Tunable joint measurements in the dispersive regime of cavity QED

Kevin Lalumire,1 J. M. Gambetta,2 and Alexandre Blais1

arXiv:0911.5322v2 [quant-ph] 5 Apr 2010

Dpartement de Physique, Universit de Sherbrooke, Sherbrooke, Qubec, Canada, J1K 2R1

2
Institute for Quantum Computing and Department of Physics and Astronomy,
University of Waterloo, Waterloo, Ontario, Canada, N2L 3G1
(Dated: April 7, 2010)

Joint measurements of multiple qubits open new possibilities for quantum information processing.
Here, we present an approach based on homodyne detection to realize such measurements in the
dispersive regime of cavity/circuit QED. By changing details of the measurement, the readout can
be tuned from extracting only single-qubit to only multi-qubit properties. We obtain a reduced
stochastic master equation describing this measurement and its effect on the qubits. As an example,
we present results showing parity measurements of two qubits. In this situation, measurement of an
initially unentangled state can yield with near unit probability a state of significant concurrence.
PACS numbers: 03.65.Yz, 42.50.Pq, 42.50.Lc, 74.50.+r

In most of the current quantum information experiments, measurements are used to extract information
only about single-qubit properties. Joint measurements
where information about both single and multi-qubit
properties can be obtained offer new possibilities. Examples are the test of quantum paradoxes [1], test of
quantum contextuality [2], realization of quantum state
tomography with weak measurements [35] and cluster
state preparation [6]. A particularly powerful type of
joint measurement is parity measurement, where information is gained only about the overall parity of the
multi-qubit state, without any single-qubit information.
This type of measurement can be used for the generation of entanglement without unitary dynamics [710],
for quantum error correction [11, 12], and deterministic quantum computation with fermions [13, 14]. In this
paper, we show how such joint measurements can be realized in the dispersive regime of cavity QED [15]. In particular, we show how the character of the measurement
can be tuned from purely single-qubit to parity readout. As a realistic example, we present results for circuit
QED [35] and show that states with large concurrence
can be obtained. Entanglement generation by measurement was previously studied in this system [1619], but
ignoring information about the parity. With parity measurements, entanglement generation by measurement can
be deterministic rather than probabilistic.
We consider a pair of two-level systems (i.e. qubits) of
frequencies aj with j = 1, 2 coupled to a high-Q cavity
of frequency r . In the dispersive limit, where |j | =
|(aj r )| |gj | with gj the coupling strength of qubit
j to the cavity, the Hamiltonian of this system takes the
form [20]
H =(r +

j zj )a a +

X
aj
j

im t

+ m (t)(a e

1 2
2 1
zj + Jq (
+ +
+ )

+ h.c.).
(1)

This result is valid to second order in the small parameter j = gj /j . Here, we have defined the dispersive
coupling strength j = gj j , the Lamb-shifted qubit frequency
aj and the strength of qubit-qubit coupling mediated by virtual photons Jq = g1 g2 (1/1 +1/2 )/2 [20].
The last term represents a coherent drive on the cavity
of amplitude m (t) and frequency m r , appropriate
for measurement of the qubits. With this choice of drive
frequency, we have safely dropped a qubit driving term
of amplitude j m [20]. In order to focus on entanglement generated by measurement only, we drop the term
proportional to Jq . This is reasonable since the possible measurement outcomes are eigenstate of the flip-flop
1 2
2 1
interaction
+ +
+ , as will be clear below.
Coupling to unwanted degrees of freedom is modeled by using a Lindblad-type master equation [21]. In
Ref. [16], a master equation for the qubits only was obtained by enslaving the cavity to the qubit dynamics.
This approach is valid only in the limit where damping
of the cavity greatly overwhelms the dispersive coupling strength j . Here, we go beyond these results by
using a polaron-type transformation to trace-out the cavity [22, 23]. Starting from Eq. (1), we find following
Ref. [22] the effective master equation
X
aj
j

