BCS Wave-Function Approach to the BEC-BCS Crossover

of Exciton-Polariton Condensates
Tim Byrnes,
1,2
Tomoyuki Horikiri,
3
Natsuko Ishida,
1,4
and Yoshihisa Yamamoto
1,3
1
National Institute of Informatics, 2-1-2 Hitotsubashi, Chiyoda-ku, Tokyo 101-8430, Japan
2
Institute of Industrial Science, University of Tokyo, 4-6-1 Komaba, Meguro-ku, Tokyo 153-8505, Japan
3
E. L. Ginzton Laboratory, Stanford University, Stanford, California 94305, USA
4
Department of Information and Communication Engineering, The University of Tokyo,
7-3-1 Hongo, Bunkyo-ku, Tokyo 113-8656, Japan
(Received 13 May 2010; published 26 October 2010)
The crossover between low and high density regimes of exciton-polariton condensates is examined
using a BCS wave-function approach. Our approach is an extension of the BEC-BCS crossover theory for
excitons, but includes a cavity photon field. The approach can describe both the low density limit, where
the system can be described as a Bose-Einstein condensate (BEC) of exciton-polaritons, and the high
density limit, where the system enters a photon-dominated regime. In contrast to the exciton BEC-BCS
crossover where the system approaches an electron-hole plasma, the polariton high density limit has
strongly correlated electron-hole pairs. At intermediate densities, there is a regime with BCS-like
properties, with a peak at nonzero momentum of the singlet pair function. We calculate the expected
photoluminescence and give several experimental signatures of the crossover.
DOI: 10.1103/PhysRevLett.105.186402 PACS numbers: 71.36.+c, 67.10.Àj, 74.78.Na
In recent years there is an increasing consensus that a
Bose-Einstein condensate (BEC) of exciton polaritons has
been realized experimentally [1,2]. For exciton BECs, it is
well known that the system crosses over into a BCS state
of weakly correlated electrons and holes at high density
[3,4]. Such a crossover was realized with fermionic atoms
by varying the attractive interaction between using
Feshbach resonances [5], providing an experimental dem-
onstration of the crossover. A natural question is, then,
does a BEC-BCS crossover also occur for exciton-
polaritons? Littlewood and co-workers have examined
this question and have predicted that with increasing
density the system transitions from a BEC state to a
photon BEC state [6–9]. In an intermediate density regime
under suitable conditions they predict a BCS-like regime
[7]. The model that they deal with is a model of non-
interacting excitons coupled to a common photonic cavity.
In this model, the excitons do not contain an internal
electron-hole structure and are treated as spins localized
on lattice sites. Our purpose here is to include the electron
and hole components as well as their Coulomb interaction.
By doing so we find that several new effects are present
which have implications on experimentally observable
quantities.
The Hamiltonian we consider is
H ¼ H
kin
þH
Coul
þH
ph
˵n,
H
kin
¼

k
_
@
2
k
2
2m
e
e
y
k
e
k
þ
@
2
k
2
2m
h
h
y
k
h
k
_
,
H
Coul
¼
1
2

k,k
0
,q
VðqÞ½e
y
kþq
e
y
k
0
Àq
e
k
0 e
k
þh
y
kþq
h
y
k
0
Àq
h
k
0 h
k
À2e
y
kþq
h
y
k
0
Àq
h
k
0 e
k
Š,
H
ph
¼

k
½e
y
k
h
y
Àk
o þo
y
h
Àk
e
k
Š þwo
y
o, (1)
where e
k
and h
k
are the fermion annihilation operators for
electrons and holes with momentum k, VðqÞ ¼
e
2
2eq
is the
Coulomb interaction, m
e
and m
h
are the electron and hole
effective masses, e is the electron charge, e is the permit-
tivity, o is the cavity photon annihilation operator, w is the
cavity photon energy, is the coupling strength of the
cavity photon to the electrons and holes, and n ¼ o
y
o þ

k
½e
y
k
e
k
þh
y
k
h
k
Š is the total number of particles per unit
area. We introduce a chemical potential µ into the
Hamiltonian in order to fix the density.
We use a BCS wave-function ansatz of the form
jÈi ¼ exp½¬o
y
À¬
2
,2Š

k
½u
k
þ¡
k
e
y
k
h
y
Àk
Šj0i, (2)
where u
2
k
þ¡
2
k
¼ 1 in analogy to the BCS wave function
used for excitons [3,4]. An additional coherent state pho-
tonic term is included of the same form as in Ref. [6]. The
photon density is n
ph
¼ ¬
2
and the electron-hole density is
n
eh
¼

k
¡
2
k
, giving a total particle density n ¼ n
ph
þn
eh
.
The BCS wave function (2) is equivalent to the solution
of the Hamiltonian (1) by mean field theory [10]. Following
the mean field derivation, we assume a pairing Hamiltonian
in Eq. (1) and restrict k
0
¼ Àk in the attractive Coulomb
interaction. In the repulsive Coulomb terms we assume a
PRL 105, 186402 (2010)
P HYS I CAL RE VI E W L E T T E RS
week ending
29 OCTOBER 2010
0031-9007,10,105(18),186402(4) 186402-1 Ó 2010 The American Physical Society
Hartree Fock approximation, and restrict q ¼ k
0
Àk.
Hartree terms (corresponding to q ¼ 0), are not present
due to the assumption of charge neutrality [4]. Expanding
the operators e
y
k
e
k
, h
y
k
h
k
, e
y
k
h
y
Àk
, and o around their mean
values and keeping only linear terms gives
H¼E
0
þDðoþo
y
Þ þðwÀµÞo
y

