Pis’ma v ZhETF, vol. 111, iss. 3, pp. 173 - 174
© 2020
February 10
Joule-Thomson cooling in graphene
K.Zarembo1)
Nordita, Kungliga Tekniska Högskolan Royal Institute of Technology and Stockholm University, SE-106 91 Stockholm, Sweden
Niels Bohr Institute, Copenhagen University, 2100 Copenhagen, Denmark
Institute of Theoretical and Experimental Physics, 117218 Moscow, Russia
Submitted 31 August 2019
Resubmitted 20 December 2019
Accepted 22
December 2019
DOI: 10.31857/S0370274X20030066
Collective effects of electron interactions may pre-
lation ǫ + P = µn + T s, where s is the entropy density,
vail over impurity scattering in a very clean sample.
then yields
Electrons then flow like a viscous fluid [1], as observed
A+T∂ŝ∂µ
α=-
,
(2)
in graphene in a certain range of temperatures [2-7].
Aŝ+ T∂ŝ
∂T
Spectacular consequences of hydrodynamic electronic
where ŝ = s/n is specific entropy and A = 1. The ra-
transport include negative local resistivity [2, 3, 4, 7],
tionale to introduce a special notation for a constant
violation of ballistic bound on conductance [2, 6, 8, 9]
equal to one is that a more accurate calculation gives
(Gurzhi effect [1]), breakdown of Wiedemann-Franz law
A ≈ 2/3, while the overall functional form of the cooling
[5, 9, 10], and negative magnetoresistance [9, 11]. An-
coefficient remains the same.
other possible manifestation of hydrodynamic mode of
Electrons in graphene form a 2d Fermi gas with lin-
transport, discussed here, is cooling of electron fluid that
ear dispersion relation, and their pressure is given by
passes through a narrow constriction. In the solid state
∑ (
)
setting electric current normally generates heat, and
d2p
qµ-vF |p|
P = 4T
ln
1+ e
T
,
(3)
cooling of electron flow may look counterintuitive, but in
(2πℏ)2
q=±
fluid mechanics this phenomenon, the Joule-Thomson
(JT) effect [12], is well known and underlies a widely
where vF is the Fermi velocity, q labels particles/holes,
used method of gas refrigeration.
and the overall factor of four takes into account valley
Consider two strips of graphene connected by a nar-
and spin degeneracy. We tacitly assume that holes and
row bridge and subject to a constant voltage δU. As-
electrons are in thermodynamic equilibrium, which is
suming that one strip is kept at temperature T1 and
not a good approximation at the neutrality point, where
denoting electron temperature in the other by T2, the
our derivation is not applicable.
cooling/heating effect can be characterized by the tem-
The rest of thermodynamic quantities can be calcu-
perature drop δT = T1 - T2 relative to the potential
lated from dP = ndµ + sdT . When applied to (2) the
difference δµ = µ1 - µ2 = eδU:
standard thermodynamic machinery gives
1
3AFF
µ
δT = αδµ,
(1)
=
- ξ,
ξ=
,
(4)
α
A+2-3FF′′
T
F′2
where
where µ1 and µ2 are chemical potentials on the two sides
F (ξ) = Li3(- eξ) + Li3(- e).
(5)
of the bridge. The dimensionless coefficient α can take
either sign and is defined such that α > 0 corresponds
In the two limiting cases we get:
to cooling.
3Aµ
T
Textbook derivations of the JT effect start with the
α
µ≫T
,
α
µ≪T≃ -
(6)
2(1 + A)π2T
(1 + A)µ
enthalpy conservation: δ[(ǫ + P )/n] = 0, where ǫ and P
are the energy density and pressure of the electron fluid
The JT effect results in cooling in the Fermi liquid
and n is the charge carrier density. Thermodynamic re-
regime (µ ≫ T) and in heating in the Dirac fluid case
(µ ≪ T ). The sign of the effect changes at an inversion
point, which for the physical case of A = 2/3 lies at
1)e-mail: zarembo@nordita.org
µinv = 3.32T.
Письма в ЖЭТФ том 111 вып. 3 - 4
2020
173
174
K.Zarembo
The thermodynamic derivation of the JT effect rests
where L is the total size of the system.
on enthalpy conservation, but enthalpy production in
In conclusion, hydrodynamic nature of electron flow
the moving electron fluid cannot be neglected. The
in biased graphene leads to JT cooling when the cur-
viscous heating turns out to be the prime source of
rent is forced through a narrow constriction. Cooling
enthalpy production and leads to order one effects,
occurs in the Fermi liquid regime, for sufficiently large
while Ohmic resistivity only gives small corrections. The
charge imbalance or at sufficiently low temperatures.
derivation is based on the hydrodynamic theory of elec-
For lower chemical potentials the JT effect results in
tron transport in the commonly used Stockes approxi-
heating, which is most pronounced in the Dirac liq-
mation:
uid regime at µ ≪ T . Although similar to conventional
η
Joule heating the mechanics behind this effect is quite
η∂2vi -
vi = ∂iP,
ivi = 0,
(7)
λ2
different, in particular the temperature increment is lin-
where vi is the electron velocity, η is the shear viscosity
ear in applied voltage and not quadratic. It is necessary
to keep in mind however that our derivation becomes
of the electron fluid and λ is the momentum-relaxation
invalid close to the neutrality point.
length.
