Letter

pubs.acs.org/NanoLett

A Low-Frequency Wave Motion Mechanism Enables Ecient Energy

Transport in Carbon Nanotubes at High Heat Fluxes
Xiaoliang Zhang,, Ming Hu,*, and Dimos Poulikakos,

Laboratory of Thermodynamics in Emerging Technologies, Department of Mechanical and Process Engineering, ETH Zurich, 8092
Zurich, Switzerland
ABSTRACT: The great majority of investigations of thermal transport in carbon
nanotubes (CNTs) in the open literature focus on low heat uxes, that is, in the regime
of validity of the Fourier heat conduction law. In this paper, by performing
nonequilibrium molecular dynamics simulations we investigated thermal transport in a
single-walled CNT bridging two Si slabs under constant high heat ux. An anomalous
wave-like kinetic energy prole was observed, and a previously unexplored, wavedominated energy transport mechanism is identied for high heat uxes in CNTs,
originated from excited low frequency transverse acoustic waves. The transported energy,
in terms of a one-dimensional low frequency mechanical wave, is quantied as a function
of the total heat ux applied and is compared to the energy transported by traditional
Fourier heat conduction. The results show that the low frequency wave actually
overtakes traditional Fourier heat conduction and eciently transports the energy at high
heat ux. Our ndings reveal an important new mechanism for high heat ux energy
transport in low-dimensional nanostructures, such as one-dimensional (1-D) nanotubes
and nanowires, which could be very relevant to high heat ux dissipation such as in micro/nanoelectronics applications.
KEYWORDS: Low-frequency wave, high heat ux, energy transport, carbon nanotube, molecular dynamics

ciency energy use is intertwined with many aspects of

the ecient performance of a plethora of technological
applications. Heat dissipation limits the performance of
electronics from hand-held devices (103 W) to massive
data centers (109 W), all primarily based on silicon micro/
nanotechnology.1 On the other hand, carbon nanotubes
(CNTs) have attracted signicant attention in recent years
for thermal management applications utilizing their favorable
properties, such as high thermal and electronic conductivities,
stemming from their unique atomic structure. So far, a
considerable amount of work has been dedicated to energy
(electrical and thermal) transport in CNTs, either measured
experimentally25 or predicted from theory,610 where electric
elds accelerate charge carriers2 (electrons or holes) or a
temperature gradient drives heat ow through phonon
transport.35 In addition, both suspended and substrate
supported nanostructures were investigated to understand the
mechanism of energy transport in such one-dimensional
devices.3,9,10
However, to date almost all of the existing investigations of
thermal transport of CNTs have focused on low heat ux; that
is, they investigate the regime where the Fourier heat
conduction law is valid. As transistors are continuously further
miniaturized and more heat needs to be dissipated, undesired
high heat ux and local hot spot regions always occur. Actually,
results from earlier investigations indicate that non-Fourier heat
conduction can be activated by fast heating.11,12 Recently
Maruyama and Shiomi11 attempted to study the nonstationary
heat conduction in a CNT by heating the tube with a short
pulse in molecular dynamics (MD) simulations, but the so-

2012 American Chemical Society

called non-Fourier conduction could not be maintained due to

the short pulse used. What remains unclear, however, is how
energy is steadily transported or dissipated in nanostructures at
high heat uxes and in what conditions non-Fourier conduction
will occur. In this paper we aim to study the mechanism of
energy transport in CNTs by applying a constant high heat ux.
To this end, we identify herein a new energy transport
mechanism in low-dimensional nanostructures, which is
responsible for ecient energy transport at high heat uxes.
Our basic model system consists of an open-ended singlewalled CNT cantilevered out of a (001) surface of bulk Si, as
shown in the top panel of Figure 1. The term open-ended
single-walled CNT means that the CNT ends are open, not
closed, and contact the Si slabs. The structure is contained in a
rectangular simulation box with a cross section of 43.44 43.44
2 (8 8 unit cells of Si with a lattice constant of 5.43 ). We
simulated both (5,5) and (10,10) armchair CNTs. The Si
crystal is composed of 8192 atoms, and the CNT contains 32
000 and 64 000 carbon atoms for (5,5) and (10,10)
congurations, respectively. Periodic boundary conditions are
used in all directions, thus resulting in two Si-CNT interfaces
normal to the z-direction (see the top panel of Figure 1). In all
MD simulations performed herein, the adaptive intermolecular
reactive empirical bond order (AIREBO) potential13 was used
to describe the CC interactions in the CNT. This potential
Received: January 20, 2012
Revised: May 30, 2012
Published: June 18, 2012
3410

