Massive thermal fluctuation of massless graphene electrons

Hosang Yoon and Donhee Ham

arXiv:1405.2356v2 [cond-mat.mes-hall] 6 Jul 2014

School of Engineering and Applied Sciences, Harvard University, Cambridge, Massachusetts 02138
(Dated: July 8, 2014)
Whereas thermal current noise hI 2 i in typical conductors is proportional to temperature T , hI 2 i
in graphene exhibits a nonlinear T dependence due to the massless nature of individual electrons.
This unique hI 2 i arising from individually massless electrons is intimately linked to the non-zero
collective mass of graphene electrons; namely, hI 2 i is set by the equipartition theorem applied to the
collective masss kinetic energy, with the nonlinear T -dependence arising from the T -dependence of
the collective mass. This link between thermal fluctuation and collective dynamics unifies hI 2 i in
graphene and typical conductors, while elucidating the uniqueness of the former at the same time.
PACS numbers: 05.30.Fk, 05.40.Ca, 72.70.+m, 72.80.Vp

Thermal agitation of electrons in a conductor creates

spontaneous current fluctuations, or Johnson noise1,2 ,
with power spectral density SI (f ) = 4kB T G (kB : Boltzmann constant; T : temperature; G: real conductance).
Nyquist explained this by equilibrating the thermal noise
energy with external macroscopic electromagnetic modes
according to the equipartition theorem2 . Alternatively,
Johnson noise can be explained by directly considering
internal microscopic thermal motions of electrons3 ; here
electrons (mass: m) are treated classically with MaxwellBoltzmann distribution, and the thermal fluctuation of
electron velocity, vf , is set by the equipartition theorem: hvf2 i = kB T /m dimensions. The aggregate of
hvf2 i causes the total current fluctuation hI 2 i T , from
which SI (f ) = 4kB T G follows.
The microscopic machinery behind the thermal noise
in graphene is then of interest. As individual graphene
electrons act as massless particles4 , the equipartition theorem cannot be applied in the way used in the traditional
microscopic approach, and thus, hI 2 i T will not hold
(SI (f ) = 4kB T G is still valid5,6 due to the fluctuationdissipation theorem7 ). Moreover, the electron/hole coexistence due to the zero-bandgap nature4 will further
enrich the behavior of hI 2 i in graphene.
Here we investigate the unique thermal fluctuation behavior, hI 2 i, in graphene. As the traditional microscopic
approach with Maxwell-Boltzmann statistics is fundamentally limited, we devise a general microscopic formalism based on Fermi-Dirac statistics, and evaluate the
nonlinear T -dependence of hI 2 i due to massless electrons
(and holes) in graphene. Interestingly, we then unveil
that this unique hI 2 i arising from individually massless
electrons is intimately linked to the non-zero collective (or
plasmonic) mass of graphene electrons, which we have recently measured8 ; i.e., hI 2 i is given by the equipartition
theorem applied to the collective masss kinetic energy,
with the nonlinear T -dependence of hI 2 i arising from the
T -dependence of the collective mass. By identifying this
link between the thermal fluctuation and collective dynamics, we explain the thermal noise hI 2 i in graphene
and typical conductors in a unified way, while delineating the uniqueness of the former at the same time.

I.

FLUCTUATION: MICROSCOPIC
FORMALISM

We first formulate the electron thermal velocity fluctuation hvf2 i in a general conductor. This formulation
is applicable to conductors in any dimensions, but for
simplicity, we consider a two-dimensional (2D) conductor, whether it be graphene with massless electrons or
2D conductors with massive (m 6= 0) electrons (e.g.,
GaAs/AlGaAs quantum well). An electron with a
wavevector k assumes an intrinsic velocity of vk : for a
massive 2D electron gas, vk = ~k/m, where k |k|;
for massless electrons in graphene, vk = vF (constant).
hvf2 i is evaluated by considering the intrinsic velocities
judiciously together with the Fermi-Dirac distribution,
fk = 1/[e(k )/kB T + 1] (k : single electron energy;
2
: chemical potential). Note that hv
R f i2 is not 2the 2av2
erage of vk over all electrons, (1/n) (d k/(2) )gvk fk
(g: spin/valley degeneracy; n: electron density). This
all-electron average counts many electron pairs moving
in opposite directions with the same velocity deep below the Fermi surface, whose velocities cancel and cannot
contribute to fluctuations. Its inadequacy is also evident
as it does not vanish at T = 0, whereas hvf2 i must.
For hvf2 i, we only consider electrons whose velocities
do not cancel. The probability that a k-state is occupied
and a k-state is not occupied is fk (1 fk ), and thus,
Z
1
d2 k
hvf2 i =
gv 2 fk (1 fk ),
(1)
n
(2)2 k
where the electron density n is
Z
d2 k
n=
gfk .
(2)2

