, 7899.
[70] U GAWA, A ., R INZLER, A . G ., and T ANNER, D . B., 1999, Phys. Rev. B, 60, R11 305.
[71] D ELANEY, P ., C HOI, H . J ., I HM, J ., L OUIE, S. G ., and C OHEN, M . L ., 1998, Nature
(London), 391, 466.
[72] M ENON, M ., and SUBBASWAMY, K . R ., 1994, Phys. Rev. B, 50, 11577.
[73] A LVAREZ , L ., R IGHI, A ., G UILLARD, T ., R OLS, S., A NGLARET, E ., L APLAZE, D ., and
SAUVAJOL, J .-L ., 2000, Chem. Phys. L ett., 316, 186.
[74] H ENRARD, L ., H ERNAíNDEZ , E ., BERNIER, P ., and R UBIO, A ., 1999, Phys. Rev. B, 60,
R8521.
[75] H IURA, H ., E BBESEN , T . W ., T ANIGAKI, K ., and T AKAHASHI, H ., 1993, Chem. Phys.
L ett., 202, 509.
[76] C HANDRABHAS, N ., SOOD, A . K ., SUNDARARAMAN, D ., R AJU, S., R AGHUNATHAN,
V. S., R AO, G . V. N ., SATRY, V. S., R ADHAKRISHNAN , T . S., H ARIHARAN, Y .,
BHARATHI, A ., and SUNDAR, C . S., 1994, PRAMANAÐ J. Phys., 42, 375.
[77] H OLDEN, J . M ., Z HOU, P ., BI, X .-X ., E KLUND, P . C ., BANDOW, S., J ISHI, R . A .,
C HOWDHURY, K . D AS, D RESSELHAUS, G ., and D RESSELHAUS, M . S., 1994, Chem.
Phys. L ett., 220, 186.
[78] K Uï RTI , J ., K UZMANY, H ., BURGER, B., H ULMAN, M ., W INTER, J ., and K RASSE, G .,
1999, Synth. Met., 103, 2508.
[79] SAITO, R ., T AKEYA, T ., K IMURA, T ., D RESSELHAUS, G ., and D RESSELHAUS, M . S.,
1999, Phys. Rev. B, 59, 2388.
Phonons in carbon nanotubes 811

[80] T AN, P .-H ., T ANG, Y ., D ENG, Y .-M ., L I , F ., W EI, Y . L ., and C HENG, H . M ., 1999,
Appl. Phys. L ett., 75, 1524.
[81] BANDOW, S., A SAKA, S., SAITO, Y ., R AO, A . M ., G RIGORIAN, L ., R ICHTER, E ., and
E KLUND, P . C ., 1998, Phys. Rev. L ett., 80, 3779.
[82] F ANG, S. L ., R AO, A . M ., E KLUND, P . C ., N IKOLAEV, P ., R INZLER, A . G ., and
SMALLEY, R . E ., 1998, J. mater. Res., 13, 2405.
[83] R AO, A . M ., BANDOW, S., R ICHTER, E ., and E KLUND, P . C ., 1998, T hin Solid Films,
331, 141.
[84] SAITO, Y ., T ANI, Y ., M IYAGAWA, N ., M ITSUSHIMA, K ., K ASUYA, A ., and N ISHINA,
Y ., 1998, Chem. Phys. L ett., 294, 593.
[85] SUGANO, M ., K ASUYA, A ., T OHJI, K ., SAITO, Y ., and N ISHINA, Y ., 1998, Chem. Phys.
L ett., 292, 575.
[86] K UZMANY, H ., BURGER, B., H ULMAN, M ., K URTI, J ., R INZLER, A . G ., and SMALLEY,
R . E ., 1998, Europhys. L ett., 44, 518.
[87] K ATAURA, H ., K IMURA, A ., O HTSUKA, Y ., SUZUKI, S., M ANIWA, Y ., H ANYU, T ., and
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

A CHIBA, Y ., 1998, Jpn. J. appl. Phys., 37, L616.