k
(
k
ðe
y
k
e
k
þh
y
k
h
k
Þ
À

k
ðÁ
k
À¬Þðe
y
k
h
y
Àk
þh
Àk
e
k
Þ (3)
where (
k
¼
@
2
k
2
2m
ÀµÀX
k
, X
k
¼

k
0 Vðk Àk
0
Þhe
y
k
0
e
k
0 i,
Á
k
¼

k
0 Vðk Àk
0
Þhe
y
k
0
h
y
Àk
0
i, D ¼

k
he
y
k
h
y
Àk
i, E
0
¼

k
½X
k
he
y
k
e
k
i þÁ
k
he
y
k
h
y
Àk
iŠ À2D¬ and we have taken
all expectation values as real, 2,m ¼ 1,m
e
þ1,m
h
,
¬ ¼ hoi, and he
y
k
e
k
i ¼ hh
y
Àk
h
Àk
i due to charge neutrality.
The photonic part may be diagonalized by introducing the
operator B
y
¼ o
y
Àc, and demanding that the off-
diagonal terms vanish. This gives the condition
¬ ¼ À
D
w˵
. (4)
The remaining part of the Hamiltonian may be diagonal-
ized by a transformation e
k
¼ u
k
¡
k0
þ¡
k
¡
y
k1
, h
y
k
¼
À¡
k
¡
k0
þu
k
¡
y
k1
. Demanding that the off-diagonal terms
disappear, we obtain the Hamiltonian
H ¼ e
0
þðwÀµÞB
y
B þ

k
E
k
ð¡
y
k0
¡
k0
þ¡
y
k1
¡
k1
Þ, (5)
where e
0
¼

k
½(
k
ÀE
k
þÁ
k
he
y
k
h
y
Àk
i þX
k
he
y
k
e
k
iŠ þ

2
D
2
,ðwÀµÞ. The parameters u
k
and ¡
k
satisfy the
standard BCS algebra [10]
he
y
k
e
k
i ¼ ¡
2
k
¼
1
2
_
1 À
(
k
E
k
_
, (6)
he
y
k
h
y
Àk
i ¼ u
k
¡
k
¼
Á
k
À¬
2E
k
, (7)
E
k
¼
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
(
2
k
þðÁ
k
À¬Þ
2
_
. (8)
The gap energy E
min
k
is defined to be the value of E
k
minimized over all momenta k. In our numerical results,
we use units such that the momentum is measured in units
of 1,o
B
, and densities are measured in units of 1,o
2
B
, where
the Bohr radius is o
B
¼ 4re@
2
,me
2
. The energy is mea-
sured in units E
0
¼ e
2
,4reo
B
(SI units used throughout).
In the low density limit with no photon field, the exciton
energy is E,E
0
¼ À1. Zero detuning therefore corresponds
to a photon energy of w,E
0
¼ À1 (the photon energy is
negative since we measure energies relative to the band
gap energy). Equations (6)–(8) with (4) are solved self-
consistently to obtain our results.
In Fig. 1(a) the occupation probability ¡
2
k
of electron-hole
pairs is shown. In the low density limit (µ % À1.4), the
distribution coincides with the exciton wave function ¡
k
/
1,ð1 þk
2
Þ
2
[4]. Furthermore, the density of electron-hole
pairs and photons is nearly equal in this limit [Fig. 1(b)].
Figure 1(c) shows the energy per electron-hole pair, which
approaches an energy of E,E
0
% À1.4. The lowering of the
energy is due to the strong coupling and anticrossing of
the exciton and photon to form a lower polariton. Finally,
the gap energy shown in Fig. 1(d) is equal to the energy of a
polariton E
min
k
% 1.4. The gap energy in this case is the
energy required to turn a polariton into a free electron-hole
pair. We thus conclude that exciton-polaritons are correctly
reproduced in the low density limit.
As the density is increased, Fig. 1(a) reveals that the
momentum distribution spreads out to higher momentum
states. This is precisely the opposite behavior to what is
expected in the excitonic BEC-BCS crossover. In a standard
BCS state, the instability towards forming a Cooper pair is
weakened with increasing density, because the electron-hole
attraction becomes increasingly screened by the surrounding
electrons and holes. This results in a ¡
k
distribution that
approaches a Fermi step for high density, for the excitonic
case. In the case of exciton-polaritons, the electron-hole
attraction in fact becomes enhanced with increasing density.
This is evidenced by the gap energy which increases with
density in Fig. 1(d), instead of decreasing in the exciton case.
What is the origin for this enhanced attraction?
Figure 1(b) reveals that at high density the photon number
greatly exceeds that of the electron-hole number. This can
be explained due to a difference in the particle statistics of
the two excitations. Photons are true bosonic particles,
hence any number of them can be excited with an energy
cost w. Meanwhile, electron-hole pairs are fermions, and
suffer a phase space filling effect. In order to excite more
fermionic particles, electrons and holes of increasingly
higher momenta and energy must be occupied in order
to increase the particle number. Thus it is favorable to excite
photons rather than electron-hole pairs to minimize the total
energy, explaining the large imbalance in these numbers.
Given that at high density there is inevitably a large
number of photons, we may return to the original
Hamiltonian (1) to see the consequences. In the high
density limit we may replace the photon operator by a
a
0 1 2 3 4 5
0
0.1
0.2
0.3
0.4
0.5
0.6
v
2
µ=-1.4
µ=-1.2
µ=-1.05
µ=-1.001
k
b
1.4 1.3 1.2 1.1 1
1.4
1.3
1.2
1.1
1
c d
1.4 1.3 1.2 1.1 1
0
5
10
15
20
25
E
k m
i
n
10
10
0
10
2
10
4
-2
n
n
n
ph
n
eh
n
e
h
E
/
n
k
1.4 1.3 1.2 1.1 1
0.0
0.2
0.4
0.6
0.8
1.
FIG. 1 (color online). (a) The occupation probability ¡
2
k
of
electron-hole pairs of momentum k for a cavity photon energy of
w ¼ À1 and photon coupling ¼ 0.1. The dotted line shows
the 1s exciton wave function for comparison. The (b) electron-
hole pair, photon, and total particle density, (c) energy per
electron-hole pair, (d) gap energy versus chemical potential.
PRL 105, 186402 (2010)
P HYS I CAL RE VI E W L E T T E RS
week ending
29 OCTOBER 2010
186402-2
classical c number, which we consider to be very large
o % ¬ ) 1. Discarding all terms that do not contain this
factor, we have
H % ¬

