Quantum Transport of Massless Dirac Fermions

Kentaro Nomura and A. H. MacDonald

Department of Physics, University of Texas at Austin, Austin TX 78712-1081, USA
(Dated: February 4, 2008)
arXiv:cond-mat/0606589v3 [cond-mat.mes-hall] 22 Dec 2006

Motivated by recent graphene transport experiments, we have undertaken a numerical study of

the conductivity of disordered two-dimensional massless Dirac fermions. Our results reveal distinct
differences between the cases of short-range and Coulomb randomly distributed scatterers. We
speculate that this behavior is related to the Boltzmann transport theory prediction of dirty-limit
behavior for Coulomb scatterers.

PACS numbers: 72.10.-d,73.21.-b,73.50.Fq

Introduction Graphene can be described at low- port on a finite-size Kubo formula analysis used to study
energies by a four component massless Dirac-fermion the quantum transport of MDFs. Over the range of sys-
(MDF) model[1] that has long attracted theoretical at- tem sizes that we can describe, we find that min is a few
tention because of appealing properties including chi- times larger for Coulomb scatterers than for short-range
ral anomalies[2, 3, 4, 5], randomness induced quantum scatterers and e2 /h.
criticality[6, 7, 8], unusual electron-electron interaction A MDF model describes graphene transport only when
effects[9, 10], relevance to high Tc superconductors[9, 11], intervalley scattering is unimportant. We argue that the
and various unusual transport properties[5, 6, 7, 8, 9, 10, MDF model likely does apply at accessible experimental
11, 12, 13, 14, 15, 16, 17, 18, 19, 20]. The recent experi- temperatures to graphene systems near the Dirac point,
mental realization of single-layer graphene sheets[21] has since the intervalley scattering length obtained within
made it possible to confirm a number of theoretical pre- the Born approximation diverges in that limit. The tem-
dictions, including unusual quantum Hall effects[22, 23], perature below which intervalley scattering is important
and has also revealed some surprises. The main find- should increase away from the Dirac point[25].
ings can be summarized as follows: i) graphenes conduc- Massless Dirac Fermion Model Graphenes honeycomb
tivity never falls below a minimum value ( min ) cor- lattice has two atoms per unit cell on sites labeled A and
responding (approximately) to a conductance quantum B. The low-energy band structure consists of Dirac cones
(e2 /h) per channel, in spite of predictions[6, 7, 11, 14, 16] located at the two inequivalent Brillouin zone corners K
based on the self-consistent Born approximation (SCBA) and K :
that min = (1/) e2 /h for a MDF channel, and predic-  
0 kx iky
tions that localization occurs[15, 17, 18] when interval- HK = ~ v k = v ~ , (1)
kx + iky 0
ley scattering is significant; ii) in gate-doped graphene
increases linearly with carrier density n away from and HK = ~v t k, where v is the graphene Fermi ve-
the charge neutrality (Dirac) point, implying a constant locity and the Pauli matrices act on the sublattice
mobility = /ne[22, 23] and not the constant con- degrees of freedom. For each wavevector k, Eq.(1) has

ductivity usually predicted for the Boltzmann transport two eigenstates |k, i = (|k, Ai ei |k, Bi)/ 2, where
regime[14]. Although these surprises have inspired a tan1 (ky /kx ), and eigenenergies Ek, = ~v|k|.
number of theoretical studies[16, 17, 18], the source of When intervalley scattering is neglected the K and K
the discrepancies between experiment and theory has not valleys contribute independently to the conductivity.
yet been conclusively identified. Boltzmann theory for doped grapheneWe start by
We have recently pointed out that the linear depen- briefly reviewing Boltzmann transport theory applied
dence of conductivity on carrier density in graphene can to graphene since this consideration motivates Colomb
be explained in the framework of Boltzmann transport scatterer models. The Boltzmann conductivity 0s =
theory by assuming Coulomb scatterers[24] rather than (e2 /h) (2EF 0 /~) = (e2 /h) 2kF is proportional to the
the short-range scatterers assumed, mainly as a practical transport relaxation time 0 . For short-range scatter-
simplification, in most theoretical work. The golden-rule ers the Born approximation gives ~/0 = 2V 2 F =
(ni u2 /~v 2 ) |EF | [14], where V (r) = u N
P i
scattering rate for Coulomb scatterers in a MDF model I (r RI )
diverges in the zero energy (Dirac point) limit, whereas is the disorder potential, ni is the density of scatterers,
it vanishes for short-range scatterers. This property sug- and F the density of states at the Fermi level. When
gests the possibility of a qualitative difference between the range of the impurity potential is much longer than
these two disorder models. Since Boltzmann theory is the lattice spacing of graphene, intervalley scattering is
not applicable in the vicinity of the Dirac point, how- weak[14, 15, 17].
ever, a fully quantum mechanical approach is required to Note that 0s is independent of the carrier density n.
address the minimal conductivity. In this Letter we re- Experiment, on the other hand, finds that the mobil-
 2