[88] C HENG, H . M ., L I, F ., SUN, X ., BROWN, S. D . M ., P IMENTA, M . A ., M ARUCCI, A .,
D RESSELHAUS, G ., and D RESSELHAUS, M . S., 1998, Chem. Phys. L ett., 289, 602.
[89] K ASUYA, A ., SUGANO, M ., M AEDA, T ., SAITO, Y ., T OHJI, K ., T AKAHASHI, H .,
SASAKI, Y ., F UKUSHIMA, M ., N ISHINA, Y ., and H ORIE, C ., 1998, Phys. Rev. B, 57,
4999.
[90] A NGLARET, E ., R OLS, S., and SAUVAJOL, J .-L ., 1998, Phys. Rev. L ett., 81, 4780.
[91] K UZMANY, H ., BURGER, B., T HESS, A ., and SMALLEY, R . E ., 1998, Carbon, 36, 709.
[92] I LIEV, M . N ., L ITVINCHUK, A . P ., A REP ALLI , S., N IKOLAEV, P ., and SCOTT, C . D .,
2000, Chem. Phys. L ett., 316, 217.
[93] K ASUYA, A ., SASAKI, Y ., SAITO, Y ., T OHJI, K ., and N ISHINA, Y ., 1997, Phys. Rev.
L ett., 78, 4434.
[94] P IMENTA, M . A ., M ARUCCI, A ., E MP EDOCLES, S., BAWENDI, M ., H ANLON, E . B.,
R AO, A . M ., E KLUND, P . C ., SMALLEY, R . E ., D RESSELHAUS, G ., and D RESSELHAUS,
M . S., 1998, Phys. Rev. B, 58, R16 016.
[95] P IMENTA, M . A ., M ARUCCI , A ., BROWN , S. D . M ., M ATTHEWS, M . J ., R AO, A . M .,
E KLUND, P . C., SMALLEY, R . E ., D RESSELHAUS, G ., and D RESSELHAUS, M . S., 1998,
J. mater. Res., 13, 2396.
[96] BROWN , S. D . M ., C ORIO, P ., M ARUCCI , A ., D RESSELHAUS, M . S., P IMENTA, M . A .,
and K NEIPP , K ., 2000, Phys. Rev. B, Rapid Commun. , 61, R5137.
[97] BARKER, A . S., and L OUDON, R ., 1972, Rev. mod. Phys., 44, 18.
[98] W HITE, C . T ., and T ODOROV, T . N ., 1998, Nature (London), 393, 240.
[99] C HARLIER, J .-C ., and L AMBIN , P H., 1998, Phys. Rev. B, 57, R15037.
[100] P ICHLER, T ., K NUPFER, M ., G OLDEN, M . S., F INK, J ., R INZLER, A ., and SMALLEY,
R . E ., 1998, Phys. Rev. L ett., 80, 4729.
[101] BROWN , S. D . M ., 2000, PhD thesis, Massachusetts Institute of Technology,
Department of Physics, USA.
[102] BROWN , S. D . M ., J ORIO, A ., C ORIO, P ., D RESSELHAUS, M . S., D RESSELHAUS, G ., and
K NEIPP , K ., submitted for publication.
[103] A NGLARET, E ., private communication.
[104] T AN, P .-H ., D ENG, Y .-M ., Z HAO, Q .-Z ., and C HENG, W .-C ., 1999, Appl. Phys. L ett.,
74, 1818.
[105] H UONG, P . V., C AVAGNAT, R ., A JAYAN, P . M ., and STEPHAN, O ., 1995, Phys. Rev. B,
51, 10 048.
[106] H UANG, F ., Y UE, K . T ., T AN , P ., Z HANG, S. L ., SHI, Z ., Z HOU, X ., and G U , Z ., 1998,
J. appl. Phys., 84, 4022.
[107] T AN, P .-H ., D ENG, Y .-M ., and Z HAO, Q ., 1998, Phys. Rev. B, 58, 5435.
[108] W ADA, N ., 1981, Phys. Rev. B, 24, 1065.
[109] M ENON, M ., and SRIVASTAVA, D ., 1997, Phys. Rev. L ett., 79, 4453.
[110] L IU, M ., and C OWLEY, J . M ., 1994, Ultramicroscopy, 53, 33.
[111] L IU, M ., and C OWLEY, J . M ., 1994, Carbon, 32, 393.
[112] SERAPHIN, S., Z HOU, D ., and J IAO, J ., 1994, Acta Microsc., 3, 45.
[113] C HARLIER, J .-C ., L AMBIN, P H., and E BBESEN , T . W ., 1996, Phys. Rev. B, 54, R8377.
812 M. S. Dresselhaus and P. C. Eklund