zj , ] +

j
1j D[
] +

X j

D[zj ]
2
2
j
j
j
X xy
X
j
xy
j ] +
(d iAc ) x y L,
+ D[

i[

xy

(2)
where D[c] = c c {c c, }/2. In this expression, 1j is
the relaxation rate of qubit j and j its pure dephasing
rate. The fourth term represents Purcell damping at the
rate 2j [23], while the last contains both measurementxy
induced dephasing (xy
d ) and ac-Stark shift (Ac ) by the
measurement photons [? ]. In Eq. (2), x (y) stands
for one of the four logical states ij with i, j {g, e}
the qubits ground and excited states and x = |xihx|.

2
a)

0.6

gg

ee

0.0

Normalized Information Gain Rates

0.2

Quadrature, Q

0.2
0.4
0.6
0.8

eg ge

1.0
1.2
0.4

0.2

0.0

0.2

0.4

average over realizations of the noise. This stochastic

equation is valid for /2 11 + 12 , which is again
easily satisfied [24].
In Eq. (6), the joint measurement operator c is

b)

0.4
0.2
0.0
0.8
0.6
0.4
0.2
0.0

c =
-4

1.0
0.8
0.6
0.4
0.2
0.0

-2

r[ ]

p
p
p
10 ()z1 + 01 ()z2 + 11 ()z1 z2 ,

(7)

where
ij () = |ij |2 cos2 ( ij ),

c)

ij = ee + (1)j eg + (1)i ge + (1)i+j gg /2,
(8)
0

In-Phase, I

FIG. 1. (Color online) a) Phase space illustration of the stationary states |ij i for: g1 = g2 = 15 and j 1.5.
The drive is at resonance with the bare cavity r = 0 and its
amplitude is = /2. b) Normalized rates of information gain
for = 0 or = /2: 01 () = 10 () (full red line), 11 ()
(dashed purple line). The vertical lines are at 2j . c) Normalized rates at r = 0 as a function of j : 01 (0) = 10 (0)
(full red line), 11 (/2) (dashed purple line). The vertical
line indicates the value of j used in panel b). Other rates
11 (0), 01 (/2), and 10 (/2) are zero and not shown.

Measurement-induced dephasing and ac-Stark shift are

given by

xy
d = (x y )Im[x y ],

(3)

(4)
Axy
c = (x y )Re[x y ],
P j j
where x = hx| j z |xi and x the amplitude of the
coherent state when the qubits are in state |xi. This
amplitude satisfies

x = i(r + x )x im (t)eim t x /2.

(5)

The reduced master equation Eq. (2) is a very good approximation to the full dynamics when /2 1j . Since
1j does not include Purcell damping, this inequality is
easily satisfied with current Purcell limited qubits [24].
To go beyond information about average evolution, we
use quantum trajectory theory of homodyne measurement on the transmitted cavity field to obtain information about single experimental runs [25]. Following the
approach of Ref. [22], we find in the multi-qubit case the
reduced stochastic master equation (SME)
J =LJ + M[c ]J (t) i[c/2 , J ](t)/2,

(6)

and the measured homodyne current is proportional to

J(t) = Tr[c J ]+(t). Here M[c] = {c, }/2Tr[c], and
(t) is Gaussian white noise satisfying E[(t)] = 0 and
E[(t)(t )] = (t t ), with E[] denoting an ensemble