k
½e
y
k
h
y
Àk
þh
Àk
e
k
Š þ¬
2
w. (9)
The solution of this Hamiltonian is the BCS wave function
(2) with u
k
¼ À¡
k
¼ 1,
ffiffiffi
2
p
, agreeing with the numerical
analysis that the average occupation number approaches
1,2 for all k. The BCS gap in this limit corresponds to a
single excitation of the Hamiltonian, which has an energy
E
min
k
% 2¬. (10)
Thus with increasing density the BCS gap continues to
increase in agreement with Fig. 1(d).
Now we ask to what extent the polariton system pos-
sesses BCS features, rather than merely a crossover be-
tween a polariton BECstate to a photon-dominated regime.
The criterion given in Ref. [7] is based on a comparison of
the energy scales of the BCS gap energy E
min
k
with the
temperature required for condensation to occur k
B
T
BEC
¼
r@
2
n,m
pol
, where m
pol
is the polariton mass. In this defi-
nition, the state can be described as ‘‘BCS-like’’ if the
energy to disassociate a polariton is lower than the thermal
excitation energy to prevent a BEC from occurring. In our
units, this gives k
B
T
BEC
,E
0
% 22 000no
2
B
, where we used
m
pol
% 10
À5
m
0
and o
B
¼ 10 nm, and m
0
is the free elec-
tron mass. In terms of Fig. 1(d), this criterion is always
much higher than the gap energy E
min
k
, for densities ex-
ceeding n % 6.5 Â10
7
cm
À2
. Experimentally, such den-
sities have already been achieved, giving the result that all
current polariton BECs are all in the ‘‘BCS-like’’ regime,
according to this criterion.
There is, however, another sense that the polaritons
can be classified as BCS-like. For an excitonic BCS state,
the singlet pair function ÉðkÞ ¼ u
k
¡
k
is peaked near
the vicinity of the Fermi momentum and has a width of
the order of the inverse of the BCS coherence [11].
Figure 2(a) shows the singlet pair function for our polariton
system, which is peaked at nonzero momentum for large
densities. The location of this peak has a nonzero momen-
tum above a critical density [Fig. 2(c)]. Such behavior is
also seen in the excitonic BEC-BCS crossover. The differ-
ence here is that instead of the singlet pair function becom-
ing sharper with increasing density, here the pair wave
function becomes broader. Using blue-detuned (more ex-
citonic) polaritons ÉðkÞ more resembles the excitonic BCS
state [Fig. 2(b)]. An alternative definition of a polariton
BCS phase may be signaled by the presence of the peak of
the singlet pair function at nonzero momentum.
We now turn to the photoluminescence (PL) character-
istics of the transition between low and high density.
Examining the high density limit first, using a similar
approximation to (9), we use the Hamiltonian
H ¼

