Sie sind auf Seite 1von 10

Transport in bilayer graphene near charge neutrality: Which scattering

mechanisms are important?

Glenn Wagner,1 Dung X. Nguyen,1 and Steven H. Simon1

Rudolf Peierls Centre for Theoretical Physics, Parks Road, Oxford, OX1 3PU, UK
Using the semiclassical quantum Boltzmann equation (QBE), we numerically calculate the DC
transport properties of bilayer graphene near charge neutrality. We find, in contrast to prior dis-
cussions, that phonon scattering is crucial even at temperatures below 40K. Nonetheless, electron-
electron scattering still dominates over phonon collisions allowing a hydrodynamic approach. We
introduce a simple two-fluid hydrodynamic model of electrons and holes interacting via Coulomb
drag and compare our results to the full QBE calculation. We show that the two-fluid model
produces quantitatively accurate results for conductivity, thermopower, and thermal conductivity.
arXiv:1905.09835v1 [cond-mat.mes-hall] 23 May 2019

Two-dimensional materials such as graphene[1] BLG[29, 30]. Compared to [29], which studies the
have attracted an enormous amount of attention case of clean BLG, we calculate the conductivity
in the last decade. While the transport properties away from CN, which requires including additional
of graphene have been extensively studied[2], other scattering mechanisms. The paper [30] does study
related materials are only beginning to be exam- the conductivity of BLG away from CN by includ-
ined. Advances in nanotechnology have recently ing the effect of disorder, however we aim to pro-
allowed electrical measurements on suspended bi- vide a more quantitative analysis allowing for com-
layer graphene (BLG) samples[3–11]. Motivated parison with experimental data. We compare our
by these advances, we theoretically examine the results for the electrical conductivity to the exper-
DC transport properties of BLG using a quantum imental results from Ref. [3] in order to extract the
Boltzmann equation (QBE) approach. value of the only free parameter of the theory: The
As in the case of monolayer graphene[12, 13], phonon scattering rate. The value we determine
we expect that electrons in BLG will have high for the phonon deformation potential lies within
mobility and scattering of electrons among them- the range reported by previous authors. We then
selves will be the dominant scattering mecha- use this value to calculate the thermal conductiv-
nism. We thus expect that the quantum Boltz- ity and the thermo-electric coefficient. Finally, we
mann equation (QBE) will reduce to some form of introduce the two-fluid model and show that it
hydrodynamics[14–20]. agrees quantitatively with the more detailed QBE
Measurement of the electrical conductivity in numerics.
BLG has been reported in Ref. [3] and we show QBE —At low energies, bilayer graphene can be
that our QBE results agree with the experimental described in a two-band model with a quadratic
data over a wide range of parameters. We focus dispersion λ (k) = λ|k|2 /2m where λ = +(−) for
our calculation on the regime explored in that ex- electrons in the conduction (valence) band [25, 31],
periment, i.e., close to charge neutrality and at m = 0.033me is the effective mass [32], and k is
temperatures T ∼ 10K – 40K (using the common the two-dimensional momentum. There are two
Bernal stacked BLG with no out of plane field). valleys (K and K 0 ) and two spin states in each
We show that in this regime, surprisingly, the ef- valley, thus giving Nf = 4 flavors. We neglect ad-
fect of phonon scattering plays an essential role ditional effects at low temperatures such as trig-
(with the effect of finite sample size playing a mi- onal warping [33] and opening of an interaction
nor role). We then use the QBE to make predic- induced gap [34], restricting the applicability of
tions for the thermopower and thermal conduc- our approach to T & 10K [3].
tivity that can verified in future experiments. We The QBE is a semiclassical approach which is
then show that our transport results can be quanti- valid when the de Broglie wavelength is much
tatively understood with an extremely simple two- shorter than the scattering lengths. This is equiv-
fluid hydrodynamic model[21–24]. This simplifica- alent to the condition αx ≡ βτx−1  1, where τx
tion allows intuitive understanding of the physics stands for the scattering time due to any of the
that is not otherwise possible. scattering mechanisms in the problem, and β is
In this paper we begin by briefly discussing our the inverse temperature.
QBE method, which we elaborate in detail in a In equilibrium the occupancy of a k state is
companion paper[25]. This approach is similar to given by the usual Fermi factor. The QBE de-
previous work on the conductivity of monolayer termines the change in occupancy of the k states
graphene [26, 27], two coupled monolayers [28] and due to small perturbations such as an external
electric field E or temperature gradient ∇T . For-
mally the QBE is derived from the Kadanoff-Baym 3.0
equations for the evolution of Green’s functions 14K
using a Born collision integral to describe scatter- 2.5 16K
ing. Once Boltzmann equations are formally de- 20K