with the phase of the local oscillator, = Arg(), and

the efficiency with which the photons leaking out of the
cavity are detected. ij represents the rate of information gained about the first qubit polarization (ij=10),
second qubit polarization (ij=01) or the parity (ij=11).
An optimal measurement occurs when c/2 = 0 since,
in this case, all the back-action arising from the measurement is associated with information gain [22]. Given the
form of c/2 , this cannot be realized, except in trivial
cases.
Given that j , r = r m and can be changed
in-situ [35], the form of the measurement operator c
can be tuned (in the dispersive approximation, changing
m only leads to an overall rescaling). There are several
useful choices of c . For example, an equally weighted
joint measurement (all |ij | equal) is ideal for quantum
state tomography since in this case both the required
single and two-qubit information are on an equal footing.
In the limit |1 2 | , this is achieved by choosing
m to match one of the four pulled cavity frequencies
r + x . As can be seen in Fig. 1b), for 1 = 2 , an
equally weighted joint measurement is realized by setting
r = 2j . For this choice of j however, at r = 0 it
is not possible to determine which qubit is excited and as
a result the measurement is either completely collective
(z1 + z2 ) for = 0 [16] or more interestingly extracts
information only about the parity (z1 z2 ) of the combined
two-qubit state for = /2.
This can be understood by considering the steady-state
cavity amplitude x . Fig. 1a) shows a phase-space plot
corresponding to the four coherent states |x i for the parameters given in the caption. Since 1 = 2 , the coherent states eg and ge overlap, while Im[ee ] = Im[gg ]
but Re[ee ] 6= Re[gg ] [? ]. As a result, measurement
of the Q ( = /2) quadrature reveals information only
about the parity and I ( = 0) the collective polarization.
Since, for these parameters, there is information in the
quadrature orthogonal to the measurement, c/2 6= 0,
and this measurement is not optimal. As illustrated in
Fig. 1c) however, as the ratio j / is increased, the
measurement becomes
optimal for parity with 01 (0),
P
normalized by ij=01,10,11 [ij (0) + ij (/2)], scaling as
(/j )2 .

3
0.5

a)

0.89

b)

0.4

0.67

0.3

0.8

0.45

0.2
0.1
0
0

0.22
0.6
50

100

150

Time, t [

200

0.00

250

-270

]
0.179

0.4

FIG. 2. (Color online) Mean fidelity to |+ i (dashed green

and full purple lines) and |+ i (dotted-dashed red and dotted blue lines) obtained from solving Eq. (2). Dashed green
and dotted-dashed red lines: (1j , j ) = (0, 0). Full purple
and dotted blue lines: (/250,0). The measurement drive is
tanh(t/) with = 1/ and the initial state (|gi + |ei)
(|gi + |ei)/2. The other parameters are g1 = g2 = 100,
j = 10, = and /2 = 5 MHz.

An application of parity measurements is the generation of entangled states from separable ones [710]. In
contrast to collective polarization measurements [1619],
this can be achieved with unit probability. For example,
with the initial separable state (|gi + |ei) (|gi + |ei)/2,
the measurement ideally
projects on the Bell

states
|+ i = (|egi+ |gei)/ 2 or |+ i = (|ggi+ |eei)/ 2. That
is, evolution under Eq. (6) shows a collapse of the separable state to |+ i or |+ i, conditioned on the record J(t)
being predominately negative or positive respectively.
There are four main causes of errors in this collapse.
The first is relaxation and damping [the dissipative terms
of Eq. (2)]. Interestingly, with the parameters of Fig. 1,
1 = 2 such that |+ i is immune from Purcell decay [16]. The second is the time-dependent ac-Stark shift
[unitary contribution from the last term of Eq. (2)] which
causes a phase accumulation between |ggi and |eei in
|+ i. This contribution can be seen as a slow oscillation of the fidelity F = h||i between the state |+ i
and those obtained by numerical integration of Eq. (2).
This is illustrated in Fig. 2. There, the mean fidelity to
|+ i is always 1/2, since half the density matrices collapse to that state, while oscillations due to the ac-Stark
shift appears in the fidelity to |+ i. However, this shift
is deterministic and can thus be undone. The third error comes from c0 6= 0 causing a stochastic phase between |ggi and |eei [last term of Eq. (6)]. For a given
experimental run, this does not reduce the concurrence
or purity of the state [because (t) is known from J(t)].
However, since this phase varies from shot to shot, the
ensemble averaged state is mixed. This error can be overcomed by performing J(t)-dependent single qubit phase
operations after the measurement or, more simply, by operating in the large j limit where its effect is negligible
as illustrated in Fig. 1c). Finally, the measurement is not
ideal in the sense that measurement-induced dephasing
affects the measurement outcome |+ i (i.e. ee,gg
6= 0).
d
However, this effect can be made negligible by increasing