20
a 10
short-range
Coulomb b 18 tion, and |ni denotes an eigenstate of the Dirac equation,

DOS ()
16
8
14
h i
i~v + V (r) = E (3)

[e2/h]
[e2/h]
12
6
0 10
-0.5 0 0.5
4 Energy 8
6
2 4
which we solve using a large momentum-space cutoff .
0
2
0
The disorder potential momentum-space matrix elements
-1 -0.5
n/ni
0 0.5 1 -1 -0.5 0
n/ni
0.5 1
are
3
c d 2
Ni
1 X
1.0 1
hk|V |k i = 2 U (k k ) ei(kk )RI , (4)
2 L
[e2/h]

[e2/h]
2
unscreened
0
0 0.2 0.4 0.6 I=1
screened
[e2/h]

0.5 1
1 where U (q) is constant for delta-impurities and given by
E=0 0
1 2 3
Usc (q) for the screened Coulomb scattering case. The
L /L
0.0
0 1 2
min

3
0
0 0.5 1 1.5
scattering center locations RI and potential signs were
chosen at random. The results shown in Fig.1 are typ-
FIG. 1: (Color online) Dirac fermion conductivities for (a)
ically averaged over 104 disorder realizations. We es-
short-range scatterers and (b) screened Coulomb scatterers. timate the bulk conductivity by evaluating Eq.(2) at a
The inset of (a) compares the densities of states for short- large number of -values. The Kubo conductivity van-
range and Coulomb cases. (c) shows Kubo formula conductiv- ishes for both small and large but there is an inter-
ities at the Dirac point as a function of for Coulomb scatter- mediate region where the dependence of on is rel-
ers for three different system sizes. The inset of (c) shows the atively weak. We use the maximum of vs. to es-
size dependence of the conductivity for unscreened Coulomb timate the conductivity at a given system size L. For
(open circles) and screened Coulomb (boxes) cases.(d) Distri-
bution function of the conductivity and resistivity (inset) at
metals, including doped graphene, physical arguments
the Dirac point for Coulomb scatterers. In the main panel, suggest that ~/TL where TL is the escape time
the solid (blue) line is for L = 1.4Lmin and the dotted (red) from the system studied numerically. It follows[26] that
line is for L = 2Lmin , where Lmin is the minimum system size hEi gT E where the Thouless energy hEi
considered. is the geometric mean of the eigenvalue difference be-
tween periodic and antiperiodic boundary conditions,
E = 1/L2F is the level spacing at the Fermi level,
and gT hEi/E is the Thouless number.
ity = /ne in graphene is nearly constant except at The reliability of the finite-size Kubo formula method
very low-densities. One plausible explanation for this described above is solidly established for diffusive metal-
behavior is that Dirac fermion scattering is dominated lic systems[26]. For conventional non-relativistic two-
by Coulomb scattering from PN ionized impurities near the dimensional electron systems with parabolic dispersion,
graphene plane, V (r) = I i e2 /|r RI |. Using Fermis numerical conductivities calculated in this way agree
golden rule and approximating the screened Coulomb closely with the theoretical expectation of a Drude
interaction by[24] Usc (q) = (2e2 )/(q + 4g kF ) conductivity 0 = (e2 /h)kF with a finite-size weak-
(~v)/(2kF ), the Boltzmann conductivity for Coulomb localization correction[27]. Subtleties, related to the
interactions is 0c (4e2 /h) (n/ni ) 32/, proportional to dependence of the numerical estimate, do however arise
density in agreement with experiment. Here g = e2 /~v when this method is applied to insulators. The Dirac
is the effective fine structure constant used to character- point of the MDF model is an intermediate case and re-
ize the ratio of Coulomb interaction and band energy lated uncertainties apply to our numerical estimate of
scales in graphene. (g 3 in vacuum and 1 when the min . The property that the Dirac point density of states
graphene sheet is placed on a SiO2 dielectric substrate.) is finite in the presence of disorder, as illustrated in the
Note that the Boltzmann conductivity for Coulomb scat- inset of Fig.1(a), may help validate the Kubo approach.
terers vanishes as n 0, contradicting experiment[22]. We find that the dependence of on at the Dirac point,
MDF model finite-size Kubo Formula Our numerical illustrated in Fig.1(c) is similar to that in a metal[26],
results, obtained by evaluating the finite-size Kubo for- indicating delocalized states. is a maximum when
mula corresponds to E at each system size as expected for
e2 /h. We have also evaluated the Thouless conduc-
i~e2 X f (En ) f (En ) hn|vx |n ihn |vx |ni tivity as a consistency check. A large number of numer-
= , (2) ical studies have demonstrated that the Thouless con-
L2 En En En En + i
n,n ductivity estimate, although perhaps not as accurate in
the diffusive metal limit ( >> e2 /h), is more univer-
are summarized in Fig.1. Here v = v is the Dirac sally applicable. It may be used for both delocalized and
fermion velocity operator, f (E) is the Fermi-Dirac func- strongly localized states[28], and should therefore be re-
 3