rived, the solution is obtained by expanding the 2.0 30K
k space occupation function f (k) in a set of ba- 35K
sis functions, which then reduces the QBE to a 40K
1.5 QBE
set of linear equations which can be numerically 2 fluid model
solved. The size of the basis set is then expanded
to convergence[27–29, 35]. This method of solu- 1.0
−1.0 −0.5 0.0 0.5 1.0
tion for the Boltzmann equation is well-known in βµ
the plasma physics community and goes by the
name Spitzer-Härm method [36]. Details of both FIG. 1. Electrical conductivity σ(βµ) for different val-
the derivation of the QBE, and the method of so- ues of the temperature. From the experimental data,
lution are presented in detail in Ref. [25] and the we fit the value αph = 0.05. The experimental data is
Supplementary Material[37]. from [3]. The black solid curve is the result of the QBE
We calculate the electrical current J and the calculation. The black dashed curve shows the result
from the two-fluid model and shows good agreement
heat current, which is defined as Q = JE − µe J,
with the full QBE calculation.
where JE is the energy current and e < 0 is the
electron charge. From this we can determine the
electrical conductivity σ, the thermal conductivity temperatures would not collapse when plotted as a
K and the thermo-electric coefficient Θ by function of βµ, and this does not agree with the ex-
perimental data. We thus disregard these two scat-
J σ Θ E
= (1) tering mechanisms. The situation, however, is dif-
Q TΘ K −∇T .
ferent for phonon scattering. The experiment is at
The open circuit thermal conductivity [38] which temperatures above the Bloch-Grüneisen temper-
is usually measured in experiments is given by κ = ature and hence the longitudinal acoustic phonons
K − T Θσ −1 Θ. have a scattering time [41, 42]
Coulomb and phonon scattering— We first con-
sider Coulomb (electron-electron, electron-hole) −1 D2 mkB T
scattering, which we expect to be dominant. We τph = , (2)
use the form of the screened Coulomb potential
valid in the experimentally-relevant regime βµ . where c is the speed of sound in graphene, D is
1. In the experimental data[3], to a good approx- the deformation potential and ρ is the mass den-
imation the conductivity only depends on the di- sity of BLG. The relevant dimensionless parame-
mensionless combination βµ. (See Fig. 1 [39]). If ter αph = βτph is then independent of tempera-
only Coulomb interactions are included then βµ is ture and should result in the conductivity being
the only dimensionless parameter of the problem a function of βµ only, in close agreement with
(the electromagnetic fine-structure constant can- experiment[3]. We emphasize the surprising result
cels out when the screened potential is used[37]). that even at these comparatively low temperatures
However, away from charge neutrality (CN), the of 12 − 40K, phonons provide the primary momen-
so-called momentum mode (i.e., a simple Galilean tum relaxation mechanism.
boost), where electrons and holes move in the same Electrical conductivity.— In the absence of a
N f e2
direction, carries electrical current, and is not re- magnetic field the conductivity σij = 2~ σ̃δij
laxed by Coulomb scattering since it conserves mo- depends on the dimensionless numbers βµ and
mentum. Thus, to obtain a finite conductivity, an- αph . The thermal density of electrons (holes)
other scattering mechanism that relaxes momen- for the free Fermi gas is given by ne(h) =
tum must be considered. We identify three pos- (Nf m/(2πβ)) ln(1 + e±βµ ) and as we increase |βµ|
sible such mechanisms: Impurity scattering, scat- there are more charge carriers. However the to-
tering off the boundary of the finite-size sample, tal number density of electrons plus holes only in-
and phonons. In the former two cases, the rel- creases by 17% between |βµ| = 0 and 1 (at fixed
evant dimensionless scattering parameter αx de- T ), and therefore this effect alone is too small to
pends on temperature[40]. Therefore, if either of account for the electrical conductivity more than
these scattering mechanisms were most important doubling in this range. To understand this we re-
(after Coulomb scattering) the curves for different alize that the primary scattering is electron-hole