k
½a
þ
k
o þa
À
k
o
y
Š þwo
y
o þ
e
g
2

k
a
z
k
, (11)
where a
þ
k
¼ e
y
k
h
y
Àk
, e
k
¼ @
2
k
2
,m, a
z
k
¼ e
y
k
e
k
¼ h
y
Àk
h
Àk
,
and we have explicitly included the semiconductor band
gap energy e
g
required to create an electron-hole pair. The
PL spectrum is calculated using [12]
IðeÞ ¼
1
r
Re
_
1
0
hA
y
ðrÞAð0Þie
À|er,@
Jr, (12)
where A is the operator that couples the system to the
external PL field. In the case of polaritons, the PL is
generally observed by leakage of the photon field through
the microcavity, hence A ¼ o. It is also possible to observe
the PL via a secondary means, from the coupling to the
exciton field A ¼ a
À
k
. This type of coupling is that mea-
sured for pure excitons and should also be present in
principle for polaritons. Experimentally the excitonic PL
is emitted homogeneously in all directions, whereas the
photonic PL is emitted perpendicularly to the sample sur-
face. Evaluating the time correlation function (12) for both
types of couplings under a mean field approximation, we
find a spectrum as shown in Fig. 3(a). For the exciton
coupling, the PL spectrum is identical to the familiar
Mollow’s triplet spectrum found in resonant fluorescence
[12], in agreement with Fig. 2 of Ref. [7]. For the photon
coupling only the central peak is present. The reason
for this difference is illustrated by the single spin version
of (11), which has eigenstates for high density jÆ, Ni ¼
ðj ", N À1i Æj #, NiÞ,
ffiffiffi
2
p
, where N is the number of
photons. For large N, the photon operator does not
cause transitions between the Æ eigenstates: ojÆ, Ni %
ffiffiffiffi
N
p
jÆ, N À1i. In contrast, the exciton coupling does
cause a transition a
À
jÆ, Ni % ðjþ, N À1i ÀjÀ,
N À1iÞ,
ffiffiffi
2
p
, giving the side peaks.
In the low density limit, either type of coupling gives the
same PL spectrum, giving the familiar emission centered at
the lower polariton energy. For the photon coupling case, in
both the low and high density limits, the action of applying
the operator A ¼ o does not cause transitions to excited
a
c
b
d
-2
10
-1
10
0
10
1
10
2
10
0 1 2 3 4 5
k
0
0.1
0.2
0.3
0.4
0.5
Ψ
0 1 2 3 4 5
0
0.1
0.2
0.3
0.4
0.5
Ψ
k
-1.5 -1 -0.5 0
µ
0
0.5
1
1.5
2
k
m
a
x
-2
µ= -1.40
µ= -1.36
µ= -1.16
µ= -1.08
µ= -0.10
µ= -0.56
µ=-0.90
µ= -1.07
ω=-E
0
ω=0
n
V
n
0
0.2
0.4
0.6
0.8
1
1.2
V
0.00
0.02
0.04
0.06
FIG. 2 (color online). The singlet pair function ÉðkÞ ¼ u
k
¡
k
for various chemical potentials as shown for (a) zero detuning
w ¼ ÀE
0
(b) blue detuning w ¼ 0. (c) The singlet pair function
peak momenta for detunings shown. (d) The particle-particle
interaction energy V and condensate interaction energy Vn as a
function of the density.
PRL 105, 186402 (2010)
P HYS I CAL RE VI E W L E T T E RS
week ending
29 OCTOBER 2010
186402-3
states. Assuming this is true for all intermediate densities,
we use the results of our ground state BCS wave function to
obtain the PL emission parameters. From the results of the
self-consistent equations (6)–(8), we may write down an
effective Hamiltonian for the BCS theory H
eff
¼ µl
y
l þ
1
2
o
2
B
A
Vl
y
l
y
ll, where l is a bosonic operator for the effec-
tive theory, µ ¼
JE
Jn
and V ¼
1
o
2
B
J
2
E
Jn
2
, where A is the sample
area and E is an energy density. The peak energy of the PL
is the energy of adding a single particle to the system,
which is by definition equal to the chemical potential µ. To
determine the linewidth of the spectrum, we use the
method presented in Ref. [13] to incorporate the effect of
self-interaction on the PL. The self-interaction energy V is
shown in Fig. 2(d). At low densities, the self-interaction
energy is in agreement with the polariton-polariton inter-
action V % 6E
0
jXj
4
[14], where X ¼ 1,
ffiffiffi
2
p
is the polariton
Hopfield coefficient at zero detuning. The condensate self-
interaction energy nV reaches a maximum at intermediate
density, which we attribute to the fact that at high density