The solution is known explicitly when λ = ∞ [8]:
Full text of the paper is published in JETP Letters
4ηu
z
journal. DOI: 10.1134/S0021364020030030
P =P0 -
Im
,
a
a2
-z2
4
2uy
z
vx =
Re
,
1.
R. Gurzhi, Sov. Phys. JETP 17, 521(1963).
a
a2
-z2
2.
I. Torre, A. Tomadin, A. K. Geim and M. Polini, Phys.
4
Rev. B 92, 165433 (2015); 1508.00363.
2
2u
a
2uy
z
vy =
Re
-z2 -
Im
(8)
3.
L. Levitov and G. Falkovich, Nat. Phys. 12, 672 (2016);
a
4
a
a2
-z2
1508.00836.
4
4.
D. A. Bandurin, I. Torre, R. K. Kumar, M. Ben Shalom,
Here z = x+iy, a is the width of the bridge, and u is the
A. Tomadin, A. Principi, G. H. Auton, E. Khestanova,
maximal velocity attained by the fluid. The square root
K. S. Novoselov, I. V. Grigorieva, L. A. Ponomarenko,
is analytic on the complex plane with a semi-infinite cut
A. K. Geim, and M. Polini, Science 351, 1055 (2016);
representing the constriction.
1509.04165.
The flow is sustained by the pressure drop:
5.
J. Crossno, J. K. Shi, K. Wang, X. Liu, A. Harzheim,
8ηu
A. Lucas, S. Sachdev, P. Kim, T. Taniguchi, K. Watan-
δP =
(9)
a
abe, T. A. Ohki, and K. C. Fong, Science 351, 1058
(2016); 1509.04713.
The entropy production rate in the moving fluid is cal-
culated according to
6.
R. K. Kumar, D. A. Bandurin, F. M. D. Pellegrino
)
et al. (Collaboration), Nat. Phys. 13,
1182
(2017);
2
η
(v
Π2
viiŝ =
+
,
Πij = ∂ivj + ∂jvi.
(10)
1703.06672.
nT λ2
2
7.
D. A. Bandurin, A. V. Shytov, L. S. Levitov, R. K. Ku-
Integrating the rate along the midflow (and setting
mar, A. I. Berdyugin, M. Ben Shalom, I. V. Grigorieva,
A. K. Geim, and G. Falkovich, Nat. Commun. 9, 4533
λ = ∞) gives:
(2018); 1806.03231.
2ηua
dy y2
16ηu
2
8.
H. Guo, E. Ilseven, G. Falkovich, and L. S. Levi-
-δŝ =
=
=
δP.
(11)
(a2
)5
tov, Proc. Natl. Acad. Sci. USA 114, 3068 (2017);
nT
2
3anT
3nT
+y2
-∞
4
1607.07269.
The relation δP = nδµ + sδT then results in the same
9.
J. Gooth, F. Menges, N. Kumar, V. Sü, C. Shekhar,
formula (2) for the JT coefficient, but with A = 2/3.
Y. Sun, U. Drechsler, R. Zierold, C. Felser, and B. Gots-
mann, Nat. Commun. 9, 4093 (2018); 1706.05925.
When momentum relaxation is taken into account,
the coefficient A starts to depend on the dimensionless
10.
A. Lucas, J. Crossno, K. C. Fong, P. Kim, and
ratio a/λ. Assuming that this ratio is small, a ≪ λ, cor-
S. Sachdev, Phys. Rev. B 93, 075426 (2016); 1510.01738.
rections due to momentum relaxation and Ohmic heat-
11.
A. Lucas, R. A. Davison, and S. Sachdev, Proc. Natl.
ing appear to be quadratic in the small parameter a/λ,
Acad. Sci. USA 113, 9463 (2016); 1604.08598.
but are logarithmically enhanced:
12.
W. Thomson and J. P. Joule, Philos. Trans. R. Soc. Lon-
(
)
don 143, 357 (1853).
2
a2
L
A=
1+
ln
,
(12)
3
32λ2
λ
Письма в ЖЭТФ том 111 вып. 3 - 4
2020