HAL Id: hal-01152336
https://hal.archives-ouvertes.fr/hal-01152336
Submitted on 15 May 2015
HAL is a multi-disciplinary open access archive for the deposit and dissemination of sci- entific research documents, whether they are pub- lished or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers.
L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.
Counterflow in a doubly superfluid mixture of Bosons and Fermions
F Chevy
To cite this version:
F Chevy. Counterflow in a doubly superfluid mixture of Bosons and Fermions. Physical Review A,
American Physical Society, 2015, 91, pp.063606. �hal-01152336�
F. Chevy
11
Laboratoire Kastler Brossel, ENS-PSL Research University, CNRS, UPMC-Sorbonne Universit´ es, Coll` ege de France
In this article, we calculate the friction between two counter-flowing bosonic and fermionic super- fluids. In the limit where the boson-boson and boson-fermion interactions can be treated within the mean-field approximation, we show that the force can be related to the dynamical structure factor of the fermionic component. Finally, we provide asymptotic expressions for weakly and strongly attractive fermions and show that the damping rate obeys simple scaling laws close to the critical velocity.
PACS numbers: 67.85.-d, 67.10.-j, 67.85.Pq, 67.85.De
I. INTRODUCTION
The onset of frictionless flow in quantum fluids is prob- ably one of the most intriguing macroscopic manifes- tations of quantum mechanics. Recent experiments on Bose-Fermi superfluid mixtures gave a new twist to this old question by probing the critical velocity of a super- fluid counterflow [1]. When two miscible superfluids flow through each-other, Castin et al. suggested a generaliza- tion of the celebrated Landau criterion where superfluid- ity is destroyed by the shedding of a pair of elementary excitations in the two systems [2]. Later-on, this scenario was supported by the study of the lifetime of the quasi- particles [3, 4] or by the calculation of the hydrodynamic spectrum [5].
The results put forward in [2] were based on heuristic arguments and were focusing on the velocity threshold above which the counter-flow is damped. In this work, we provide a full microscopic treatment of the friction in a superfluid counterflow and we determine the explicit velocity dependence of the damping force above the crit- ical velocity. Assuming that the boson-boson and boson- fermion interactions can be treated within the mean-field approximation, we show that the force can be related to the dynamic structure factor of the fermionic superfluid.
Although the general expression of the structure factor of an attractive Fermi gas is not known exactly in the crossover between the Bardeen-Cooper-Schrieffer (BCS) and molecular Bose-Einstein Condensate (BEC) regimes, we provide asymptotic expressions in the limits of weak and strong interactions where the fermionic component behaves respectively as an ideal Fermi gas and an hydro- dynamic Bose-Einstein condensate of dimers [6]. We find that close to the critical velocity, the force obeys simple scaling laws, with exponent depending on the value of the fermion-fermion scattering length.
II. GENERAL SETTING
We consider a mixture between a fermionic superfluid and a weakly interacting Bose-Einstein condensate mov- ing at velocities V
fand V
brespectively. The Hamiltonian
of the system can be written as
H b = H b
b+ H b
f+ H b
bf, (1) where H b
b,fare the Hamiltonians of the bosons and the fermions and H b
bfdescribes the coupling between the two species. In the mean-field approximation, we have
H b
bf= g
bfZ
d
3r X
σ
ϕ(r) b
†ϕ(r) b ψ b
σ†(r) ψ b
σ(r), (2) where ψ b
σis the field operator of spin σ fermions, and ϕ b that of the bosons. Assuming that the BEC can also be described within the mean-field approximation, then the bosonic field operator can be expanded over the Bogoli- ubov modes
ϕ(r) = b e
ikb·r"
√ n
b+ 1
√ Ω X
q
u
qb b
qe
iq·r− v
qb b
†qe
−iq·r# , (3) where Ω is the quantization volume, ~ k
b= mV
b, n
bis the density of bosons and the (u
q, v
q) are the usual Bogoli- ubov coefficients. The exponential prefactor describes the Galilean boost associated with the the motion of the condensate (see Appendix A). In addition, due to the motion of the BEC, the energies of the Bogoliubov modes are shifted by Doppler effect and their spectrum is given by
E
q,b= E
q,b(0)+ ~ q · V
b, (4) where E
q,b(0)= ~ c
bq p
1 + q
2ξ
b2/2 is the Bogoliubov spec- trum (here c
bis the sound velocity and ξ
bis the healing length) [7].