(2)

With k = k , we rewrite fk (1 fk ) as
fk (1 fk ) =

fk
fk
=
,
(/kB T )
(k /kB T )

(3)

which we will make use of later. At low T , since fk (1

fk ) in k-space peaks around the Fermi surface with a
vanishing width for T 0, hvf2 i of Eq. (1) vanishes at
T = 0, as it should.

2
hvf2 i leads to the total current thermal fluctuation hI 2 i.
Consider a 2D conductor of width W and length l along
the x axis, with hI 2 i measured along the length. Only the
x-component of vf , or vf,x , is relevant to the measured
fluctuation. As a single electron contributes a fluctuation current of evf,x /l, and as there are a total of nW l
electrons,
hI 2 i = nW l

W 2
e2 2
hv i = ne2 hvf,x
i,
l2 f,x
l

(4)

2
where hvf,x
i = hvf2 i/2 (2 degrees of freedom). SI (f ) readily follows from hI 2 i. The autocorrelation of the stationary process I is9 hI(0)I(t)i = hI 2 ie|t|/ ( : Drude
scattering time), because electron scatterings randomize initial momenta at an average rate of 1/ . The
single-sided
power spectral density is then SI (f ) =
R
4 0 dthI(0)I(t)i cos(t) with = 2f , or,

SI (f ) = 4hI 2 i

.
1 + 2 2

(5)

Before applying this formalism to graphene, we first

apply it to a massive 2D electron gas, as the result can
be compared to the traditional microscopic approach3
valid for the massive electron gas. Using k = ~2 k 2 /2m,
2
vk = ~k/m, Eqs. (2) and (3), and hvf,x
i = hvf2 i/2 in
Eq. (1), we find
2
hvf,x
i

kB T
=
m

d f ( )
R
,
df ( )
0

(6)

where R k /kB T , /kB T , and f () 1/(e + 1).

Using 0 d s f ( ) = (1 + s)Li1+s (e ), where

P
(z) is the gamma function and Lin (z) = k=1 z k /k n is
the polylogarithm function, we reduce Eq. (6) to
2
hvf,x
i=

kB T
kB T Li2 (e )
=
,
m Li1 (e )
m

(7)

where we have used (d/dx)Lin (x) = x1 Lin1 (x). This

is consistent with the traditional microscopic approach3
based on Maxwell-Boltzmann statistics, in which Eq. (7)
results from the equipartition theorem. Eq. (4) then
yields
hI 2 i =

ne2 W
kB T T.
m l

II.

THERMAL FLUCTUATION IN GRAPHENE

We now apply the formalism to graphene with10 k =

~vF k and vk = vF . The constant vk , arising from
the massless nature of individual electrons and holes,
2
will yield a nonlinear T -dependency of hvf,x
i and hI 2 i,
sharply contrasting the linear T -dependency of the massive case. The electron/hole coexistence will further en2
rich the thermal fluctuation behavior. hvf,x
i and n of
Eqs. (1) and (2) are calculated separately for electrons
in the conduction band and holes in the valence band:
2
hvef,x
i=

vF2 Li1 (e )
vF2 Li1 (e )
2
;
hv
i
=
,
hf,x
2 Li2 (e )
2 Li2 (e )

(9)

g(kB T )2
g(kB T )2

Li
(e
);
n
=
Li2 (e ),
2
h
2(~vF )2
2(~vF )2
(10)
where subscripts e and h indicate electrons and holes,
and for the hole case, we have used fk = f ( + ) and
fk (1 fk ) = (/)fk = (/)fk .
To first see the massless effect without the complication
from the electron-hole coexistence, consider a fictitious
graphene with the conduction band only (electrons only)
with k = ~vF k. The T -dependency depends on whether
the chemical potential or electron density ne is fixed
for varying T . We consider the constant ne case, as it
is practically achieved with electrostatic gating. Then
ne = constant condition
[Eq. (10)] determines (T ) with
p
(0) = F = ~vF 4ne /g > 0 [Table I]. With this par2
i first grows linearly with T just as in
ticular (T ), hvef,x
the massive case, but eventually saturates to vF2 /2, deviating from the persistent linear T -dependence of the
massive case [Fig. 1].
This low-T similarity, high-T difference between the
massless and massive case can be explained with Eq. (1).
For kB T  F , fk (1 fk ) peaks sharply around the
Fermi surface, so vk = vF for graphene coincides with
vk vF for the massive case, while this peaks width
grows linearly with T . So Eq. (1) is linear to T in both
massless and massive cases. For kB T  F with
[Table I] (the same holds for a massive gas), in the
conduction band, fk (1 fk ) fk e(k )/kB T is
the far tail of the Fermi-Dirac distribution. So vk2 = vF2
(massless) and vk2 k 2 (massive) makes a difference in
2
Eq. (1); in the former, hvef,x
i saturates; in the latter,
2
hvef,x i T persists.
ne =