liable at the Dirac point. We find consistency between ture of delocalized states. Although this argument is not
the two estimates. quantitative, it makes it clear that there is an essential
Fig.1 compares Kubo conductivity estimates for (a) difference between the Coulomb and short-range cases.
the short-range scatterer and (b) the screened Coulomb A related difference is also seen in the inset in Fig.1(a)
scatterer cases. For the short-range disorder potential which compares densities of states near the Dirac point
case, the density dependence of the conductivity is non- (E = 0) for short-range scatterers (green dashed line)
linear, approaching the constant Boltzmann conductivity and Coulomb scatterers (blue solid line). The prominent
for |EF | ~/ . The estimated value of (E = 0) is close dip at E = 0 in the short-range case is replaced by a
to the SCBA value (1/)e2 /h predicted in earlier theo- smooth minimum at a larger value in the Coulomb case.
retical studies[6, 7, 11, 14, 16]. For the screened Coulomb One interpretation of the increase in Dirac point density
scatterer case, on the other hand, the conductivity in- of states is that the carrier density fluctuates spatially
creases linearly with increasing density |n| as Boltzmann in the smooth Coulomb potential. Non-zero local car-
theory predicts. At the Dirac point, however, the con- rier densities at the Dirac point could explain both the
ductivity remains finite with the minimum value e2 /h, increase in density of states and the increase in conduc-
which is few times larger than the SCBA value for the tivity. For short-range scatterers, our simulations have
short-range model. To illustrate the dependence on car- focused on the Boltzmann dominated kF >> 1 regime.
rier density, we have smoothed the curves in Fig.1(a) and We note that systems with kF 1 or smaller behave
(b) by averaging over Fermi energy interval containing more like Coulomb scatterer systems at low densities,
typically 1-30 levels, and over boundary conditions, in and have larger minimum conductivities and densities of
addition to over approximately 104 disorder realizations. states.
If we neglect intervalley scattering and account for Although our theory does not account for electron-
graphenes spin and valley degeneracies, MDF proper- electron interactions, the issue of screening effect near
ties can be compared with graphene experimental results. the Dirac point requires comment. We use the T =
Both the linear dependence of the conductivity on density 0 Thomas-Fermi approximation U (q) = 2e2 /(q +
and the shift of min suggest that long-range scattering 8e2 F ) to give an indication of the importance of screen-
similar to that produced by ionized impurities is present ing, although this approximation must fail for EF 0
in experimental samples. We note that ionized impuri- since it applies strictly only for disorder potential range
ties in the substrate or at samples edges that are sep- much longer than the Fermi wavelength and kB T EF .
arated from the conduction channel by a distance that The density of states at the Dirac point F vanishes in the
is large compared to the graphene lattice constant but clean limit whereas it can be evaluated self consistently
small compared to the Fermi wave length in the regime in the disordered case, leading to the small but finite
studied experimentally (|n| . 7 1012[cm2 ][22, 23]) will value seen in the inset in Fig.1(a). The inset in Fig.1 (c)
act like Coulomb scatterers in the MDF model but will compares the conductivity for unscreened and screened
not produce strong intervalley scattering. case at several different system sizes. We find that the
The difference between Coulomb and short-range min conductivities evaluated with screened interactions are
values can be rationalized by the following argument. In suppressed somewhat compared to the unscreened case.
the short-range case, the golden-rule relaxation time for It appears likely that the precise value of min may de-
delta-function scatterers diverges (0 1/|EF | ) pend indirectly on electron-electron interactions.
at the Dirac point and kF remains finite, which sug- We emphasize that we are able to evaluate the Kubo
gests that standard transport theory considerations may formula only over a relatively small range of finite sys-
apply at least qualitatively once the non-vanishing den- tem sizes. If we use the mobility to convert to physi-
sity of states in disordered systems is accounted for. In cal length units the minimum system size we study is
contrast, Boltzmann transport theory suggests that the 0.1m, not much smaller than the size of the crystal-
Dirac point of the Coulomb scatterer model is in the lites studied experimentally.[22] Over the range of sizes
strong disorder limit because c |EF | 0. We ex- we are able to study the conductivity does increase very
amined this idea by varying the carrier density n and the slowly with system size, even at E = 0, as shown in
ionized impurity density ni independently, finding that the inset of Fig.1(c). This weak size dependence might
the conductivity for Coulomb scattering model appears be hinting that the infinite conductivity predicted for
to be a function of n/ni only. Letting n 0 at finite two-dimensional systems with symplectic symmetry[29]
ni and letting ni at finite n are equivalent be- by scaling arguments will emerge in the MDF model at
cause min e2 /h. This finding is consistent with extremely large system sizes even at E = 0. It is not
the idea that at the Dirac point MDF Coulomb scatter- obvious to us that the scaling theory applies at the Dirac
ers are always in the strongly disordered limit. As long point for either Coulomb or short-range scatterers since
as these states remain delocalized, however, the conduc- there is no length scale on which e2 /h. Alternately,
tivity cannot vanish; the non-zero conductivity at the the weak size dependence we find might be related to en-
Dirac point is purely due to the quantum mechanical na- hancement of carrier-density spatial fluctuations, which
 4