Expanding H b
bfto first order in fluctuations of the bosonic field, we have
H b
bf= H b
bf(0)+ H b
bf(1)+ H b
bf(1)†+ ... (5) with
H b
bf(0)= g
bfn
bZ
d
3r X
σ
ψ b
σ†ψ b
σ(r), (6)
2 and
H b
bf(1)= g
bfr n
bΩ Z
d
3r X
q,σ
(u
q− v
q)e
−iq·rψ b
σ†(r) ψ b
σ(r)b b
†q. (7) H b
bf(0)being proportional to the number of fermions, it is a constant of motion, and can therefore be gauged away.
H b
bf(1)and H b
bf(1)†respectively create and destroy Bogoli- ubov excitations in the BEC. In the weak coupling limit we can expand the damping rate in power of g
bf. To first order, only H b
bf(1)plays a role and using Fermi’s Golden Rule the rate of creation of excitations of momentum ~ q in the BEC and in an eigenstate |α
fi of the fermionic Hamiltonian H b
fis given by
Γ(α, q) = g
bf2n
bΩh (u
q− v
q)
2δ(E
α,f+ E
q,b)×
hα
f| Z
d
3r X
σ
ψ b
σ†ψ b
σ(r)e
−iq·r|0
fi
2
, (8)
where E
α,fis the energy of the state |α
fi relatively to that of the ground state |0
fi of the moving fermionic superfluid. Since the creation of a Bogoliubov excitation imparts a momentum ~ q to the BEC, the force acting on the bosons can be written as
F = X
α,q
~ qΓ(α, q) (9)
= 2π g
bf2n
bΩ
X
q
q(u
q− v
q)
2S(−q, −E
q,b), (10) where we introduced the dynamical structure factor of the Fermi gas
S(q, E) = X
α
δ(E−E
α,f)
hα
f| Z
d
3r X
σ
ψ b
σ†ψ b
σ(r)e
iq·r|0
fi
2
. (11) Note that the expression for the force is galilean invariant and depends only on the relative velocity V
b−V
f. Indeed, one can show that the dynamical structure factor S of a system moving at velocity V
fis the related to the one of a system at rest (denoted as S
st) by
S(q, E) = S
st(q, E − ~ q · V
f) (12) (this result is a direct consequence of Eq. (A12) giving the energy in the moving frame, and the fact that the operator R
d
3r P
σ
ψ b
†σψ b
σ(r)e
iq·rgives a kick ~ q to the system). Inserting this expression in Eq. (10) and using the expression for the Bogoliubov spectrum of a moving BEC, Eq. (4) we have for the force
F = 2π g
2bfn
bΩ
X
q
q(u
q− v
q)
2S
st(−q, −E
(0)q,b− ~ q · V ), (13)
with V = V
f−V
b. Finally, using the explicit form of the Bogoliubov coefficients, we have
(u
q− v
q)
2= ε
q,bE
q,b(0)(14)
with ε
q,b= ~
2q
2/2m
b, hence F = 2π g
2bfn
bΩ X
q
q ε
q,bE
q,b(0)S
st(−q, −E
q,b(0)− ~ q · V ). (15) Eq. (15) is the main result of this paper and shows that the friction between the two superfluids is directly related to the dynamic structure factor of the fermionic component of the system. The dynamical structure fac- tor describes the response of a given system to a sinu- soidal perturbation and can be related to two-body cor- relations. In strongly correlated gases, it was measured using Bragg spectroscopy [8] and was used to determine the value of Tan’s contact parameter describing the tail of the momentum distribution [9].
The properties of strongly correlated Fermi systems are notoriously hard to calculate, and approximate ex- pression for the dynamic structure factor have been ob- tained using the approximate methods [10, 11]. However, its general properties can be summarized on the sketch of Fig. 1. At low momenta, the excitation spectrum is dom- inated by a phonon branch E = ~ c
fk, with c
fthe sound velocity of the fermionic superfluid. Since the ground state defines the energy zero, S
stvanishes for E < 0.
However, the presence of a pairing gap implies of a re- gion where S
stvanishes at positive energies. Landau’s critical velocity v
Lis defined as the highest value of v such that S
st(E, q) vanishes for all E ≤ vq (see Fig. 1).