collisions[3] and therefore the conductivity at large
chemical potential is large both because we have
many electrons to carry the charge and also be-
cause there are not many holes to scatter from. A
non-zero αph makes the conductivity finite and, as
mentioned above, the curves for different temper-
atures approximately collapse when plotted as a
function of βµ. We treat the deformation poten-
tial D as a fit parameter being that various dif-
ferent approaches have given different estimates of
this quantity[43–47]. Our best fit value αph = 0.05
corresponds to D ≈ 27eV, which is consistent with
prior expectation that it lies in the range 10−30eV.
Including the effect of finite size (boundary scat-
tering) reduces the phonon scattering required to
match the experiment. Taking account of the fact
that the typical size of the system is around 3µm,
and there will be additional momentum-relaxing
scattering off the boundary of the sample, the best
FIG. 2. Top: The dimensionless open-circuit ther-
fit D may be reduced by around 30%. 2
mal conductivity κ̃ = (Nf kB T /2~)−1 κ calculated us-
In Fig. 1 we show σ as a function of βµ. We show ing QBE (blue) and two-fluid approximation (yellow).
both the result of QBE calculation using the above Bottom: The dimensionless thermoelectric coefficient
discussed fit value of D as well as the experimental is defined as Θ̃ = (Nf ekB T /2~)−1 Θ calculated us-
data from Ref. [3]. ing QBE (blue) and two-fluid approximation (yellow).
The two curves show near perfect overlap. Both figures
Exactly at CN, our prediction for the magni- use αph = 0.05.
tude of σ matches prior calculations by Ref. [29]
to within about 1.5% between 10K –100K. Com-
parison to experiment is more difficult because dif- tering between electrons and holes is suppressed
ferent samples give different precise values of con- due to both matrix element effects and energy-
ductivity — differing from each other by factors of momentum conservation constraints[25], we be-
up to about 4. Our prediction lies acceptably in lieve treating the electron fluid and the hole fluids
the middle of the experimental range. as weakly interacting with each other will be ac-
Thermal Properties.—Including only Coulomb curate. In this limit, due to the strong scattering
scattering, the thermal conductivity K diverges within each species, each of the two fluids should
since the momentum mode carries thermal cur- have a well defined temperature, chemical poten-
rent and cannot be relaxed by Coulomb interac- tial, and velocity. We thus introduce a two-fluid
tions. Phonons again regulate this divergence. model [21–24], which shows excellent agreement
One might expect that the thermal conductivity with our detailed numerical calculation[48]. The
increases with increasing |βµ| since the total heat evolution of the mean fluid velocities of electrons
transport carrier density ne + nh increases. How- (holes) ue (uh ) can then be derived explicitly from
ever, we note that K actually decreases with in- the QBE (See Supplementary Material[37] for de-
creasing |βµ| which is counter-intuitive. In the tailed derivation) giving[49].
limit of weak phonon scattering we will see below m e m e
that K ∼ 1/(ne + nh ). The open circuit thermal mu̇e = − (u − uh ) − u + eE − Λe kB ∇T
N k2 T
τeh τph
conductivity κij = f2~B κ̃δij , plotted in Fig. 2 m h m h
(top), decreases faster than K away from CN since mu̇h = − (u − ue ) − u − eE − Λh kB ∇T
τhe τph
the momentum mode carries electric current and
hence does not contribute to κ. The thermoelec- Here τeh = τ0 (ne + nh )/nh is the scattering time
N ek T
tric coefficient Θij = f 2~B Θ̃δij plotted in Fig. 2 for electrons to scatter off holes and ne τeh = nh τhe
(bottom) vanishes at CN and increases as we in- guarantees momentum conservation. By evaluat-
crease βµ, as the momentum mode now carries ing the Coulomb collision integral we extract the
both heat and charge. value βτ0−1 = 0.15 (See Supplement[37]). Here,
Two-fluid model.—Since Coulomb scattering is Λe (Λh ) is the entropy per electron (hole) of the
dominant we expect a hydrodynamic description free Fermi gas characterized by βµ. An explicit
will be appropriate. Further, since the scat- expression for these quantities is given in Supple-