dx.doi.org/10.1021/nl300261r | Nano Lett. 2012, 12, 34103416

Nano Letters

Letter

Figure 1. (top) Schematic of a (5,5) single-walled CNT with 1600 unit cells (400 nm long) connected to Si slabs of 8 8 16 unit cells. Color
coding: cyan, C; yellow, Si. (bottom) Corresponding total kinetic energy, external energy, and real temperature distribution for the CNT. The real
temperature is calculated by only using the internal energy. The total kinetic energy and the external energy are normalized by 3/2NkB to facilitate
comparison with the real temperature. See denitions of these energies in the text.

actual reported data represent the steady state averages over the
last 0.5 ns of each production run.
A typical temperature prole of a (5,5) CNT under the high
heat ux of 464 GW/m2 is presented in the bottom panel of
Figure 1 (the range of heat ux values that can be termed
high will be discussed later in the paper). For the CNT crosssection area, we follow the standard practice of treating the
single-walled CNT as a hollow tube of diameter d and a wall
thickness of h = 3.4 .8,26 The diameter values are d = 6.9
and 13.8 for (5,5) and (10,10) CNT, respectively. This gives
the cross-section area of dh. Surprisingly, an anomalous wavelike overall temperature prole is observed (see black line in
Figure 1). Strictly speaking, this is the total kinetic energy
prole and not the real temperature prole of the CNT, as we
will explain later. We studied a video of the CNT motion in
VMD software27 and found that the CNT vibrates signicantly
in this case. The same wave-like overall temperature prole was
found in the (10,10) CNT for heat ux around 500 GW/m2.
We rst excluded the artifacts possibly induced by the Muller
Plathe method employed and conrmed our results by reducing
the heat ux to 16 and 122 GW/m2 for (5,5) and (10,10)
CNT, respectively. The simulation shows that as heat ux is
reduced no noticeable CNT vibration occurs and the overall
temperature prole of the CNT is practically linear on both
sides of the heat source region with two well-dened constant
temperature gradients (not shown for brevity), from which we
obtained the thermal conductivity of the (5,5) and (10,10)
CNTs to be 196 4 and 229 6 W/mK, respectively, in good
agreement with previous studies.2830 This underpins our
ndings of the wave-like overall temperature prole occurring
only at high heat uxes.
We analyze the anomalous temperature prole starting with
the velocities of atoms. We consider the CNT atoms in each
slice as an ensemble. For each slice we decompose the total
velocity of each CNT atom (tot) in the slice into a translational

has been widely used by other researchers to calculate thermal

transport properties of carbon nanomaterials.1416 The Terso
potential17 was used to model SiSi and SiC interatomic
interactions. This potential has been successfully used to
simulate the thermal conductivity of Si18,19 and Si-CNT20
systems.
All MD calculations were performed using the LAMMPS21
package. In the rst stage of the MD simulations, we relaxed the
system at constant atmospheric pressure and T = 300 K for 1 ns
with time step 1.0 fs, using a NoseHoover thermostat and
barostat.22,23 After NPT (constant pressure and temperature)
relaxation, we continued to relax the system with NVT
(constant volume and temperature) ensemble for 0.5 ns,
using a Langevin thermostat and NVE (constant volume
without thermostat) ensemble for additional 0.5 ns. Following
equilibration, we applied a constant heat ux to the system
using the MullerPlathe method.24,25 The heat source and heat
sink was placed at the center of the CNT and Si slab,
respectively, as illustrated in the top panel of Figure 1. It is
worth pointing out that, for the cases of very high heat ux, a
reduced time step of 0.1 fs was used to ensure energy
conservation within a precision of 105 in relative energy
change in the course of heat-source heat-sink (NVE) run.
Once a steady state is reached, a temperature prole is
established perpendicular to the Si slab. The temperature
prole is obtained by dividing the simulation domain into slices
of 9.75 thickness along the z-direction, that is, along the heat
current direction. Each slice contains approximately 82 atoms
for the (5,5) CNT and 164 atoms for the (10,10) CNT. The
local temperature (we will discuss later that this is not the real
temperature) of each slice was calculated from its kinetic
energy. We average the temperature proles over 60 ps and
obtained a series of about 70 temperature proles for all of the
production runs of typically 4 ns. After 0.52 ns (depending on
the diameter of CNT and the heat ux applied) we observe that
the temperature prole does not change systematically. The
3411