the particles are more likely to be present in a photonic
state, which have no interactions with other particles.
Inputting the condensate interaction energy into the theory
of Ref. [13] gives the PL spectrum Fig. 3(b). The PL
spectrum gradually shifts from the lower polariton energy
to the cavity photon energy, with a asymmetric linewidth
with an exponential tail towards high energy. The linewidth
narrows again in the high density limit, due to the de-
creased interactions.
In the excitonic PL, the side peaks of the Mollow’s
triplet should disappear at intermediate densities, when
the saturation effect of the excitons becomes negligible.
The central peak of the excitonic PL should exhibit a
similar behavior to that shown in Fig. 3(b). We note that
only the zero center of mass momentum PL is considered
in our analysis and we leave calculations of dispersion
characteristics as future work.
We have analyzed the crossover between low and high
density limits of exciton-polariton condensates using a BCS
wave-function approach. Contrary to the exciton case, the
electron-hole pairs have a reduced separation in the high
density limit due to the dominant cavity photon field.
Intuitively we picture this state as a strong cavity photon
field continuously creating and destroying electron-hole
pairs at localized positions in the quantum well, resulting
in a half occupancy of ¡
2
k
. In the intermediate density regime,
the system has BCS-like properties in the sense that the pair-
breaking energy is less than the energy required for destroy-
ing the condensate, and a peak in the singlet pair function
develops. The photonic PL shifts from the lower polariton
energy towards the cavity photon energy with a broadening
of the linewidth due to the increased interactions. In the high
density limit the excitonic PL should exhibit a Mollow’s
triplet type structure, originating from the saturation of the
electron-hole occupancy. Inclusion of nonequilibrium effects
resulting from a coupling to an external bath is left as future
work. We expect that inclusion of this effect will reduce the
energy gap due to pair-breaking dephasing processes [15].
The densities required for observing the crossover should be
observable in current experimental systems.
This work is supported by the Special Coordination
Funds for Promoting Science and Technology, the FIRST
program for JSPS, Navy/SPAWAR Grant No. N66001-09-
1-2024, MEXT, and NICT.
Note added in proof.—Recently, we became aware that
the strong binding of electrons and holes due to the photon
field was also independently found in Ref. [16].
[1] J. Kasprzak et al., Nature (London) 443, 409 (2006).
[2] H. Deng et al., Science 298, 199 (2002).
[3] L. V. Keldysh and A. N. Kozlov, Sov. Phys. JETP 27, 521
(1968).
[4] C. Comte and P. Nozie`res, J. Phys. (Paris) 43, 1069 (1982).
[5] C. A. Regal and D. S. Jin, Adv. At. Mol. Opt. Phys. 54, 1
(2006).
[6] P. R. Eastham and P. B. Littlewood, Phys. Rev. B 64,
235101 (2001).
[7] J. Keeling, P. R. Eastham, M. H. Szymanska, and P. B.
Littlewood, Phys. Rev. B 72, 115320 (2005).
[8] J. Keeling, P. R. Eastham, M. H. Szymanska, and P. B.
Littlewood, Phys. Rev. Lett. 93, 226403 (2004).
[9] J. Keeling, F. M. Marchetti, M. H. Szymanska, and P. B.
Littlewood, Semicond. Sci. Technol. 22, R1 (2007).
[10] M. Tinkham, Introduction to Superconductivity (McGraw-
Hill, New York, 1996).
[11] J. R. Waldram, Superconductivity of Metals and Cuprates
(IOP Publishing, Bristol and Philadelphia, 1996).
[12] M. Scully and M. Zubairy, Quantum Optics (Cambridge
University Press, Cambridge, U.K., 1997).
[13] D. Porras and C. Tejedor, Phys. Rev. B 67, 161310(R)
(2003).
[14] C. Ciuti, V. Savona, C. Piermarocchi, A. Quattropani, and
P. Schwendimann, Phys. Rev. B 58, 7926 (1998).
[15] M. H. Szymanska and P. B. Littlewood, Solid State
Commun. 124, 103 (2002).
[16] K. Kamide and T. Ogawa, Phys. Rev. Lett. 105, 056401
(2010).
0 0.2 0.4 0.6 0.8 1
PL intensity
ω ω +2Ωλ ω −2Ωλ
Ι
max
Ι /3
max