mentary Material[37]. experiment with us. This work was supported by
We are interested in DC transport so we may set EP/N01930X/1. Statement of compliance with
the left of these two equations to zero, and solve for EPSRC policy framework on research data: This
the velocities ue,h . The electric current is then J = publication is theoretical work that does not re-
e(ne ue − nh uh ) and the thermal current is JE = quire supporting research data.
kB T (Λe ne ue + Λh nh uh ). The advantage of the
two-fluid model is its simplicity which allows us to
obtain analytic expressions (see Supplement[37])
for transport coefficients. In the limit τph  τ0 the
[1] K. S. Novoselov, A. Mishchenko, A. Carvalho,
expressions simplify even more giving, for exam-
−1 and A. H. Castro Neto, Science 353 (2016).
ple, K̃ = (2π 2 mkB T /9)αph /(ne + nh ). In Figs. 1 [2] N. Peres, Reviews of Modern Physics 82 (2010).
and 2 we compare the results from the QBE and [3] Y. Nam, D.-K. Ki, D. Soler-Delgado, and A. F.
the two-fluid model. In both cases the agreement Morpurgo, Nature Physics 13, 1207 (2017).
is extremely good. [4] D.-K. Ki, V. I. Fal’ko, D. A. Abanin, and A. F.
Since heat is carried by the momentum mode, Morpurgo, Nano Letters 14, 2135 (2014).
the Coulomb drag between electrons and holes [5] R. T. Weitz, M. T. Allen, B. E. Feldman, J. Mar-
tin, and A. Yacoby, Science 330, 812 (2010).
will not affect it, and the thermal conductivity
[6] B. E. Feldman, J. Martin, and A. Yacoby, Nature
will mainly be limited by the phonon scattering. Physics 5, 889 (2009).
The electrical conductivity on the other hand will [7] M. T. Pettes, I. Jo, Z. Yao, and L. Shi, Nano
be limited by both phonon scattering and the Letters 11, 1195 (2011).
Coulomb drag between electrons and holes. There- [8] D.-K. Ki and A. F. Morpurgo, Nano Letters 13,
fore, the electrical conductivity is suppressed com- 5165 (2013).
pared to the thermal conductivity, leading to a [9] F. Freitag, J. Trbovic, M. Weiss, and
C. Schönenberger, Phys. Rev. Lett. 108, 076602
large Lorenz number L ≡ κ/(σT ). At charge
neutrality we find L ≈ 25(kB /e)2 which is much [10] J. Velasco Jr, L. Jing, W. Bao, Y. Lee, P. Kratz,
larger than the value of π 2 /3(kB /e)2 predicted by V. Aji, M. Bockrath, C. N. Lau, C. Varma,
the Wiedemann-Franz law. The violation of the R. Stillwell, D. Smirnov, F. Zhang, J. Jung, and
Wiedemann-Franz law for BLG has been previ- A. H. MacDonald, Nature Nanotechnology 7, 156
ously pointed out by [30] and has been experimen- (2012).
tally observed for monolayer graphene in [50]. [11] Y. Nam, D.-K. Ki, M. Koshino, E. McCann, and
A. F. Morpurgo, 2D Materials 3, 045014 (2016).
Conclusion.— In this paper we have calculated
[12] K. Bolotin, K. Sikes, Z. Jiang, M. Klima, G. Fu-
the transport properties of bilayer graphene. Our denberg, J. Hone, P. Kim, and H. Stormer, Solid
results for the electrical conductivity match the ex- State Communications 146, 351 (2008).
perimental results in [3]. From the experimental [13] S. V. Morozov, K. S. Novoselov, M. I. Katsnelson,
data we deduce that even at low temperatures, the F. Schedin, D. C. Elias, J. A. Jaszczak, and A. K.
scattering off phonons is crucial. Nonetheless, the Geim, Phys. Rev. Lett. 100, 016602 (2008).
dominant scattering mechanism is between charge [14] D. Svintsov, V. Vyurkov, S. Yurchenko, T. Otsuji,
and V. Ryzhii, Journal of Applied Physics 111,
carriers of the same species, which justifies a two
083715 (2012).
fluid approach which shows excellent agreement [15] F. M. D. Pellegrino, I. Torre, and M. Polini, Phys.
with the detailed numerical results of the QBE and Rev. B 96, 195401 (2017).
provides a simple way of calculating the transport [16] R. Bistritzer and A. H. MacDonald, Phys. Rev. B
properties analytically. One can adapt our two 80, 085109 (2009).
fluid model to different experimental setups with [17] M. Müller, J. Schmalian, and L. Fritz, Phys. Rev.
slight modifications [25]. Lett. 103, 025301 (2009).
[18] A. V. Andreev, S. A. Kivelson, and B. Spivak,
It would be interesting to test the predictions
Phys. Rev. Lett. 106, 256804 (2011).
for the thermopower and thermal conductivity in [19] B. N. Narozhny, I. V. Gornyi, M. Titov,
upcoming experiments. Our formalism can be ex- M. Schütt, and A. D. Mirlin, Phys. Rev. B 91,
panded to address a variety of other quantities 035414 (2015).
of interest, including finite frequency effects, spin [20] M. Müller and S. Sachdev, Phys. Rev. B 78,
transport, and Hall viscosity. 115419 (2008).
Acknowledgements.— We would like to acknowl- [21] V. Gantmakher and I. Levinson, Sov. Phys. JETP
133 (1978).
edge helpful discussions with Philipp Dumitrescu
[22] J. C. W. Song, D. A. Abanin, and L. S. Levitov,
and Lars Fritz. We would also like to thank the au- Nano Letters 13, 3631 (2013).
thors of [3] for kindly providing the data from that [23] D. A. Abanin, R. V. Gorbachev, K. S. Novoselov,
work to us. We are very grateful to David Soler A. K. Geim, and L. S. Levitov, Phys. Rev. Lett.
Delgado and Alberto Morpurgo for discussing their 107, 096601 (2011).