dx.doi.org/10.1021/nl300261r | Nano Lett. 2012, 12, 34103416

Nano Letters

Letter

velocity (t r) of the slice plus a random atomic motion velocity

(ran):31
tot, = tr , + ran,
(1a)
tr , =

1
N

N
i
tot,

(1b)

i=1

where denotes the direction (x, y, or z), i a specic atom, and

N the total number of atoms in the slice. The subscripts tot,
tr, and ran denote the total, translational, and random
velocities, respectively. We dene the total kinetic energy (Etot
k ),
the external mechanical energy (Eext), and the internal thermal
energy (Eint, from which a real temperature can be dened) as
follows:
N

Ektot =

i=1
N

Eext =

E int =

i=1

1
i
2
mi(tot,
)
2

(2a)

1
mi(tr i , )2
2

(2b)

3
NkBT =
2

i=1

1
i
2
mi(ran,
)
2

(2c)

where kB is the Boltzmann's constant. By normalizing the total

kinetic energy and the external energy by (3/2)NkB, we
compare the energy distribution with the real temperature for
the CNT in Figure 1. It is clearly seen that, after subtracting the
translational velocities, the wave-like total nominal temperature
prole approaches the real temperature prole of the vibrating
CNT with two well-dened constant temperature gradients at
both sides of the heat source region, while the remaining
temperature prole is a regular sinusoidal-like curve, which has
the character of a transverse wave.
We further calculated the frequency of the transverse wave
present in the CNT performing a Fourier transform of the xand y-translational displacement (Figure 2a) of a selected CNT
ring (about 1 unit cell long). The results of wave amplitude
versus vibrational frequency are shown in Figure 2b. The low
frequency around 11.44 GHz was clearly identied. We also
conrmed the results of such low frequency by calculating the
vibrational density of states (VDOS) of the CNT. The VDOS
are calculated by a Fourier transform of the translational
velocity (t r) autocorrelation function, and the result is shown
in Figure 2c. A sharp peak at exactly the same low frequency
was observed (11.44 GHz). This frequency is far below the
vibrational frequencies due to random atom motion, which can
reach as high as 50 THz; see the inset of Figure 2c. The left and
right (relative to the source in the center) parts of the CNT
have the same vibrational frequency, but the energy transported
by the mechanical wave travels from the center to the two ends
of the CNT in two opposite directions. It is worth pointing out
that such low-frequency vibration has been observed in recent
experiments for dierent types of low-dimensional nanostructures, for example, (10,10) single-walled CNT32 and twodimensional graphene.33 None of these studies focused on the
investigation of the thermal transport nor quantied the role of
the low-frequency motion in energy transport. We would also
like to emphasize that, due to the excited low-frequency modes,
the temperature measured experimentally is apparent and the
real temperature distribution can only be obtained after the
translational movement is subtracted.

Figure 2. (a) Time evolution of x- and y-translational displacements of

a selected layer in (5,5) CNT; (b) corresponding amplitude of the low
frequency wave obtained from Fourier transform of a; (c) vibrational
density of states of CNT (low-frequency part) obtained from Fourier
transform of the translational velocity autocorrelation function. Inset:
full spectrum of vibrational density of states obtained from random
thermal velocities.