can be imagined in terms of puddles of electrons and port the view that graphene is described by a MDF model
holes, at larger system sizes in the long-range disorder near the Dirac point. Our numerical results suggest that
potential case. Since Dirac fermions are not strongly the MDF conductivity is finite and e2 /h at the Dirac
localized, these puddles are not isolated but effectively point over the relevant range of the system size and the
percolate[30] and are constrained by boundary condition coherence length.
in finite size systems. The large length scale limit of Acknowledgment The authors acknowledge helpful
the conductivity at zero temperature is more obvious for interactions with A. Geim, T. Hughes, Z. Jiang, J. Jung,
real graphene than for its (two-component) MDF model, P. Kim, V. Falko and A. Castro-Neto. This work has
since intervalley scattering will always become relevant been supported by the Welch Foundation and by the De-
and lead to localization.[15, 17, 18] partment of Energy under grant DE-FG03-02ER45958.
Finally we comment on statistical properties of the
conductivity and the resistivity at the Dirac point. Fig- Note added After submitting the present paper,
ure.1 (d) shows the distribution function of the con- we became aware of related work in which the impor-
ductivity (main panel) and the resistivity (inset). The tance of the long-range nature of disorder scatterers is
distribution function of has a sharp peak near = discussed.[33]
(1/)e2 /h, and a large tail on the the large regime. On
the other hand, the resistivity distribution function has
a broad peak near h/e2 , as seen in experiment.[22]
Discussion We find that a MDF model with Coulomb [1] J. C. Slonczewski and P. R. Weiss, Phys. Rev. 109, 272
scatterers is able to account for two key findings of exper- (1958).
iments on graphene sheets, namely that the conductivity [2] S. Deser, R. Jakiew, and S. Templeton, Phys. Rev. Lett.
is proportional to carrier density away from the Dirac 48, (1982) 975.
point and that the minimal conductivity per channel is [3] G.W. Semenoff, Phys. Rev. Lett. 53, 2449 (1984)
finite and larger than the SCBA value obtained theo- [4] E. Fradkin, E. Dagotto D. Boyanovsky, Phys. Rev. Lett.
57, (1986) 2967.
retically for a MDF model with short-range scatterers. [5] F.D.M. Haldane, Phys. Rev. Lett. 61, 2015 (1988);
The impurity density ni in the Coulomb model should [6] E. Fradkin, Phys. Rev. B 33, 3263 (1986)
be associated with the density of ionized impurities that [7] A. W. W. Ludwig, M. P. A. Fisher, R. Shankar, and G.
are located in the substrate within a Fermi wavelength Grinstein, Phys. Rev. B 50, 7526 (1994).
of the graphene plane. In this interpretation, the mobil- [8] C. de C. Chamon et al., Phys. Rev. Lett. 77, 4194 (1996).
ities measured in current samples[22, 23] correspond to [9] D.H. Kim et al., Phys. Rev. Lett. 79, 2109 (1997).
ni 5 1011 [cm2 ]. The property that the conductiv- [10] D. V. Khveshchenko, Phys. Rev. Lett. 87, 206401 (2001);
J. Gonzalez et al., Phys. Rev. B 63, 134421 (2001);
ity is proportional to carrier density in the Boltzmann T. Stauber et al., Phys. Rev. B 71, 041406 (2005);
regime suggests dominant smooth intravalley scattering Khveshchenko, cond-mat/0604180.
that could be Coulomb in character. If so intervalley [11] P.A. Lee, Phys. Rev. Lett. 71, 1887 (1993).
scattering is likely to be irrelevant in graphene in the [12] Y. Hatsugai and P. A. Lee, Phys. Rev. B 48, 4204 (1993).
Boltzmann transport regime, although it will always be [13] Y. Morita and Y. Hatsugai, Phys. Rev. Lett. 79, 3728
important at finite carrier densities in the weak local- (1997); Phys. Rev. B 58, 6680 (1998).
[14] N.H. Shon and T. Ando, J. Phys. Soc. Jpn. 67 (1998)
ization regime when the temperature is low enough that
2421.
the phase coherence length exceeds the intervalley scat- [15] H. Suzuura and T. Ando, Phys. Rev. Lett. 89, 266603
terling length, l0 1/|EF |. When intervalley scattering (2002).
is weak, the MDF model applies and momentum space [16] N. M. R. Peres, F. Guinea, A. H. Castro Neto, Phys.
Berry phases change constructive interference of back Rev. B 73, 125411 (2006).
scattering particles into destructive interference[15, 31], [17] E. McCann et al., Phys. Rev. Lett. 97, 146805 (2006);
implying weak antilocalization. When local scatterers D.V.Khveshchenko, Phys. Rev. Lett. 97, 036802 (2006);
dominate in graphene intervalley and intravally scatter- A. F. Morpurgo, F. Guinea, Phys. Rev. Lett. 97, 196804
(2006)
ing rates[14, 15, 17] are comparable and the accumu- [18] M.I. Katsnelson, Eur. J. Phys B 51, 157 (2006); J.
lated Berry phases are randomized , so that all states are Tworzydlo et al., Phys. Rev. Lett. 96, 246802 (2006);
localized.[13, 20] Qualitative studies of electron-phonon K. Ziegler, cond-mat/0604537; I.L. Aleiner, K.B. Efetov,
interaction[32] and electron-electron interaction[10] in Phys. Rev. Lett. 97, 236801 (2006); A. Altland, Phys.
graphene give important information on the coherence Rev. Lett. 97, 236802 (2006).
length. In a weak magnetic field, the magnetic length [19] A.M.J. Schakel, Phys. Rev. D 43, 1428 (1991); Y. Zheng
and T. Ando, Phys. Rev. B 65, 245420 (2002); V.P.
limits the coherence length. One recent magnetoresis-
Gusynin and S.G. Sharapov, Phys. Rev. Lett. 95, 146801
tance study has shown that highly doped graphene ex- (2005).
hibits a negative magneto resistance indicating weak- [20] D.N. Sheng et al., Phys. Rev. B 73, 233406 (2006).
localization, while states remain delocalized in the vicin- [21] K.S. Novoselov et al., Science 306, 666 (2004).
ity of the Dirac point [25]. These properties further sup- [22] K.S. Novoselov et al., Nature 438, 197 (2005)
 5