In the BEC regime of the BEC-BCS crossover, we have v
L= c
F, while in the BCS regime, Landau’s velocity is driven by pair breaking excitations and we have v
Lc
f. These general features implies that if for all q we have
−E
q,b(0)− ~ q · V < qv
L, the force is zero. To get some damping, there must be values of q for which −E
(0)q,b−
~ q · V ≥ ~ qv
L. Using the general Landau argument, we get that |V | ≥ v
L+ min
q(E
q,b(0)/ ~ q) = v
L+ c
b. Note that this bound sets a necessary, but not sufficient, condition on the relative velocity. As a consequence v
L+c
bis only a lower bound for the real critical velocity.
Another consequence of this inequality is that the val- ues of q contributing to the integral are bounded. Indeed, from the inequality E ≤ ~ qv
L, we get that
qξ
b≤ p
2x(x + 2),
where xc
b= V − c
b− v
L. In other words, when the relative velocity gets close to c
b+ v
L, the damping is dominated by long wave-length excitations.
In what follows, we compute the force in the deep BEC
and BCS limits where analytical expressions can be ob-
tained.
0.0 0.5 1.0 1.5 2.0 0.0
0.5 1.0 1.5 2.0
q @ a.u. D
E 2 D
FIG. 1: Sketch of the structure factor of an interacting Fermi gas: S
stvanishes outside the gray area and ∆ is the excitation gap. Dotted line: low momentum phonon branch E = ~ c
fq.
Dashed line: Landau’s critical velocity E = ~ v
Lq below which the structure factor vanishes. Note that this sketch corre- sponds to a weakly interacting system where Landau’s veloc- ity is fixed by pair breaking, and v
L< c
f. In the molecular regime of the crossover, v
L= c
f[2, 12].
III. BCS LIMIT
Let’s consider first the deep BCS limit where the fermions can be described as an ideal gas. Since there is a vanishingly small gap in the fermionic spectrum the lower bound derived above is actually saturated and we have V
c= c
b. Working close to c
b, the relevant values of q much smaller than k
F, and we can therefore replace the expression of the dynamical structure factor by its low-energy approximation (see for instance [13], and Ap- pendix B)
S
st(q, E) = m
2fΩE 2π
2~
4q ,
this replacement being valid as long as the condition
k
Fξ s
2
V
2− c
2bc
2bis fulfilled.
Replacing the discrete sum over q by an integral, one gets
F
Ω = − 2g
2bfn
bm
2f(2π)
4~
4Z
d
3q qε
b,qqE
b,q(0)(E
b,q(0)+ ~ q·V )Θ(−E
b,q(0)− ~ q·V ).
(16) where Θ is Heaviside’s step function. The integral can be computed analytically and taking V
0= V /c
b, the force becomes
F
Ω = − g
2bfn
bm
2f(2π)
3m
b~
2ξ
5I(V
0), (17) with
I(V
0) = 1 105 √
2V
02h
105V
03sinh
−1p
V
02− 1 + p V
02− 1 6V
06− 39V
04− 80V
02+ 8 i
. (18)
Close to V
0= 1, this expression can be expanded as I(V
0) ' 128
315 (V
0− 1)
9/2+ ... (19) Even though this scaling is strictly-speaking obtained for a non-interacting Fermi gas, it can be extended to the case of a BEC moving inside an interacting Fermi gas in its normal phase when it can be described within the Fermi Liquid Theory. Indeed, close the critical velocity, the damping mechanism described involves only low en- ergy excitations of the system which behave as fermionic quasiparticles with renormalized physical parameters. In particular, the scaling S(q, E) ∝ E/q is still valid [14].
IV. HYDRODYNAMIC LIMIT
In the BEC limit of the cross-over, the excitations of the fermionic superfluid are dominated by phonons of energy E
f,q. In this case, the dynamical structure factor can be approximated by a Dirac function and we have in this case
S
st(q, E) = Aδ(E − E
f,q(0)) (20) where the normalization constant A is fixed by the f-sum rule [7, 15] and is given by
A = N
fε
q,fE
f,q(0). (21)
In this limit, we obtain for the force F = 2πg
2bfn
bn
fX
q
q ε
q,bε
q,fE
q,b(0)E
−q,f(0)δ(E
(0)q,b+ E
−q,f(0)+ ~ q · V ).