Next, we calculate how much thermal power is transported

by the mechanical wave and by Fourier conduction,
respectively, by the following formulas34
Ptot = PW + PF
3412

(3a)
dx.doi.org/10.1021/nl300261r | Nano Lett. 2012, 12, 34103416

Nano Letters
PW =

1 2 2
A Vg
2

T
PF = 2 Ac
z

Letter

software.35 The results are shown in Figure 3. Then, the

phonon group velocity was obtained by

(3b)

Vg =
(3c)

(4)

where k is the wave vector. We found that the transverse

acoustic (TA) phonon mode is much stronger than the
longitudinal acoustic (LA) phonon mode, according to the
wave amplitude and the VDOS of the CNT in three directions,
so we only consider the wave energy in x- and y-directions, that
is, only the transverse acoustic wave. For the case shown in
Figures 1 and 2, the input heat power was Ptot = 0.34 W,
which translates into a total heat ux of 464 GW/m2, and PW is
calculated to be 0.20 W, meaning that 59% of the total energy
is transported by the mechanical wave. From eq 3c we obtained
the thermal conductivity of CNT to be 231 W/mK, which is in
good agreement with the nonvibrational case.
In Figure 4a we plot the frequency of the excited transverse
acoustic wave in a CNT as a function of the total heat ux
applied. It is clearly shown that for both (5,5) and (10,10)
CNTs the frequency increases linearly with heat ux. The same
trend was found in a previous study on electrically biased 2-D
graphene.33 We also noticed that the (10,10) CNT frequency
increases faster with increasing heat ux than the (5,5) CNT,
possibly because the (10,10) CNT is more exible in the
transverse modes. The percentage of the transported energy by
the wave motion versus total heat ux is presented in Figure 4b.
At low heat uxes, the wave energy is negligible, and the heat is
entirely conducted according to Fouriers law. As the heat ux
increases, the percentage of the transported energy by the wave
motion steeply increases since the wave energy is proportional
to 2 according to eq 3b, indicating that the low-frequency
wave becomes dominant in energy transport. For example, the
percentage of wave-transported energy becomes as high as 59%
(total heat ux 464 GW/m2) and 69% of the total energy (total
heat ux 656 GW/m2) for (5,5) and (10,10) CNTs,
respectively. Note that the wave-transported energy does not
increase in a power fashion because the slope of the phonon
group velocity versus frequency gradually decreases, as shown
in Figure 3c,d. For a (5,5) 400 nm long CNT, we rerun the
simulations using the reactive empirical bond order (REBO)
potential36 with a heat ux 412 GW/m2. The same behavior
was observed, and the percentage of wave-transported energy is
obtained to be 40.0%, which is in good agreement with that
using the AIREBO potential.
To prove that our ndings are not dependent on the
MullerPlathe method with which the high heat ux is applied,
for a selected test case, that is, (5,5) 400-nm-long CNT, we run
two additional simulations with dierent high ux methods: (1)
two thermostats and (2) constant heat ux. For the thermostat
method, a slice (thickness: 9.94 ) at the center of the CNT
and an identical slice at the Si walls at its ends serve as the heat
source and heat sink, respectively. The thermostat to the heat
source/sink is controlled by the Langevin method, and the
temperatures of the heat source and sink are maintained at 460
K and 139 K, respectively. These are exactly the same
temperature values as for the time-averaged steady state
temperature in the MullerPlathe method, for the sake of a
direct comparison under practically the same heat ux. Note
that in this case both the total energy of the system and the
heat ux are uctuating around a constant value. We also run
the simulation with yet another method. In this case the energy
entering the system equals the energy exiting the system. We

where Ptot is the total input heat power into the CNT,
determined by the input parameter of the MullerPlathe
method, PW and PF the heat power transported by the
mechanical wave and by Fourier conduction, respectively, the
mass of the CNT per unit length, the angular frequency, A
the wave amplitude, Vg the phonon group velocity, the
thermal conductivity of the CNT, T/z the real temperature
gradient along the tube axis, and Ac the cross-sectional area of
the CNT. The factor of 2 in eq 3c accounts for the Fourier heat
ow in the left and right directions. To obtain the phonon
group velocity Vg, we rst calculated the phonon dispersion
curves of the (5,5) and (10,10) CNTs using the PHONOPY

Figure 3. (top) Phonon dispersion curves of (5,5) (a) and (10,10) (b)
CNT. a is the lattice constant of the CNT in the axial direction, and k
is the wave vector. The thick red lines identify the two degenerated
transverse acoustic (TA) phonon modes. (bottom) Corresponding
phonon group velocity (low frequency part) of (5,5) (c) and (10,10)
(d) CNT as a function of frequency. Insets: full phonon group
velocity.
3413

dx.doi.org/10.1021/nl300261r | Nano Lett. 2012, 12, 34103416

Nano Letters

Letter

heat ux method and the MullerPlathe method is that: (a)