3γ/2
γ
photonic PL
excitonic PL
a
1.5
1.4
1.3
1.2
1.1
1
0.9
E
10
-2
10
-1
10
0
10
1
10
2
b
n
FIG. 3 (color online). (a) The PL intensity of the BCS model in
the high density limit for the photonic PL and excitonic PL. A
reservoir coupling ¡ is assumed for both cases, giving the
linewidths as shown. (b) The photonic PL for the BCS model
at all densities. Zero detuning is assumed in all cases.
PRL 105, 186402 (2010)
P HYS I CAL RE VI E W L E T T E RS
week ending
29 OCTOBER 2010
186402-4

1 0 0 1 µ=-1.4 1. and total particle density.4 a 0.05 µ=-1. k Àk The photonic part may be diagonalized by introducing the operator By ¼ ay À c.3 0. Expanding the operators ey ek . In the low density limit with no photon field. Hartree terms (corresponding to q ¼ 0).2 1. 0. In Fig. The gap energy in this case is the k energy required to turn a polariton into a free electron-hole pair.2 0. As the density is increased. (7) hey hy i ¼ uk vk ¼ k k Àk 2Ek qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi Ek ¼ 2 þ ðÁk À Þ2 : (8) k The gap energy Emin is defined to be the value of Ek k minimized over all momenta k. The energy is measured in units E0 ¼ e2 =4aB (SI units used throughout). we obtain the Hamiltonian X H ¼ 0 þ ð! À ÞBy B þ Ek ð y k0 þ y k1 Þ. the electron-hole attraction in fact becomes enhanced with increasing density.4 1. In order to excite more fermionic particles. We thus conclude that exciton-polaritons are correctly reproduced in the low density limit. E0 ¼ k Àk k Àk P y y y k ½Xk hek ek i þ Ák hek hÀk iŠ À 2D and we have taken all expectation values as real.8 0. This is evidenced by the gap energy which increases with density in Fig. and demanding that the offdiagonal terms vanish.4 0.PRL 105.2 1. energy is due to the strong coupling and anticrossing of the exciton and photon to form a lower polariton. ey hy .2 µ=-1. (6) k k 2 Ek Á À  .1 1 15 10 5 0 1. and densities are measured in units of 1=a2 . and a around their mean k k k Àk values and keeping only linear terms gives X H ¼ E0 þDðaþay Þþð!ÀÞay a þ k ðey ek þhy hk Þ k k X À ðÁk ÀÞðey hy þhÀk ek Þ k Àk k @2 k2 2m k b n n ph 1. 2=m ¼ 1=me þ 1=mh . This results in a vk distribution that approaches a Fermi step for high density. the exciton energy is E=E0 ¼ À1. Finally. (5) k0 k1 k P y y where 0 ¼ k ½k À Ek þ Ák hek hÀk i þ Xk hey ek iŠ þ k 2 D2 =ð! À Þ.5 0. (c) energy per electron-hole pair. and restrict q ¼ k0 À k. 186402 (2010) PHYSICAL REVIEW LETTERS 0.3 1. explaining the large imbalance in these numbers. Xk ¼ k0 Vðk À k0 Þhey0 ek0 i. the distribution coincides with the exciton wave function vk / 1=ð1 þ k2 Þ2 [4].4 1. In the high density limit we may replace the photon operator by a 186402-2 n eh n .3 1. The parameters uk and vk satisfy the standard BCS algebra [10]   1  hey ek i ¼ v2 ¼ 1 À k . In the low density limit ( % À1:4). photon. electron-hole pairs are fermions. and suffer a phase space filling effect. which approaches an energy of E=E0 % À1:4. 1(d) is equal to the energy of a polariton Emin % 1:4. hy ¼ k1 k Àvk k0 þ uk y . Photons are true bosonic particles. 1(a) the occupation probability v2 of electron-hole k pairs is shown. the density of electron-hole pairs and photons is nearly equal in this limit [Fig. This gives the condition D : (4) ¼À !À The remaining part of the Hamiltonian may be diagonalized by a transformation ek ¼ uk k0 þ vk y .6 0. This can be explained due to a difference in the particle statistics of the two excitations. The (b) electronhole pair. hence any number of them can be excited with an energy cost !. Furthermore. 1 (color online).1 1 1. This is precisely the opposite behavior to what is expected in the excitonic BEC-BCS crossover. Thus it is favorable to excite photons rather than electron-hole pairs to minimize the total energy. 1(a) reveals that the momentum distribution spreads out to higher momentum states. Demanding that the off-diagonal terms k1 disappear. Figure 1(c) shows the energy per electron-hole pair. the gap energy shown in Fig. the instability towards forming a Cooper pair is weakened with increasing density. In the case of exciton-polaritons.1 1 P where k ¼ À  À Xk . are not present due to the assumption of charge neutrality [4].1 c d E/n (3) 1.2 10-2 2 k 3 4 5 25 20 min Ek 2 week ending 29 OCTOBER 2010 Hartree Fock approximation. Meanwhile. (a) The occupation probability v2 of k electron-hole pairs of momentum k for a cavity photon energy of ! ¼ À1 and photon coupling  ¼ 0:1. Equations (6)–(8) with (4) are solved selfconsistently to obtain our results. Zero detuning therefore corresponds to a photon energy of !=E0 ¼ À1 (the photon energy is negative since we measure energies relative to the band gap energy). The dotted line shows the 1s exciton wave function for comparison. and hey ek i ¼ hhy hÀk i due to charge neutrality. instead of decreasing in the exciton case.  ¼ hai. for the excitonic case. because the electron-hole attraction becomes increasingly screened by the surrounding electrons and holes. electrons and holes of increasingly higher momenta and energy must be occupied in order to increase the particle number. Given that at high density there is inevitably a large number of photons.2 1.4 0. hy hk .3 1.4 1. In a standard BCS state. 1(b)]. What is the origin for this enhanced attraction? Figure 1(b) reveals that at high density the photon number greatly exceeds that of the electron-hole number.001 104 10 µ=-1. where B the Bohr radius is aB ¼ 4@2 =me2 . k P P Ák ¼ k0 Vðk À k0 Þhey0 hy 0 i.0 0.3 1. Fig. 1(d). we use units such that the momentum is measured in units of 1=aB . we may return to the original Hamiltonian (1) to see the consequences.6 vk 2 100 1 1. In our numerical results. (d) gap energy versus chemical potential.2 n eh 1. The lowering of the FIG. D ¼ k hey hy i.