[24] M. Titov, R. V. Gorbachev, B. N. Narozhny, [50] J. Crossno, J. K. Shi, K. Wang, X. Liu,
T. Tudorovskiy, M. Schütt, P. M. Ostrovsky, I. V. A. Harzheim, A. Lucas, S. Sachdev, P. Kim,
Gornyi, A. D. Mirlin, M. I. Katsnelson, K. S. T. Taniguchi, K. Watanabe, T. A. Ohki, and
Novoselov, A. K. Geim, and L. A. Ponomarenko, K. C. Fong, Science 351, 1058 (2016).
Phys. Rev. Lett. 111, 166601 (2013).
[25] D. X. Nguyen, G. Wagner, and S. H. Simon, arXiv
e-prints (2019), arXiv:1901.07039.
[26] L. Fritz, J. Schmalian, M. Müller, and S. Sachdev,
Phys. Rev. B 78, 085416 (2008).
[27] M. Müller, L. Fritz, and S. Sachdev, Phys. Rev.
B 78, 115406 (2008).
[28] J. Lux and L. Fritz, Phys. Rev. B 86, 165446
[29] J. Lux and L. Fritz, Phys. Rev. B 87, 075423
[30] M. Zarenia, G. Vignale, T. B. Smith, and A. Prin-
cipi, arXiv e-prints (2019), arXiv:1901.05077.
[31] E. McCann and M. Koshino, Reports on Progress
in Physics 76, 056503 (2013).
[32] M. Koshino and T. Ando, Phys. Rev. B 73,
245403 (2006).
[33] E. McCann and V. I. Fal’ko, Phys. Rev. Lett. 96,
086805 (2006).
[34] Y. Nam, D.-K. Ki, D. Soler-Delgado, and A. F.
Morpurgo, Science 362, 324 (2018).
[35] P. T. Dumitrescu, Phys. Rev. B 92, 121102
[36] L. Spitzer and R. Härm, Phys. Rev. 89, 977
[37] G. Wagner, D. X. Nguyen, and S. H. Simon, Sup-
plementary Material to this paper.
[38] κ measures the heat current in the absence of elec-
trical current.
[39] We also note that in [3], in three of four samples,
σ(µ = 0) does not vary more than about 10%
over the full range of measured temperatures 10 −
[40] The dimensionless parameters for impurity scat-
tering, scattering off the boundary and phonons