With the constant heat ux method the heat ux determined by
the input parameters is strictly the same throughout an entire
simulation, while with the MullerPlathe method the heat ux
may vary (uctuate) a bit but will nally stabilize. With the
MullerPlathe method, one can tune the heat ux and the
temperature gradients, adjusting the time lag between velocity
swapping, and have an exact estimate of the exchanged energy.
Yet the stationary ux is an average quantity and is determined
a posteriori by the response of the system. (b) The constant
heat ux method adds/subtracts energy to all of the atoms in
the heat source/sink at each time step, while the MullerPlathe
method exchanges continuously the velocity of a pair of atoms
(the coldest in the hot region and the hottest in the cold
region), thus maybe aecting the MaxwellBoltzmann velocity
distribution in these regions. The results using the above two
additional high ux methods are included in Figure 4b. It is
clearly shown that the low-frequency waves are present in both
additional methods and that the percentage of wave-transported energy is in good agreement with that of the Muller
Plathe method under a very similar heat ux (375411 GW/
m2). This comparison provides solid evidence that the
mechanism of low-frequency waves facilitating thermal transport in a CNT is not an artifact of the MullerPlathe method.
Figure 4b provides direct evidence that the low-frequency
wave occurs at considerably high heat uxes and is the major
mechanism responsible for the high heat ux energy transport
in CNTs, as well as that the contribution from traditional
Fourier conduction is reduced as the heat ux increases. Figure
4b also shows that there is a critical heat ux beyond which the
CNT starts to vibrate signicantly in low frequency; thus, the
low frequency mechanical wave dominates the energy transport. To this end, heat uxes termed as high in this paper
have values above the critical heat ux limit. Using the phonon
gas model presented in a previous work37 the critical heat ux
can be written as
qc =

22 C 3T0 3

(5)

where is the Gruneisen constant, the mass density of the

CNT, C the specic heat, and T0 the average temperature of the
CNT. Employing the following parameter values: = 2236 kg/
m3 (to be consistent with our heat ux denition, dh length
is used to calculate the volume), C = 600 J/kgK,38 and T0 =
300 K, we obtained the critical heat ux to be qc = 241 GW/m2,
as indicated by the dashed line in Figure 4b. The critical heat
ux determined in our MD simulations is in good agreement
with previous theoretical predictions.37
The observed phenomenon obeys the second law of
thermodynamics. A violation of the second law of thermodynamics for our system would mean eciency in converting the
added thermal energy to mechanical energy is higher than the
corresponding Carnot eciency. In our problem of course no
mechanical energy is removed out of the system. It stays in, and
it is removed nally as heat also from the cold end of the CNT.
We checked this issue in detail. If one considers the total
mechanical energy of a CNT, that is, the vibrational motion of
low-frequency waves produced by the applied input heat,
according to the second law of thermodynamics the following
must hold:

Figure 4. Frequency of the excited transverse acoustic wave in a 400nm-long CNT (a), percentage of wave-transported energy (b), and
thermal conductivity of CNT (c) as a function of the total heat ux. In
b, : the eciency of exciting the mechanical wave; : the maximum
eciency for a Carnot cycle; : two-thermostat method; : constant
heat ux method; : MullerPlathe method with REBO potential;
the black dashed line denotes the critical heat ux (qc) beyond which
CNT starts to vibrate signicantly; thus, the low frequency mechanical
wave dominates the energy transport.

call it the constant heat ux method. This method adds

nontranslational kinetic energy (heat) to a group of atoms such
that their aggregate momentum is conserved. The same heat
source and heat sink were used as in the two-thermostat
method mentioned above. The dierence between this constant

wave
3414

Eext|wave
T
max
1 C Carnot
Q
TH

(6)