(8)

In sum, for the massive electron gas, our general microscopic approach and the traditional microscopic approach
2
agree; importantly, hvf,x
i T and hI 2 i T . Incidentally, Eq. (5) then yields SI (f ) = 4kB T [(ne2 /m)(1 +
2 2 )1 ](W/l), where the real part of the Drude conductivity = (ne2 /m)/(1 + i ) appears inside the
square brackets. As G = Re[W/l], we arrive at SI (f ) =
4kB T G.

TABLE I. (T ) and (T ) for conduction- and valence-bandonly graphene, and actual graphene electron-doped at T = 0.
T =0
T
Bands
Held constant

Conduction only
ne
F +
Valence only
nh
F + +
Conduction and valence
ne nh
F
0
0

3
2

10

0.5
F

0.8

T=0

or
T=0

2
hvf,x
i /vF2

10

10

0.6

10

10

0.4
2
i /vF2
hvf,x

10

0.3

0.4

0.2
2
hvef
,x i

0.2

or

kB T /F

2
hv hf
,x i

Electrons
Holes
kB T /F
1/2
6 ln(2)/ 2

T=0

0.1

1/2
0

0
0

0.5

1.5

0.5

kB T /F

1.5

We can also consider a fictitious graphene with the

valence band only (holes only) with k = ~vF k. In
this case, nh = constant [Eq. (10)] determines (T ) with
(0) = F < 0 [Table I]. The resulting T -dependence of
2
2
hvhf,x
i is exactly the same as that of hvef,x
i [Fig. 1].
Now consider the actual graphene with both the conduction and valence bands. Let graphene be electrondoped at T = 0 and the total charge density ne (T )
nh (Tp) be fixed via electrostatic gating. (0) = F =
~vF 4ne (0)/g > 0, nh (0) = 0, and ne (T ) nh (T ) =
ne (0) for any T . Using ne (T ) and nh (T ) from Eq. (10),
this last expression can be rewritten as

2
2
FIG. 2.
T -dependence of hvef,x
i and hvhf,x
i for electrondoped graphene ((0) > 0) with k = ~vF k, assuming constant charge density (i.e., ne nh = constant).

10

1.5 10

Lk,0 hI 2 i /F

2
FIG. 1.
T -dependence of hvf,x
i for fictitious conductionband-only graphene with constant ne , or valence-band-only
graphene with constant nh . Inset: the same plot, log scales.

g(kB T )2
[Li2 (e ) + Li2 (e )] = ne (0).
2(~vF )2

1
kB T /F

10

10

10

10

Eq. (12)

0.5

kB T /F
(kB T /F )2 ln 4
0

(11)

Eq. (11) determines (T ) [Table I]. 0 for T

contrasts the electron- or hole-only case; this is because
ne and nh grow increasingly similar (ne /nh 1) with T
2
2
i are still
despite their fixed difference. hvef,x
i and hvhf,x
given by Eq. (9), but due to the new (T ), T -dependence
2
2
of hvef,x
i and hvhf,x
i [Fig. 2] now deviates from Fig. 1.
2
hvef,x i is still linear to small T , as the actual electrondoped graphene in this regime is no different from the fic2
titious, electron-only graphene. For T , hvef,x
i also
2 2
2
saturates, but to (6 ln(2)/ )vF instead of vF /2, because
(T ) = 0 now, while (T ) in the
2
electron-only graphene. hvhf,x
i in Fig. 2 more drastically
differs from Fig. 1, as we start from an electron-doped
graphene. The small number of holes in the valence band
at low T are at the far tail of the Fermi-Dirac distribution
2
(similar to the T case of Fig. 1), so hvhf,x
i vF2 /2
2
for low T . For large T , 0, so hvhf,x i approaches
2
(6 ln(2)/ 2 )vF2 just like hvef,x
i.
2
2
These behaviors of hvef,x i and hvhf,x
i lead to a complex
2
nonlinear T -dependence of hI i. Considering both the
electron and hole current fluctuations, Eq. (4) is now

0.5

1.5

kB T /F

FIG. 3. hI 2 i vs. T for electron-doped graphene with constant charge density (i.e., ne nh = constant); Lk,0 =
4~2 l/ge2 F W . Inset: same plot, log scales.