02 Vn The solution of this Hamiltonian is the BCS wave function pffiffiffi (2) with uk ¼ Àvk ¼ 1= 2. N À 1i.90 b µ = -0. Now we ask to what extent the polariton system possesses BCS features.2 0.6 0.56 µ =-0. (11) k k 2 k k k c ω=0 d k max 1 0. z ¼ ey ek ¼ hy hÀk . In contrast.1 0 µ = -1. the photon operator does not cause pffiffiffiffi transitions between the Æ eigenstates: ajÆ. NiÞ= 2.08 week ending 29 OCTOBER 2010 classical c number.PRL 105. using a similar approximation to (9). here the pair wave function becomes broader.5 0 -2 -1. the action of applying the operator A ¼ a does not cause transitions to excited 186402-3 . 2(c)]. For the photon coupling case. the PL spectrum is identical to the familiar Mollow’s triplet spectrum found in resonant fluorescence [12]. Using blue-detuned (more excitonic) polaritons ÉðkÞ more resembles the excitonic BCS state [Fig.5 0. For the exciton coupling. Ni ¼ pffiffiffi ðj ".5 µ = -1. In terms of Fig.3 0. [7]. Experimentally the excitonic PL is emitted homogeneously in all directions.4 0. We now turn to the photoluminescence (PL) characteristics of the transition between low and high density.36 µ = -1. Discarding all terms that do not contain this factor. In the case of polaritons.04 0. There is. the PL is generally observed by leakage of the photon field through the microcavity. For large N. 1(d).5 0 10 -2 10 -1 10 n 0 10 1 10 2 0. 2 of Ref. which has eigenstates for high density jÆ. whereas the photonic PL is emitted perpendicularly to the sample surface. It is also possible to observe the PL via a secondary means. which we consider to be very large a %  ) 1. [7] is based on a comparison of the energy scales of the BCS gap energy Emin with the k temperature required for condensation to occur kB TBEC ¼ @2 n=mpol . (12) IðÞ ¼ Re  0 where A is the operator that couples the system to the external PL field. in both the low and high density limits. giving the familiar emission centered at the lower polariton energy. we find a spectrum as shown in Fig.5 -1 V ω=-E 0 µ -0. either type of coupling gives the same PL spectrum. The singlet pair function ÉðkÞ ¼ uk vk for various chemical potentials as shown for (a) zero detuning ! ¼ ÀE0 (b) blue detuning ! ¼ 0. (c) The singlet pair function peak momenta for detunings shown. The PL spectrum is calculated using [12] Z1 1 hAy ðÞAð0ÞieÀi=@ d. where we used B mpol % 10À5 m0 and aB ¼ 10 nm. we use the Hamiltonian X g X z H ¼  ½þ a þ À ay Š þ !ay a þ  . the state can be described as ‘‘BCS-like’’ if the energy to disassociate a polariton is lower than the thermal excitation energy to prevent a BEC from occurring. which is peaked at nonzero momentum for large densities. we have X H %  ½ey hy þ hÀk ek Š þ 2 !: (9) k Àk k a Ψ 0. the singlet pair function ÉðkÞ ¼ uk vk is peaked near the vicinity of the Fermi momentum and has a width of the order of the inverse of the BCS coherence [11]. 2 (color online).8 0. For the photon coupling only the central peak is present. 1(d). where mpol is the polariton mass. this gives kB TBEC =E0 % 22 000na2 . for densities exk ceeding n % 6:5  107 cmÀ2 . This type of coupling is that meak sured for pure excitons and should also be present in principle for polaritons. giving the result that all current polariton BECs are all in the ‘‘BCS-like’’ regime. In this definition. 186402 (2010) PHYSICAL REVIEW LETTERS 0. Experimentally. such densities have already been achieved. hence A ¼ a. The difference here is that instead of the singlet pair function becoming sharper with increasing density. p N À 1iÞ= 2.2 0 0 1 2 k 3 4 5 0. For an excitonic BCS state. Ni % ðjþ. Figure 2(a) shows the singlet pair function for our polariton system. in agreement with Fig. An alternative definition of a polariton BCS phase may be signaled by the presence of the peak of the singlet pair function at nonzero momentum. Such behavior is also seen in the excitonic BEC-BCS crossover. Evaluating the time correlation function (12) for both types of couplings under a mean field approximation. N À 1i Æ j #.3 0. another sense that the polaritons can be classified as BCS-like. however.16 µ = -1. the exciton coupling does cause a ffiffiffi transition À jÆ. k ¼ @2 k2 =m.4 0. k k Àk k Àk k and we have explicitly included the semiconductor band gap energy g required to create an electron-hole pair. according to this criterion. The location of this peak has a nonzero momentum above a critical density [Fig. The BCS gap in this limit corresponds to a single excitation of the Hamiltonian. N À 1i À jÀ. rather than merely a crossover between a polariton BEC state to a photon-dominated regime. and m0 is the free electron mass. In the low density limit. this criterion is always much higher than the gap energy Emin . agreeing with the numerical analysis that the average occupation number approaches 1=2 for all k. where N is the number of photons.07 µ = -0. Ni % N jÆ. The reason for this difference is illustrated by the single spin version of (11).40 µ = -1.5 0. from the coupling to the exciton field A ¼ À . where þ ¼ ey hy . 2(b)].4 0.06 0.2 0. Examining the high density limit first. The criterion given in Ref.2 1 0.1 0 0 2 1.00 FIG. In our units. which has an energy Emin % 2: k (10) Thus with increasing density the BCS gap continues to increase in agreement with Fig. giving the side peaks. (d) The particle-particle interaction energy V and condensate interaction energy Vn as a function of the density. 3(a).10 Ψ 1 2 k 3 4 5 1.