dx.doi.org/10.1021/nl300261r | Nano Lett. 2012, 12, 34103416

Nano Letters

Letter

where wave is the eciency of exciting the mechanical energy

wave, Eext|wave is the mechanical energy of the CNT with the
excited low frequency waves, Q is the total energy added to the
heat source, and TC and TH are the temperatures of heat source
and heat sink, respectively. To calculate the left-hand-side of
the above equation, we consider the time period in which the
mechanical wave travels from the heat source, that is, the center
of CNT, to the cold end of the CNT (the Si/CNT interface).
Then Q = Lz/2/VgPtot, where Lz is the total length of the CNT,
Vg is the group velocity of the mechanical wave, and Ptot is the
total input heat power into the CNT. Eext|wave is explicitly
calculated in the molecular dynamics simulations by adding the
translational energy of all CNT atoms together. The results of
wave and max
Carnot are reported in Figure 4b for the (5,5) CNT for
selected heat uxes. The results clearly show that the eciency
of exciting the mechanical wave with heat is smaller than the
Carnot eciency and proves that the observed phenomenon
does not violate the second law of thermodynamics. It is worth
pointing out that wave is higher than the percentage of wavetransported energy calculated by eq 3. This is because in eq 3
we only consider the translational energy contribution from a
single frequency which is the dominant peak in the VDOS
shown in Figure 2c, while in calculating wave we include the
contributions from all frequencies.
We also calculated the thermal conductivity of CNTs from
eq 3c for dierent heat uxes. The results are presented in
Figure 4c. The thermal conductivity stabilizes around 214 W/
mK and 239 W/mK for (5,5) and (10,10) CNT, respectively.
Again, these values are in good agreement with previous
studies.2830 The heat ux independent conductivity indicates
that Fourier conduction and the excited low frequency wave are
uncoupled and transport energy independently even for very
high heat uxes.
It is worth pointing out that the current study is
fundamentally dierent from the previous work by Wang et
al.37 who analyzed CNTs electrically heated by high-bias
current ows using the phonon gas model. They claimed that
the heat ow will be choked and temperature jumps will be
observed at the tube ends when the phonon gas velocity
reaches the thermal speed. Our atomistic simulation results
provide a completely dierent, new scenario: as the heat ux
increases the heat ow will not be choked. Instead, lowfrequency (long wavelength) waves will be excited in the CNT,
and through them the CNT becomes an energy conduit since
the mechanical wave can transfer the energy more eciently
than the traditional Fourier conduction. We also checked the
temperature jump at the Si-CNT interfaces and found that the
temperature jump does not increase as fast as the heat ux; thus
the interfacial thermal conductance (heat ux divided by
temperature jump) increases steeply, indicating that the SiCNT interface becomes more energy conductive as well.
In addition, we studied the length dependence of the low
frequency wave in the CNT under same total heat ux. Figure 5
shows the result for a (10,10) CNT for the heat ux of 500
GW/m2. It can be seen that the contribution of wave
transported energy decreases as the CNT becomes longer
and there is a threshold length beyond which the low-frequency
mechanical waves are hardly excited at this heat ux. Hence,
even higher heat uxes are required to trigger the vibration of
longer CNTs. This can be understood in terms of the higher
thermal conductivity of longer CNTs transporting the energy
eectively by Fourier heat conduction. It is worth pointing out
that, for the cases with same heat uxes shown in Figure 5, the

Figure 5. Percentage of wave-transported energy as a function of

(10,10) CNT length under the same total heat ux of 500 GW/m2.

frequency is found to decrease with increasing CNT length.

Generally speaking, when Fourier heat conduction cannot
transport the applied energy eectively, a lower frequency wave
will be excited rst and then a higher frequency wave, because
the transported energy by the wave is proportional to square of
the frequency. Thus a higher frequency wave can transport
more energy.
With reference to wave behavior vs energy transport, a
comment regarding possible soliton-like behavior39,40 of the
observed waves is appropriate. We see no apparent reason
pointing in the direction that the low-frequency waves
identied in this work has the form of a soliton. A brief
examination (not shown for brevity) yielded that the lowfrequency mechanical wave excited in the CNT is described
well by a second-order linear wave equation and not by the
nonlinear soliton equation.41 It was found that the CNT
vibration is exactly the same as the transverse vibration of a
string, which is dierent than that of a soliton where the
disturbance at any point in the domain disappears after the
wave passes through. A detailed analysis of this issue is outside
the scope of our work.
Before closing, we would like to emphasize that in this paper
we applied the heat source at the center of the CNT. We also
run an additional case with same heat ux but with reversed
heat current direction; that is, the heat sink was placed at the
center of the CNT. In the latter case, we did not nd as
signicant low-frequency waves as found in the case with the
heat source at the CNT center. This is understandable
considering that, when the CNT center serves as the heat
sink, the temperature of the entire CNT will be signicantly
lower. At this lower temperature, the CNT has high thermal
conductivity, which is sucient to facilitate the thermal energy
transport. Hence, in this case it is not necessary (there is no
reason) for the low-frequency wave to be excited.
In summary, nonequilibrium molecular dynamics simulations
were performed to investigate thermal transport in a singlewalled CNT bridging two Si slabs under a constant high heat
ux. An anomalous wave-like nominal temperature prole not
reported before is observed, and a new wave-dominated energy
transport mechanism is identied for high heat uxes in CNTs,
due to excited low-frequency transverse acoustic waves. By
decomposing the atomic velocity into a collective translational
3415