(22) Using the usual Landau argument, we see that the force vanishes if the velocity is smaller than a critical velocity V
cgiven by
V
c= min
q
E
q,b(0)+ E
−q,f(0)~ q
!
. (23)
4 In case of linear dispersion relation, the critical velocity
is simply
V
c= c
b+ c
f, (24) where c
f,bare the sound velocities of the fermions and the bosons.
To calculate explicitly the force, we assume that both the bosons and the fermions follow the Bogoliubov dis- persion relation, an assumption valid in the far-BEC limit of the BEC-BCS crossover. Taking ξ
b,fthe healing lengths of the two superfluids, we have after performing the angular integral
F
Ω = − g
2bfn
bn
f~ 8πm
fm
bc
bc
fV × Z
qM0
q
4dq
cb
V
p 1 + q
2ξ
b2/2 +
cVfq
1 + q
2ξ
f2/2 q (1 + q
2ξ
b2/2)(1 + q
2ξ
f2/2)
(25) with q
Mgiven by the condition
c
bq
1 + q
2Mξ
b2/2 + c
fq
1 + q
M2ξ
f2/2 = V (26) For V ' V
c, q
Mis vanishingly small, and is given by
q
M' 2
s V − V
cc
bξ
b2+ c
fξ
f2, hence
F
Ω ' − 4g
2bfn
bn
f~ 5πm
fm
bc
bc
fV
cV − V
cc
bξ
2b+ c
fξ
2f!
5/2(27)
V. CONCLUSION
In this work we have shown that the damping between two bosonic and fermionic superfluids could be related to the dynamical structure factor of the fermions. In the weakly and strongly attractive limits, we find a power law dependence of the damping force vs velocity close to the critical velocity. Like previous findings [3], we find that the exponent varies across the BEC-BCS crossover and is different from the scaling F ∝ (V
0−1)
2found for the drag of a particle in a Bose-Einstein condensate [16]. These different scalings highlight the many-body nature of the two counter-flowing systems and the role of the collective excitations in the damping mechanism. Our prediction can be directly tested by measuring the damping of the relative oscillations of two superfluids, as presented in [1]
for a unitary fermions. In particular, comparison with Eq. 15 provides a quantitative test of theoretical pre- dictions for the structure factor of a strongly correlated fermionic superfluid.
Acknowledgments
The authors thank X. Leyronas, C. Salomon, Y.
Castin, I. Ferrier-Barbut, S. Laurent, M. Delehaye and S.
Stringari for stimulating discussions. This work was sup- ported by R´ egion ˆIle de France (IFRAF) and European Union (ERC grant ThermoDynaMix).
Appendix A: Bogoliubov Transformation in a Moving Frame
Consider a ensemble of N particles described by a state
|ψi. The action of a velocity V boost is described by the operator
U b = exp
i
N
X
j=1
k · r b
i
, (A1) with k = mV / ~ . In second quantization, this operator is represented by
U b = exp
i Z
d
3r
0ϕ(r b
0)
†ϕ(r b
0)k · r
0. (A2) Define ϕ b
0(r) the field operator in the moving frame.
By definition, we must have for |ψ
0i=1,2i = U b |ψ
i=1,2i, hψ
10| ϕ b
0(r)|ψ
02i = hψ
1| ϕ(r)|ψ b
2i, hence
hψ
1| U b
†ϕ b
0(r) U b |ψ
2i = hψ
1| ϕ(r)|ψ b
2i. (A3) This equation being valid for any |ψ
ii, we must have U b
†ϕ b
0(r) U b = ϕ(r) hence b
ϕ b
0(r) = U b ϕ(r) b U b
†. (A4) ϕ b
0can be calculated using the Campbell-Hausdorff for- mula. Taking U b = exp( T b ), we have indeed
ϕ b
0(r) = ϕ(r) + [ b T , b ϕ(r)] + b 1
2 [ T , b [ T , ϕ(r)]] + b ... (A5) The commutator [ T , b ϕ(r)] is readily calculated and yields b [ T , b ϕ(r)] = b −ik · r ϕ(r). After resumming the Campbell- b Hausdorff formula, we obtain
ϕ b
0(r) = e
−ik·rϕ(r). b (A6) In the moving frame, the BEC is stationary and we can therefore decompose ϕ
0(r) using a Bogoliubov transform
ϕ b
0(r) = √
n
0+ 1
√ Ω
X
q
h u
qe
iq·rb b
q− v
qe
−iq·rb b
†qi . (A7) We thus have in the lab frame
ϕ(r) = b e
ik·rϕ b
0(r) (A8)
= e
ik·r"
√ n
0+ 1
√ Ω X
q
u
qe
iq·rb b
q− v
qe
−iq·rb b
†q#
.