2
2
hI 2 i = (W e2 /l)[ne hvef,x
i + nh hvhf,x
i], or

hI 2 i =

ge2 W
(kB T )2 [Li1 (e ) Li1 (e )],
4~2 l

(12)

using Eqs. (9) and (10). Figure 3 plots hI 2 i vs. T with

(T ) set by Eq. (11). At low T , as ne (T ) ne (0),
2
2
nh (T ) 0, and hvef,x
i T , we have hI 2 i hvef,x
i
2
2
T . At high T , as both hvef,x i and hvhf,x i converge to
(6 ln(2)/ 2 )vF2 , and as both ne and nh grow with T 2 (see
Eq. (11) with 0 for T ), hI 2 i (ne + nh ) T 2 .
In sum, the massless nature of electrons and holes and
their coexistence yield unique thermal fluctuation dy2
2
namics in graphene. Particularly, hvef,x
i, hvhf,x
i, and
2
hI i vary nonlinearly with T , contrasting the linear T dependence in massive electron gases.

4
Incidentally, graphene intraband conductivity is11
Z
 f ( ) f ( ) 
ge2 kB T
d

=
4~2 ( 1 + i) 0

ge2 kB T
=
[Li1 ( e ) Li1 (e )], (13)
4~2 ( 1 + i)
where the conduction and valence band contributions are
separated. Comparing the real part of the above with
Eq. (12) and using G = Re[W/l], we attain hI 2 i =
kB T G(1 + 2 2 )/ . By plugging this into Eq. (5), we
arrive at SI (f ) = 4kB T G. That is, despite the nonlinear
T -dependence of hI 2 i, as shows the same nonlinear T dependence except for the extra kB T factor, the Johnson
noise still holds. This is how the fluctuation-dissipation
relation7 manifests in graphene.

III.

The unique thermal fluctuation behavior hI 2 i has a

fundamental connection to the massive collective dynamics of individually massless graphene electrons. To explain this, we first briefly discuss the collective motion of
graphene electrons8 , while setting aside the fluctuation
problem. Let graphene electrons collectively move by a
voltage V along the x axis. Individual electron velocity
vF remains constant, but their wavevectors change along
the x axis; let this change be (same for all electrons)
at a certain time. The total kinetic energy of the electron
gas then has grown by a certain amount Ee ; the larger
the ||, the larger the Ee whether > 0 or < 0. So Ee
assumes a (smooth8 ) minimum at = 0, thus Ee 2
for small . On the other hand, the collective momentum follows Pe . So Ee Pe2 and the collective
motion exhibits a mass Me satisfying Ee = Pe2 /(2Me ),
while individual electrons are massless.
Thus in the collective motion, the voltage V accelerates
Me according to the Newtons second law, increasing its
velocity Vec Pe /Me . The frequency-domain equation
of motion is (ne W l)(eV /l) = iMe Vec . As the current
is Ie = ne eW Vec , V /Ie = i[Me /(ene W )2 ] iLek ,
where the kinetic inductance Lek emerges as another
manifestation of the collective inertia Me :
Lhk = (e2 n2h W 2 )1 Mh . (14)

Here we have also written the same relation for holes. In

sum, while graphene electrons are individually massless,
their collective motion is of massive nature and described
by Me (Mh ) or equivalently by Lek (Lhk ). Note that Ee =
2
Me Vec2 /2 = Lek Ie2 /2 and Eh = Mh Vhc
/2 = Lhk Ih2 /2. We
can find the expressions of Lek and Lhk in graphene from
Eq. (13). As Lk = Im[l/W ], we have
Lk =

4~2
1
l
.
ge2 kB T [Li1 (e ) Li1 (e )] W

4~2
l/W
4~2 l/W
,
L
=
.
hk
2

2
ge kB T Li1 (e )
ge kB T Li1 (e )
(16)
We now return to the fluctuation problem and find its
deep-seated connection to the collective dynamics. By
inspection of Eqs. (12) and (15), we see that
Lek =

1
1
Lk hI 2 i = kB T.
2
2

(17)