[114] BROWN, S. D . M ., CORIO, P ., M ARUCCI , A ., D RESSELHAUS, M . S., P IMENTA, M . A .,


and K NEIPP , K ., 2000, Phys. Rev. B, 61 R5137.
[115] C ORIO, P ., BROWN, S. D . M ., M ARUCCI, A ., P IMENTA, M . A ., K NEIPP , K .,
D RESSELHAUS, G ., and D RESSELHAUS, M . S., 2000, Phys. Rev. B, 61, 13202.
[116] D RESSELHAUS, M . S., D RESSELHAUS, G ., SUGIHARA, K ., SPAIN, I . L ., and G OLDBERG,
H . A ., 1998, Graphite Fibers and Filaments, Vol. 5, Springer Series in Materials Science
(Berlin: Springer-Verlag).
[117] M ATTHEWS, M . J ., P IMENTA, M . A ., D RESSELHAUS, G ., D RESSELHAUS, M . S., and
E NDO, M ., 1999, Phys. Rev. B, 59, R6585.
[118] P IMENTA, M . A ., H ANLON, E . B., M ARUCCI, A ., C ORIO, P ., BROWN, S. D . M .,
E MP EDOCLES, S., BAWENDI, M ., D RESSELHAUS, G ., and D RESSELHAUS, M . S., 2000,
Brazilian J. Phys. (in the press).
[119] J EANMAIRE, D . L ., and D UYNE, R . P . V., 1977, J. Electroanal. Chem., 84, 1.
[120] A LBRECHT , M . G ., and C REIGHTON, J . A ., 1977, J. Am. chem. Soc., 99, 5215.
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

[121] O TTO, A ., 1984, L ight Scattering in Solids IV . Electronic Scattering, Spin E€ ects,
SERS and Morphic E€ ects, edited by M. Cardona and G . G untherodt (Berlin:
Springer-Verlag).
[122] M OSKOVITS, M ., 1985, Rev. mod. Phys., 57, 783.
[123] K NEIPP , K ., W ANG, Y ., K NEIPP , H ., I TZKAN, I ., D ASARI, R . R ., and F ELD, M . S.,
1996, Phys. Rev. L ett., 76, 2444.
[124] K NEIPP , K ., W ANG, Y ., K NEIPP , H ., P ERELMAN, L . T ., I TZ KAN, I ., D ASARI, R . R .,
and F ELD, M . S., 1997, Phys. Rev. L ett., 78, 1667.
[125] N IE, S., and E MORY, S. R ., 1997, Science, 275, 1102.
[126] K NEIPP , K ., K NEIPP , H ., K ARTHA, V. B., M ANOHARAN, R ., D EINUM, G ., I TZKAN, I .,
D ASARI , R . R ., and F ELD , M . S., 1998, Phys. Rev. E, 57, R6281.
[127] K NEIPP , K ., K NEIPP , H ., M ANOHARAN, R ., H ANLON, E . B., I TZKAN, I ., D ASARI,
R . R ., and F ELD, M . S., 1998, Appl. Spectrosc., 527, 1493.
[128] M ARKEL, V. A ., SHALAEV, V. M ., Z HANG, P ., H UYNH, W ., T AY, L ., H ASLETT, T . L .,
and M OSKOVITS, M ., 1999, Phys. Rev. B, 59, 10 903.
[129] L EFRANT, S., BALTOG, I ., L AMY DE LA C HAPELLE, M ., BAIBARAC, M ., L OUARN, G .,
J OURNET, C ., and BERNIER, P ., 1999, Synthetic Met., 100, 13.
[130] K NEIPP , K ., K NEIPP , H ., C ORIO, P ., BROWN, S. D . M ., SHAFER, K ., M OTZ , J .,
P ERELMAN , L . T ., H ANLON, E . B., M ARUCCI, A ., D RESSELHAUS, G ., and
D RESSELHAUS, M . S., 2000, Phys. Rev. L ett., 84, 3470.
[131] W EBER , A ., 1979, Raman Spectroscopy in Gases and L iquids, Vol. 11, Series in T opics
in Current Physics (Heidelberg, New York: Springer-Verlag).
[132] STOCKMAN , M . I ., SHALAEV, V. M ., M OSKOVITS, M ., BOTET, R ., and G EORGE, T . F .,
1992, Phys. Rev. B, 46, 2821.
[133] D UESBERG , G . S., BLAU , W . J ., BYRNE, H . J ., M USTER, J ., BURGHARD, M ., and
R OTH, S., 1999, Chem. Phys. L ett., 310, 8.
[134] SHALAEV, V. M ., and STOCKMAN , M . I ., 1987, Sov. Phys. JET P, 65, 287.
[135] R OJAS, R ., and C LARO, F ., 1993, J. chem. Phys., 98, 998.
[136] R AO, A . M ., J ORIO, A ., P IMENTA, M . A ., D ANTAS, M . S. S., SAITO, R ., D RESSELHAUS,
G ., and D RESSELHAUS, M . S., 2000, Phys. Rev. L ett., 84, 1820.
[137] SUN, H . D ., T ANG, Z . K ., C HEN, J ., and L I , G ., 1999, Solid State Commun. , 109, 365.
[138] SAITO, R ., D RESSELHAUS, G ., and D RESSELHAUS, M . S., 1999, Science and T echnology
of Carbon Nanotubes, edited by K. Tanaka, T. Yamabe and K. Fukui (Oxford:
Elsevier Science Ltd), p. 51.
[139] A NDREWS, R ., J ACQUES, D ., R AO, A . M ., D ERBYSHIRE, F ., Q IAN, D ., F AN , X .,
D ICKEY, E . C ., and C HEN, J ., 1999, Chem. Phys. L ett., 303, 467.
[140] A JIKI, H ., and A NDO, T ., 1994, Physica B, Condens. Matter, 201, 349.
[141] A JIKI, H ., and A NDO, T ., 1995, Jpn. J. appl. Phys., Suppl. 34-1, 107.
[142] VIDANO, R . P ., F ISHBACH, D . B., W ILLIS, L . J ., and L OEHR, T . M ., 1981, Solid State
Commun., 39, 341.
[143] M ERNAGH, T . P ., C OONEY, R . P ., and J OHNSON, R . A ., 1984, Carbon, 22, 39.
[144] R AMSTEINER, M ., and W AGNER, J ., 1987, Appl. Phys. L ett., 51, 1355.
[145] W ANG, Y ., A LSMEYER , D . C ., and M C C REERY, R . L ., 1990, Chem. Mater., 2, 557.
Phonons in carbon nanotubes 813