[23] Y.B. Zhang et al., Nature 438, 201 (2005). Asada et al., Phys. Rev. B 73, 041102 (2006); Physica E,
[24] K. Nomura, A.H. MacDonald, Phys. Rev. Lett, 92 34, 228 (2006).
256602 (2006). [30] M. I. Katsnelson, K. S. Novoselov, and A. K. Geim, Na-
[25] S.V. Morozov et al., Phys. Rev. Lett. 97, 016801 (2006). ture Phys. 2, 620 (2006).
[26] D. J. Thouless and S. Kirkpatrick, J. Phys. C, 14, 235 [31] T. Ando, T. Nakanishi, R. Saito, J. Phys. Soc. Jpn. 67,
(1981), Y. Imry, Introduction to Mesoscopic Physics, Ox- 2857 (1998).
ford University. [32] A. H. Castro Neto, F. Guinea, cond-mat/0608543.
[27] P.A. Lee and T.V. Ramakrishnan, Rev. Mod. Phys. 57, [33] H. Kumazaki and D. S. Hirashima: J. Phys. Soc. Jpn.
287 (1985). 75, 53707 (2006); T. Ando, J. Phy. Soc. Jpn., 75, 74716
[28] T. Ando, J. Phys. Soc. Jpn. 52, 1740 (1983); 53 3101 (2006); E.H. Hwang, S. Das Sarma, cont-mat/0610561.
(1984); 53 3126 (1983).
[29] S. Hikami et al., Prog. Theor. Phys. 63 707 (1980); Y.