dx.doi.org/10.1021/nl300261r | Nano Lett. 2012, 12, 34103416

Nano Letters

Letter

component and a random thermal uctuation component, the

transported energy in terms of a one-dimensional transverse
acoustic wave is quantied as a function of the total heat ux
applied. The atomistic simulation results suggest that at low
heat uxes Fourier heat conduction is dominant and the
contribution of low frequency waves is negligible. However, as
the heat ux exceeds a critical value, low-frequency waves are
excited, and the wave transport energy mechanism overtakes
the traditional Fourier conduction, rendering the CNT
signicantly more energy conductive. Moreover, the critical
heat ux determined in our atomistic simulations is in good
agreement with predictions of the phonon gas model. We also
found that under same total heat ux the contribution of wave
transported energy decreases with CNT length and the low
frequency mechanical waves are hardly excited for long CNT.
Higher heat uxes are needed for their activation with
increasing length. Our ndings put forth a previously
unexplored mechanism for high heat ux transport in lowdimensional nanostructures, such as 1-D nanotubes and
nanowires, which could be relevant to energy transport and
high heat ux dissipation in applications such as micro/
nanoelectronics.

(13) Stuart, S. J.; Tutein, A. B.; Harrison, J. A. J. Chem. Phys. 2000,

112, 6472.
(14) Grujicic, M.; Cao, G.; Roy, W. N. J. Mater. Sci. 2005, 40, 1943.
(15) Ong, Z. Y.; Pop, E. Phys. Rev. B 2010, 81, 155408.
(16) Varshney, V.; Patnaik, S. S.; Roy, A. K.; Froudakis, G.; Farmer,
B. L. ACS Nano 2010, 4, 1153.
(17) Terso, J. Phys. Rev. B 1989, 39, 5566.
(18) Lee, J.-H.; Grossman, J. C.; Reed, J.; Galli, G. Appl. Phys. Lett.
2007, 91, 223110.
(19) Donadio, D.; Galli, G. Nano Lett. 2010, 10, 847.
(20) Diao, J.; Srivastava, D.; Menon, M. J. Chem. Phys. 2008, 128,
164708.
(21) Plimpton, S. J. Comput. Phys. 1995, 117, 1.
(22) Nose, S. J. Chem. Phys. 1984, 81, 511.
(23) Hoover, W. G. Phys. Rev. A 1985, 31, 1695.
(24) Muller-Plathe, F.; Reith, D. Comput. Theor. Polym. Sci. 1999, 9,
203.
(25) Nieto-Draghi, C.; Avalos, J. B. Mol. Phys. 2003, 101, 2303.
(26) Maruyama, S. Microscale Thermophys. Eng. 2003, 7, 41.
(27) Humphrey, W.; Dalke, A.; Schulten, K. J. Mol. Graphics 1996,
14, 33.
(28) Lee, J. W.; Meade, A. J.; Barrera, E. V.; Templeton, J. A. Proc.
ImechE, Part N: J. Nanoeng. Nanosys. 2010, 224, 41.
(29) Padgett, C. W.; Brenner, D. W. Nano Lett. 2004, 4, 1051.
(30) Thomas, J. A.; Iutzi, R. M.; McGaughey, A. J. H. Phys. Rev. B
2010, 81, 045413.
(31) Vallabhaneni, A. K.; Rhoads, J. F.; Murthy, J. Y.; Ruan, X. J. Appl.
Phys. 2011, 110, 034312.
(32) Kumar, S.; Kamaraju, N.; Karthikeyan, B.; Tondusson, M.;
Freysz, E.; Sood, A. K. J. Phys. Chem. C 2010, 114, 12446.
(33) Jo, I.; Hsu, I.-K.; Lee, Y. J.; Sadeghi, M. M.; Kim, S.; Cronin, S.;
Tutuc, E.; Banerjee, S. K.; Yao, Z.; Shi, L. Nano Lett. 2011, 11, 85.
(34) Fung, Y.-C. Foundations of solid mechanics; Prentice-Hall: Upper
Saddle River, NJ, 1965.
(35) Togo, A.; Oba, F.; Tanaka, I. Phys. Rev. B 2008, 78, 134106.
(36) Brenner, D. W.; Shenderova, O. A.; Harrison, J. A.; Stuart, S. J.;
Ni, B.; Sinnott, S. B. J. Phys.: Condens. Matter 2002, 14, 783.
(37) Wang, H. D.; Cao, B. Y.; Guo, Z. Y. Int. J. Heat Mass Transfer
2010, 53, 1796.
(38) Zhang, S.; Xia, M.; Zhao, S.; Xu, T.; Zhang, E. Phys. Rev. B 2003,
68, 075415.
(39) Chang, C. W.; Okawa, D.; Garcia, H.; Majumdar, A.; Zettl, A.
Phys. Rev. Lett. 2007, 99, 045901.
(40) Li, N.; Li, B.; Flach, S. Phys. Rev. Lett. 2010, 105, 054102.
(41) Dodd, R. K. Solitons and Nonlinear Wave Equations; Academic
Press: New York, 1982.