[146] BARANOV, A . V., BEKHTEREV, A . N ., BOBOVICH, Y . S., and P ETROV, V. I ., 1987, Opt.
Spectrosc. USSR, 62, 612.
[147] P OCSIK, I ., H UNDHAUSEN, M ., K OOS, M ., and L EY, L ., 1998, J. non-cryst. Solids, 227±
230 B, 1083.
[148] M ARCUS, B., F AYETTE, L ., M ERMOUX, M ., A BELLO, L ., and L UCAZEAU , G ., 1994,
J. appl. Phys., 76, 3463.
[149] K ASTNER, J ., W INTER, J ., and K UZMANY, H ., 1995, Mater. Sci. Forum, 191, 161.
[150] D RESSELHAUS, M . S., P IMENTA, M . A ., K NEIPP , K ., BROWN, S. D . M ., C ORIO, P .,
M ARUCCI, A ., and D RESSELHAUS, G ., 2000, Science and Applications of Nanotubes,
edited by D. Toma nek, and R. J. Enbody (New York: Kluwer Academic) ; 2000,
Proceedings of the International W orkshop on the Science and Applications of
Nanotubes, Michigan State University, East Lansing, MI, USA, 24± 27 July, p. 253.
[151] K ELLY, B. T ., 1981, Physics of Graphite (London: Applied Science Publishers).
[152] D ONG, Z . H ., Z HOU, P ., H OLDEN, J . M ., E KLUND, P . C ., D RESSELHAUS, M . S., and
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

D RESSELHAUS, G ., 1993, Phys. Rev. B, 48, 2862.