are αimp = (8π 2 Z/Nf r )2 βnimp /m
β −1
αL = √mL , αph = βτphonon . Here L is the size of
the sample, Z is the charge of the impurities, r is
the relative permeability and nimp is the impurity
[41] J. K. Viljas and T. T. Heikkilä, Phys. Rev. B 81,
245404 (2010).
[42] H. Ochoa, E. V. Castro, M. I. Katsnelson, and
F. Guinea, Phys. Rev. B 83, 235416 (2011).
[43] S. S. Kubakaddi, Phys. Rev. B 79, 075417 (2009).
[44] R. Bistritzer and A. H. MacDonald, Phys. Rev.
Lett. 102, 206410 (2009).
[45] H. Suzuura and T. Ando, Phys. Rev. B 65, 235412
[46] H. Ochoa, E. V. Castro, M. I. Katsnelson, and
F. Guinea, Phys. Rev. B 83, 235416 (2011).
[47] K. M. Borysenko, J. T. Mullen, X. Li, Y. G. Se-
menov, J. M. Zavada, M. B. Nardelli, and K. W.
Kim, Phys. Rev. B 83, 161402 (2011).
[48] We note that the two-fluid model is equivalent to
performing the Spitzer-Harm calculation with two
basis functions.
[49] By considering the Lorenz force we can also in-
clude the magnetic field in this equation.

— Supplementary Material —
Transport in bilayer graphene near charge neutrality: Which scattering
mechanisms are important?
Glenn Wagner, Dung X. Nguyen and Steven Simon


In this section, we briefly summarize the formalism [25] used to calculate transport coefficients in
BLG near the neutral point. To calculate the transport coefficients, we solve the QBE equation in the
perturbed background to obtain the perturbation of distribution function fλ (p) and then use the result
to derive the linear response. The Boltzmann equation is
∂ ∂ ∂
+ vλ (k) · + eE(x, t) · fλ (k, x, t) = −Iλ [{fλi }](k, x, t), (S1)
∂t ∂x ∂k

where the group velocity of band λ is defined as

vλ (k) = ∂k λ (k). (S2)

The distribution function is expanded as

fλ (p) = fλ0 (p) + fλ0 (k)[1 − fλ0 (p))]hλ (p). (S3)

about the equilibrium Fermi distribution

fλ0 (p) = , (S4)
1+ eβ(λ (p)−µ)

where λ (p) = λp2 /(2m) is the energy in band λ. The linearized quantum Boltzmann equation can then
be derived to be
λβ 1
− Iλ [{hλi (ki )}](p) = − fλ0 (p)[1 − fλ0 (p)] eE · p − ∇T · p(λ (p) − µ) , (S5)
m T

where E is the electric field and ∇T is the thermal gradient. The collision integral includes contributions
from quasi-particle interactions, scattering on phonons and finite size effect respectively

Iλ [hλi (ki )](p) = Iλ,int [hλi (ki )](p) + Iλ,phonon [hλi (ki )](p) + Iλ,size [hλi (ki )](p). (S6)

The first term in (S6) accounts for the Coulomb scattering of charge carriers (electrons and holes) off
each other. The second term describes the scattering of charge carriers off acoustic phonons. The last
term accounts for scattering of the charge carriers off the boundary of the necessarily finite-size sample.
The linearized collision integral for scattering between charge carriers is derived in [25] as

X Z d2 k d2 q
Iλ,int [hλi (ki )](p) = −(2π) δ(λ (p) + λ1 (k) − λ2 (p + q) − λ3 (k − q))
(2π)2 (2π)2
λ1 λ2 λ3

× Nf |Tλλ1 λ3 λ2 (p, k, q)|2 − Tλλ1 λ3 λ2 (p, k, q)Tλλ 1 λ2 λ3
(p, k, k − p − q)
× [1 − fλ0 (p)][1 − fλ01 (k)]fλ02 (p + q)fλ03 (k − q)
× − hλ (p) − hλ1 (k) + hλ2 (p + q) + hλ3 (k − q) . (S7)

Nf = 4 is the number of fermion species and the matrix elements are

Tλ1 λ2 λ3 λ4 (k, k0 , q) = VC (−q)Mλ1 λ4 (k + q, k)Mλ2 λ3 (k0 − q, k0 ) (S8)

Mλλ0 (k, k0 ) = 1 + λλ0 ei(2θk0 −2θk ) (S9)
where θk is the angle between k and the x-axis. The screened Coulomb potential is given according to
[25] by
VC (q) = (S10)
q + qT F (q)
with the Thomas-Fermi screening momentum
βq 2
qT F (q) = αEM mNf (1 + ), (S11)
where αEM = e2 /0 is the electromagnetic fine structure constant. In Eq. S10 we can take the de-
nominator to be only qT F if q  αEM mNF . This condition is equivalent to the de Broglie wavelength
being much greater than the Bohr radius (∼ 15 Angstrom), which is true for the entire temperature
βq 2 −1
range we consider. We can thus take the approximation VC (q) = 2π/Nf m(1 + 12m ) . Note that the
electromagnetic fine structure constant has dropped out when calculating the screened potential. The
contribution to collision integral due to scattering off phonons is
Iλ,phonon [hλi (ki )](p) = τphonon fλ0 (p)[1 − fλ0 (p)]hλ (p) (S12)