From the results of the self-consistent equations (6)–(8). B 72.3 1. V. A. Szymanska. Rev. which have no interactions with other particles. In the excitonic PL. H. B. which we attribute to the fact that at high density the particles are more likely to be present in a photonic state. Sov. Littlewood. and P.4 0. E [1] J. Deng et al. Szymanska and P. 3(b). Regal and D. M.4 1. [10] M. B 64. H. 1996). H. Phys. B. The condensate selfinteraction energy nV reaches a maximum at intermediate density. The densities required for observing the crossover should be observable in current experimental systems. [12] M. [2] H. Introduction to Superconductivity (McGrawHill. Jin. Phys. Kamide and T. where A is the sample dn dn2 B 2 1 aB 2 A area and E is an energy density. [16] K. 7926 (1998). we use the method presented in Ref. Bristol and Philadelphia.5 10-2 10-1 100 n 101 102 week ending 29 OCTOBER 2010 b Ιmax a excitonic PL γ photonic PL 2γ 3γ/2 Ι max /3 ω −2Ωλ ω ω +2Ωλ FIG. [13] to incorporate the effect of self-interaction on the PL. where X ¼ 1= 2 is the polariton Hopfield coefficient at zero detuning. Zero detuning is assumed in all cases. and P. 54. ` [4] C. the electron-hole pairs have a reduced separation in the high density limit due to the dominant cavity photon field.  ¼ dE and V ¼ a12 d E . Waldram. 124..1 1. H. In the high density limit the excitonic PL should exhibit a Mollow’s triplet type structure. [8] J. The linewidth narrows again in the high density limit. [11] J. 1997). Savona. where b is a bosonic operator for the effec2 tive theory. Note added in proof. J. [9] J. Solid State Commun. Rev. 2(d). To determine the linewidth of the spectrum. resulting in a half occupancy of v2 . states. 161310(R) (2003). A reservoir coupling is assumed for both cases. Ciuti. Eastham and P. M. Scully and M. 105. R. Keldysh and A. with a asymmetric linewidth with an exponential tail towards high energy. [13] D. Inclusion of nonequilibrium effects resulting from a coupling to an external bath is left as future work. Porras and C.K. P. F. N. B 58. [5] C. 3(b). Comte and P. Tejedor. [14] C. Marchetti. 199 (2002).. the self-interaction energy is in agreement with the polariton-polariton interpffiffiffi action V % 6E0 jXj4 [14]. B.6 0.. Quattropani. [7] J. Quantum Optics (Cambridge University Press. New York. (Paris) 43. Opt. R. Schwendimann. Littlewood. R1 (2007). P. M. Phys. [3] L. Contrary to the exciton case. Eastham. 1069 (1982). Phys. B. Phys. Mol. Nozieres. Kasprzak et al. The central peak of the excitonic PL should exhibit a similar behavior to that shown in Fig. 235101 (2001). The PL spectrum gradually shifts from the lower polariton energy to the cavity photon energy. (b) The photonic PL for the BCS model at all densities. MEXT. S. Keeling. Cambridge. R. V. 1996). At low densities. Technol.PRL 105. This work is supported by the Special Coordination Funds for Promoting Science and Technology. and a peak in the singlet pair function develops.9 1 1. and P. [13] gives the PL spectrum Fig. Ogawa. We have analyzed the crossover between low and high density limits of exciton-polariton condensates using a BCS wave-function approach. originating from the saturation of the electron-hole occupancy. Lett. Keeling. k the system has BCS-like properties in the sense that the pairbreaking energy is less than the energy required for destroying the condensate.2 1. the side peaks of the Mollow’s triplet should disappear at intermediate densities. 409 (2006). the FIRST program for JSPS. C. 93. and NICT. N66001-091-2024. 226403 (2004). Phys. 521 (1968). [6] P. Nature (London) 443. 186402-4 . we use the results of our ground state BCS wave function to obtain the PL emission parameters. Phys. which is by definition equal to the chemical potential . Tinkham. Kozlov. we became aware that the strong binding of electrons and holes due to the photon field was also independently found in Ref. Littlewood.—Recently. Szymanska. 115320 (2005). The self-interaction energy V is shown in Fig. when the saturation effect of the excitons becomes negligible. (a) The PL intensity of the BCS model in the high density limit for the photonic PL and excitonic PL. U. A. [15] M. Science 298. Eastham. Piermarocchi. 056401 (2010). Littlewood. R. due to the decreased interactions. Rev.2 0. Sci. Zubairy. Intuitively we picture this state as a strong cavity photon field continuously creating and destroying electron-hole pairs at localized positions in the quantum well.8 1 PHYSICAL REVIEW LETTERS 0. 1 (2006). Rev. We expect that inclusion of this effect will reduce the energy gap due to pair-breaking dephasing processes [15]. Semicond. Phys. Rev. 186402 (2010) PL intensity 0 0. The peak energy of the PL is the energy of adding a single particle to the system. and P. 103 (2002). M. Keeling. Rev. JETP 27. Adv. Phys. Littlewood. 22. Navy/SPAWAR Grant No. Assuming this is true for all intermediate densities. B 67. The photonic PL shifts from the lower polariton energy towards the cavity photon energy with a broadening of the linewidth due to the increased interactions. 3 (color online). we may write down an effective Hamiltonian for the BCS theory Heff ¼ by b þ Vby by bb. In the intermediate density regime. At. B. Superconductivity of Metals and Cuprates (IOP Publishing. [16]. We note that only the zero center of mass momentum PL is considered in our analysis and we leave calculations of dispersion characteristics as future work. giving the linewidths as shown. Lett. Szymanska. Inputting the condensate interaction energy into the theory of Ref.