[153] W ANG, K . A ., R AO, A . M ., E KLUND, P . C ., D RESSELHAUS, M . S., and D RESSELHAUS,
G ., 1993, Phys. Rev. B, 48, 11375.
[154] D RESSELHAUS, M . S., E KLUND, P . C ., and P IMENTA, M . A ., 2000, Raman Scattering in
Materials Science, edited by W. Weber and R. Merlin (Berlin: Springer-Verlag), in
press.
[155] G RIGORIAN, L ., W ILLIAMS, K . A ., F ANG, S., SUMANASEKERA, G . U ., L OPER , A . L .,
D ICKEY, E . C ., P ENNYCOOK, S. J ., and E KLUND, P . C ., 1998, Phys. Rev. L ett., 80,
5560.
[156] R AO, A . M ., BANDOW, S., R ICHTER, E ., and E KLUND, P . C ., 1998, T hin Solid Films,
331, 141.
[157] G RIGORIAN, L ., SUMANASEKERA, G . U ., L OPER, A . L ., F ANG, S., A LLEN, J . L ., and
E KLUND, P . C ., 1998, Phys. Rev. B, 58, R4195.
[158] G AO, B., K LEINHAMMES, A ., T ANG, X . P ., BOWER, C ., F LEMING, L ., W U, Y ., and
Z HOU, O ., 1999, Chem. Phys. L ett., 307, 153.
[159] SUMANASEKERA, G . U ., A LLEN, J . L ., F ANG, S., L OPER, A . L ., R AO, A . M ., and
E KLUND, P . C ., 1999, J. phys. Chem., 103, 4292.
[160] D RESSELHAUS, M . S., and D RESSELHAUS, G ., 1981, Adv. Phys., 30, 139.
[161] R AO, A . M ., E KLUND, P . C ., BANDOW, S., T HESS, A ., and SMALLEY, R . E ., 1997,
Nature (London), 388, 257.
[162] D RESSELHAUS, M . S., and D RESSELHAUS, G ., 1982, L ight Scattering in Solids III,
Vol. 51, T opics in Applied Physics, edited by M. Cardona and G. GuÈntherodt (Berlin:
Springer-Verlag), p. 3.
[163] E KLUND, P . C ., and D OLL, G . L ., 1992, Graphite Intercalation Compounds II:
T ransport and Electronic Properties, Vol. 18, Springer Series in Materials Science,
edited by H. Zabel and S. A. Solin (Berlin: Springer-Verlag), p. 105.
[164] M IZEL , A ., BENEDICT, L . X ., C OHEN, M . L ., L OUIE, S. G ., Z ETTL , A ., BUDRA, N . K .,
and BEYERMANN, W . P ., 1999, Phys. Rev. B, 60, 3264.
[165] N ICKLOW , R ., W AKABAYASHI, N ., and SMITH , H . G ., 1972, Phys. Rev., 5, 4951.
[166] E KLUND, P . C ., H OLDEN, J . M ., and J ISHI, R . A ., 1995, Carbon, 33, 959.
[167] R OLS, S., A NGLARET, E ., SAUVAJOL, J .-L ., C ODDENS, G ., and D IANOUX, A . J ., 1999,
Appl. Phys. A, 81, 4780 ; R OLS, S., 2000, Physica B, 276, 276.
[168] Y I, W ., L U , L ., D IAN-LIN , Z HANG, P AN , Z . W ., and X IE, S. S., 1999, Phys. Rev. B,
Rapid Commun. , 59, R9015.
[169] L IN, M . F ., and SHUNG, K . W .-K ., 1996, Phys. Rev. B, 54, 2896.
[170] H ONE, J ., W HITNEY, M ., P ISKOTI, C ., and Z ETTL , A ., 1999, Phys. Rev. B, 59,
R2514.
[171] H EREMANS, J ., 1985, Carbon, 23, 431.
[172] H EREMANS, J ., and BEETZ J R, C . P ., 1985, Phys. Rev. B, 32, 1981.
[173] H ONE, J ., E LLWOOD, I ., M UNO, M ., M IZEL, A RI, C OHEN, M ARVIN L ., Z ETTL , A .,
R INZLER , A NDREW G ., and SMALLEY, R . E ., 1998, Phys. Rev. L ett., 80, 1042.
[174] P EIERLS, R ., 1955, Quantum Theory of Solids (Oxford: Oxford University Press).
[175] L IN, M . F ., CHUU, D . S., and SHUNG, K . W .-K ., 1996, Phys. Rev. B, 53, 11186.
814 Phonons in carbon nanotubes

[176] A SHCROFT, N . W ., and M ERMIN, N . D ., 1976, Solid State Physics (New York: Holt,
Rinehart and Winston).
[177] SUMANASEKERA, G . U ., ADU, C ., F ANG, S., and E KLUND, P . C ., 2000, Phys. Rev.
L ett., in press.
[178] G RIGORIAN, L ., SUMANASEKERA, G . U ., L OPER, A . L ., F ANG, S., A LLEN, J . L ., and
E KLUND, P . C ., 1999, Phys. Rev. B, 60, R11 309.
Downloaded By: [USC University of Southern California] At: 00:52 11 November 2009

Вам также может понравиться