and due to the finite size effect is

p 0
Iλ,size [hλi (ki )](p) = f (p)[1 − fλ0 (p)]hλ (p), (S13)
mL λ
where L is a characteristic length scale of the sample. We present more detailed numerical calculations
for each√ transport coefficient in the companion paper [25]. We can define the dimensionless parameters
β −1
αL = √mL and αph = βτphonon to characterize the strength of the two scattering mechanisms. Here,
αx is the ratio of the de Broglie wavelength of a particle travelling at the thermal speed divided by the
mean free path of the scattering mechanism x.


We introduce the two-fluid model, which reproduces the salient features of our numerical results. We
start with the Boltzmann equation (S5). We multiply the Boltzmann equation by λp/m and integrate
over momentum space in order to derive the evolution of the mean fluid velocities as
m e mue mue
m∂t ue = − (u − uh ) − − + eE − Λe kB ∇T (S14)
τeh τph τLe

m e muh muh
m∂t uh = (u − uh ) − − − eE − Λh kB ∇T, (S15)
τhe τph τLh
where τph is the scattering time due to collisions with phonons. τeh and τLe are the scattering times of
electrons by holes and by the boundary respectively. τhe and τLh are the scattering times of holes by
electrons and by the boundary respectively. We defined the electron and hole velocities as
R d2 p p R d2 p p
e (2π)2 m f+ (p) h (2π)2 m (1 − f− (p))
u = R d2 p , u = − R d2 p . (S16)
0 0
(2π)2 f+ (p) (2π)2 (1 − f− (p))

The scattering times off the boundary are

d2 p 3 0 0
−1 β (2π)2 p f+ (p)[1 − f+ (p)]
τLe = R d2 p 0 (S17)
2m2 L
(2π)2 f+ (p)

d2 p 3 0 0
−1 β (2π)2 p f− (p)[1 − f− (p)]
τLh = R d2 p (S18)
2m2 L 0
(2π)2 (1 − f− (p))

The coefficients Λe,h account for the fact that the average entropy per particle is ΛkB
R d2 p 2 0 0
e (2π)2 p (+ (p) − µ)f+ (p)[1 − f+ (p)]
kB T Λ = R d2 p 0 (S19)
(2π)2 f+ (p)

d2 p 2 0 0
h (2π)2 p (−− (p) + µ)f− (p)[1 − f− (p)]
kB T Λ = R d2 p (S20)
(2π)2 (1 − f− (p))

One can derive explicitly equations (S19) and (S20) from the ∇T term in Boltzmann’s equation (S5). It
is easy to show that these integrals are in fact the entropy per particle
R d2 p  0
 0 0

e (2π)2 1 − f+ (p) ln 1 − f+ (p) + f+ (p) ln f+ (p)
Λ =− R d2 p 0 , (S21)
(2π)2 f+ (p)

d2 p
R  0
 0 0

h (2π)2 1 − f− (p) ln 1 − f− (p) + f− (p) ln f− (p)
Λ =− R d2 p , (S22)
(2π)2 [1 − f− (p)]

which were used previously in [22]. The Coulomb drag term can be derived explicitly from the collision
( e
d2 p λp (1) − mn e h
τeh (u − u ) λ = +
u − uh
Iλ,int h λi (k i ) = λi βk i · (p) = mn h
e h
(2π)2 m 2 τhe (u − u ) λ=−