AUTHOR INFORMATION

Corresponding Author

*E-mail: hum@ethz.ch (M.H.).

Present Address

Permanent address: Center for Heat and Mass Transfer,

Institute of Engineering Thermophysics, Chinese Academy of
Sciences, Beijing 100190, China, 5 and Graduate University of
Chinese Academy of Sciences, Beijing 100049, China.

Notes

The authors declare no competing nancial interest.

E-mail: dpoulikakos@ethz.ch (D.P.).

ACKNOWLEDGMENTS
X.Z. gratefully acknowledges the scholarship of the State
Scholarship Fund of the China Scholarship Council that
allowed him to be a visiting student at ETH Zurich and
contribute to this work. Computational support from the
Brutus Cluster at ETH Zurich and Supercomputing Center of
CAS in China is also gratefully acknowledged. This work was
supported by a grant from the Swiss National Supercomputing
Centre-CSCS under project ID s243 and s359.

REFERENCES

(1) Pop, E. Nano Res. 2010, 3, 147.

(2) Pop, E.; Mann, D.; Cao, J.; Wang, Q.; Goodson, K.; Dai, H. Phys.
Rev. Lett. 2005, 95, 155505.
(3) Kim, P.; Shi, L.; Majumdar, A.; McEuen, P. L. Phys. Rev. Lett.
2001, 87, 215502.
(4) Fujii, M.; Zhang, X.; Xie, H.; Ago, H.; Takahashi, K.; Ikuta, T.;
Abe, H.; Shimizu, T. Phys. Rev. Lett. 2005, 95, 065502.
(5) Choi, T. Y.; Poulikakos, D.; Tharian, J.; Sennhauser, U. Nano Lett.
2006, 6, 1589.
(6) Che, J.; C agin, T.; Goddard, W. A. Nanotechnology 2000, 11, 65.
(7) Wang, J.; Wang, J.-S. Appl. Phys. Lett. 2006, 88, 111909.
(8) Lukes, J. R.; Zhong, H. J. Heat Transfer 2007, 129, 705.
(9) Donadio, D.; Galli, G. Phys. Rev. Lett. 2007, 99, 255502.
(10) Hu, M.; Keblinski, P.; Wang, J.-S.; Raravikar, N. J. Appl. Phys.
2008, 104, 083503.
(11) Shiomi, J.; Maruyama, S. Phys. Rev. B 2006, 73, 205420.
(12) Greaney, P. A.; Lani, G.; Cicero, G.; Grossman, J. C. Nano Lett.
2009, 9, 3699.
3416

dx.doi.org/10.1021/nl300261r | Nano Lett. 2012, 12, 34103416