(A9)
The eigenenergies of the Bogoliubov excitations of the moving Bose-Einstein condensate are obtained by apply- ing the Galilean boost operator U b to the eigenstates |βi
bof the condensate at rest. Taking |β
0i
b= U b |βi
b, and consider the quantum many-body Hamiltonian
H b
b= − Z
ϕ(r) b
†~
2∇
22m ϕ(r)d b
3r +
Z
V (r − r
0) ϕ(r) b
†ϕ(r b
0)
†ϕ(r b
0) ϕ(r) b
†d
3rd
3r
0. (A10) By using Eq. A6), we see readily that if |βi
bis a com- mon eigenstate of H b
band the momentum and particle number operators, then |β
0i
bis an eigenstate of H b
bfor the eigenvalue
E
0β= E
β+ ~ q
β· V + N
bmV
22 , (A11)
where ~ q
βis the momentum of state |βi
b. Compared to the energy E
00of the boosted ground-state Bose-Einstein condensate, the energy of the excitation |β
0i
bis therefore E
β0− E
00= E
β+ ~ q
β· V , (A12) and is simply Doppler-shifted.
Appendix B: Dynamic structure factor of the ideal Fermi gas.
In this appendix, we rederive the dynamical structure factor for zero-temperature, non-interacting fermions. A more general derivation including finite temperature ef- fects can be found in [13].
The dynamic structure factor is given by the matrix elements of the excitation operator F b
q= P
σ
R d
3re
iq·eψ b
σ(r)
†ψ b
σ(r). Expanding the field op- erators over the plane wave basis as ψ b
σ(r) = P
k
e
ik·rb c
k,σ/ √
Ω, where Ω is a quantization volume, yields
F b
q= X
k,σ
b c
†k+q,σb c
k,σ.
In other words, the operator F b
qkicks one particles of the system and gives it a momentum q. The states
|ni excited by F b
qfrom the ground state |0i correspond to particle-hole pairs of momenta k (hole) and k + q (particle). The energy of such an excitation is E
n= ε
k+q− ε
k= ~
2q
2/2m + ~
2k · q/m and we can write the structure factor as
S(q, E) = 2 Z
k<kF,|k+q|>kF
d
3kΩ (2π)
3δ
E − ~
2q
22m − ~
2k
zq m
,
Κ kz
Κ1
Kz
q
Κ kz
Κ1
Κ2
Kz
q
Κ kz
Κ1 Κ2
Kz
q
FIG. 2: Construction of the dynamic structure factor in mo- mentum space. The shaded area represents the Fermi sphere.
Energy-momentum conservation fixes the initial k
zto a value K
z(see text). Since the initial state must lie inside the Fermi Sea, we must have κ ≤ κ
1= p
k
2F− K
z2. Moreover, the final momentum must be above the Fermi surface, hence κ ≥ p
k
2F+ (K
z+ q)
2. The thick solid line represents the mo-
menta contributing to the dynamical structure factor. From
top to bottom. Upper panel, case (1): K
z+ q larger than k
f.
The final momentum is always above the Fermi surface. Mid-
dle panel, case 2: K
z+ q is lower than k
F, with κ
2≤ κ
1. The
permitted values of k are located in the ring κ
2≤ κ ≤ κ
1.
Lower panel, case 3. For κ
2≥ κ
1, the contribution to the
structure factor vanishes.
6 where the z axis is chosen along q and where the factor
2 comes from the two spin states.
We perform first the integral over k
z. Take K
z(q, E) = (E −
~2m2q2)m/ ~
2q. The delta-function selects k
z= K
z(q, E) and taking κ = (k
x, k
y) we have
S (q, E) = 2m
~
2q Z
|Kz|<kF,|k+q|>kF