This allows us to calculate τeh and τhe . We perform the calculation at charge neutrality and then use
(S39) to extrapolate. τph is taken from the fit with the experimental data (the deformation potential D
being fit to experiment, as discussed in the main text). Aside from this, our model has no free parameters.
We consider the steady state ∂t ue = ∂t uh = 0 and solve the fluid equations (S14) and (S15) for ue and
uh . Using these solutions, we calculate the electric current
J = e(ne ue − nh uh ) (S24)
where e < 0 is the electron charge. The number densities calculated from the Fermi distribution are
Nf m Nf m
ne = ln(1 + eβµ ), nh = ln(1 + e−βµ ) (S25)
2πβ 2πβ
The heat current is defined as Q = JE − µe J, where JE is the energy current. Thus, the heat current is
given by
Q = kB T (Λe ne ue + Λh nh uh ) (S26)
We define the electrical conductivity σ, the thermal conductivity K and the thermo-electric coefficient
Θ by
J σ Θ E
= (S27)
Q TΘ K −∇T .
The open circuit thermal conductivity κ measuring the heat current in the absence of electrical current
and which is usually measured in experiments is given by κ = K − T Θσ −1 Θ. With these definitions we
can derive the transport coefficients from the two fluid model:
−1 −1 −1 −1
e2 ne + nh + τph [ne τLh + nh τLe + (τhe − τeh )(ne − nh )]
σ= −1 −1 −1 −1 −1 −1 −1 −1 −1 −1 −1 (S28)
m(τph + τeh + τhe + τLh + τLe + τph (τeh τLh + τhe τLe + τLe τLh ))

ekB ne Λ̃e − nh Λ̃h
Θ= −1 −1 −1 −1 −1 −1 −1 −1 −1 −1 −1 (S29)
m(τph + τeh + τhe + τLh + τLe + τph (τeh τLh + τhe τLe + τLe τLh ))

kB T Λe ne Λ̃e + Λh nh Λ̃h
K= −1 −1 −1 −1 −1 −1 −1 −1 −1 −1 −1 (S30)
m(τph + τeh + τhe + τLh + τLe + τph (τeh τLh + τhe τLe + τLe τLh ))

−1 −1 −1
Λ̃e = Λe (1 + τph (τhe + τLh )) + Λh τph τeh (S31)

−1 −1 −1
Λ̃h = Λh (1 + τph (τeh + τLe )) + Λe τph τhe (S32)

Neglecting the scattering off the boundary we find the simpler expressions

e h −1 −1 e h
σ= −1 −1 −1 n + n + τph (τhe − τeh )(n − n ) (S33)
m(τph + τeh + τhe )

ekB e e h h
Θ= −1 −1 −1 n Λ̃ − n Λ̃ (S34)
m(τph + τeh + τhe )

kB T e e e h h h
K= −1 −1 −1 Λ n Λ̃ + Λ n Λ̃ (S35)
m(τph + τeh + τhe )

−1 −1
Λ̃e = Λe (1 + τph τhe ) + Λh τph τeh (S36)

−1 −1
Λ̃h = Λh (1 + τph τeh ) + Λe τph τhe (S37)

For momentum conservation we require

ne τeh = nh τhe (S38)

We verify explicitly that the Onsager relations for the thermoelectric coefficients are satisfied if equation
(S38) is satisfied. Thus we can choose

ne + nh ne + nh
τeh = τ0 , τhe = τ0 (S39)
nh ne
By evaluating the collision integral in (S23) at CN numerically, we find

α0 ≡ βτ0−1 = 0.15 (S40)

From the experimental

p data (see main text) we extracted αph = 0.05 and from L ∼ 3µm we have
αL = 0.03/ T [K] < 0.01 for T > 10K. Therefore we find that αph , αL  α0 which means that the
scattering rate due to electron-electron collisions dominates over the electron-phonon scattering and the
scattering of electrons off the boundary. This justifies the use of the hydrodynamic theory. In the limit
τph  τ0 and again neglecting the scattering off the boundary we find the simpler expressions
e h 2
−1 (n − n )
2~ β −1 e h
σ̃ ≡ σ = α (n + n ) + α (S41)
Nf e 2 2m 0 ph
ne + nh

2~ π −1 ne − nh
Θ̃ ≡ Θ = αph (S42)
Nf ekB T 3 ne + nh

2~ 2π 2 −1 m
K̃ ≡ K = α (S43)
Nf kB 9 ph β(ne + nh )

where the tilde quantities are dimensionless as defined in the main text. In deriving the above equation
we have made use of the identity
2π m
ne Λe + nh Λh = , (S44)
3 β

which can be proven from the explicit integral expressions for Λe,h .



−1.0 −0.5 0.0 0.5 1.0

FIG. S1. Thermal conductivity coefficient K̃(βµ) (blue) and two-fluid approximation to K̃(βµ) (yellow) for
N k2 T
αph = 0.05. We have defined Kij = f2~B K̃δij . The prediction from the two-fluid model and the detailed
numerical calculation roughly differ by a constant.