This can be broken into electron and hole contributions,

1
1
Lek hIe2 i = kB T,
2
2

1
1
Lhk hIh2 i = kB T,
2
2

(18)

1
1
as hI 2 i = hIe2 i + hIh2 i and L1
k = Lek + Lhk . Or equivalently, in terms of Me and Mh , and their thermal velocity
2
fluctuations hVec2 i and hVhc
i,

FLUCTUATION AND COLLECTIVE

DYNAMICS

Lek = (e2 n2e W 2 )1 Me ,

This is the overall kinetic inductance combining Lek and

1
1
Lhk in parallel as L1
k = Lek + Lhk , with

(15)

1
1
Me hVec2 i = kB T,
2
2

1
1
2
Mh hVhc
i = kB T.
2
2

(19)

Eqs. (17)(19) are the same statement on the intimate

relation between thermal fluctuations and collective dynamics. Although individual graphene electrons and
holes act as massless relativistic particles, their thermal
fluctuations are governed by the classical kinetic energies
of the collective electron mass Me and of the collective
hole mass Mh , with each receiving a thermal energy of
kB T /2, satisfying the equipartition theorem [Eq. (19)],
thus determining the collective velocity thermal fluctuations hVec2 i and hVec2 i. These directly translate to the
thermal current fluctuations of electrons and holes, hIe2 i
and hIh2 i [Eq. (18)]. Eq. (17) expresses this most concisely; the total current thermal fluctuation hI 2 i is determined by the total kinetic inductance storing an average
collective kinetic energy of kB T /2.
The relationship between thermal fluctuation and collective dynamics captured by Eq. (17) also holds for
the massive electron gas, as one can see from Eq. (8)
where Lk = (m/ne2 )(l/W ) is the kinetic inductance of
the massive electron gas. However, this massive case is
less surprising, as each electron already follows equipartition and the collective mass is their simple aggregate
(M = nW lm). The more interesting, and unifying, observation is that even in graphene with massless electrons, hI 2 i arises from their non-zero collective mass storing an averaged kinetic energy of kB T /2. As much as
the relation hI 2 i = kB T /Lk [Eq. (17)] offers a unified
picture for the thermal fluctuation in the massless and
massive electron gas, it also directly explains the unique
nonlinear T -dependence of hI 2 i in graphene, as Lk is decisively temperature dependent in graphene [Eq. (15)],
whereas in the massive electron gas Lk is constant and
thus hI 2 i T .
Interrogation of the collective (plasmonic) dynamics of
graphene electrons via noise measurement based upon

5
this study may be an interesting point of future investigation. In addition, the present study may in the future
be expanded to take into account the quantum radiation
regime.

1
2
3
4

donhee@seas.harvard.edu
J. B. Johnson, Phys. Rev. 32, 97 (1928).
H. Nyquist, Phys. Rev. 32, 110 (1928).
R. Kubo, Rep. Prog. Phys. 29, 255 (1966).
K. S. Novoselov, A. K. Geim, S. V. Morozov, D. Jiang,
M. I. Katsnelson, I. V. Grigorieva, S. V. Dubonos, and
A. A. Firsov, Nature 438, 197 (2005).
A. C. Betz, F. Vialla, D. Brunel, C. Voisin, M. Picher,
A. Cavanna, A. Madouri, G. F`eve, J.-M. Berroir,
B. Placais, and E. Pallecchi, Phys. Rev. Lett. 109, 056805
(2012).
A. C. Betz, S. H. Jhang, E. Pallecchi, R. Ferreira, G. F`eve,

ACKNOWLEDGMENTS

We thank the support of this research by the Air Force

Office of Scientific Research (FA9550-13-1-0211) and Office of Naval Research (N00014-13-1-0806).

7
8

10

11

J.-M. Berroir, and B. Placais, Nat. Phys. 9, 109 (2013).

H. B. Callen and T. A. Welton, Phys. Rev. 83, 34 (1951).
H. Yoon, C. Forsythe, L. Wang, N. Tombros, K. Watanabe, T. Taniguchi, J. Hone, P. Kim, and D. Ham, Nature
Nanotechnology advance online publication (2014),
10.1038/nnano.2014.112.
R. K. Pathria and P. D. Beale, Statistical mechanics, 3rd
ed. (Elsevier, Oxford, 2011).
S. Das Sarma, S. Adam, E. H. Hwang, and E. Rossi, Rev.
Mod. Phys. 83, 407 (2011).
L. A. Falkovsky, Phys. Usp. 51, 887 (2008).