0.134

-90

90

270

c)
+

0.089

0.2

0.045
0

0.000
0

12.5

25

37.5

Time, t [

50

-895

-224

447

Integrated signal, s

FIG. 3. (Color online) a) Concurrence as a function of time

for 104 trajectories. Dotted-dashed red line: (1j , j , ) =
(0, 0, 1), dashed green line: (/250,0,0.8), dotted blue line:
(/250,0,0.2) and full purple line: (/250,0,0.05). = /2.
All other parameters are the same as in Fig. 2. b-c) Histograms of the integrated current s(t) at b) t = 1.6/ and c)
t = 6.3/.

the ratio j / since 11 (/2)/ee,gg

(j /)2 .
d
To show that the system collapses to |+ i or |+ i,
and that entanglement is generated with unit probability,
Fig. 3a) shows the mean concurrence E[C(J )], averaged
over 104 trajectories. There, departure from unit concurrence in the dotted-dashed red line (no damping, unit
detector efficiency = 1) is only due to measurementinduced dephasing. The dashed green, dotted blue and
full purple lines take into account relaxation with /1j
250, and detection efficiency of = 4/5, = 1/5, =
1/20, respectively. The latter corresponds to current experimental values [24]. The ratio /1j is slightly out of
reach of current experiments when taking into account
that j = 10 is also required. This cannot be achieved
with transmons as current experiments have reached the
maximal possible coupling [26]. However, new ideas to
increase the qubit-cavity coupling can help in achieving
these parameters [27].
Small detection efficiency reduces the ratio 11 /ee,gg
,
d
which in turns corrupts |+ i. As illustrated in Fig. 3a),
this results in lower concurrences when < 1. Interestingly, |+ i is not affected by this detection efficiency [16].
Nevertheless, improvement in detection efficiency is required to match concurrences that can be realized with an
entangling Hamiltonian [4]. Recent improvements with
near quantum-limited amplifiers are a good step in this
direction [28].
Having generated one of the two orthogonal entangled
state, it is necessary to distinguish them efficiently. Using the experimental record J(t) to compute J (t) from
the SME Eq. (6) is not efficient since the record is widely

4
1
0.8
0.6
0.4
0.2
0
0

200

400

600

800

FIG. 4. (Color online) Average concurrence (dashed red line),

average fidelity to |+ i and |+ i (dotted green line) and success probability (full blue line) as a function of the threshold
at time t = 18.5/. (1j , j , ) = (/250,0,0.05). The other
parameters are the same as in Fig. 2. The ac-Stark shift affecting |+ i has been corrected before evaluating the fidelity.

fluctuating. As a result, a useful and more efficient quantity to distinguish the states is the integrated current
Z t
p
s
s(t) = 11
J(t )dt ,
(9)
0

where s11 is the steady-state value of 11 (/2). Fig. 3b)

and c) show two histograms of s(t) at times t = 1.6/
and t = 6.3/. These results are for = 1 and exclude
damping for illustration purposes. The full blue lines
are Gaussians fits to the histograms. These separate at
a rate 11 . At times large compared to 1/11 , but
short compared to T1 and T2 , the distributions are well
separated and correspond to |+ i and |+ i.
As shown in Fig. 3c), we introduce a threshold sth to
distinguish these states. All outcomes with s(t) < s0 sth
(condition c = ) are assigned to |+ i, while those with
s(t) > s0 + sth (condition c = +) to |+ i where s0 is
the median of s. Values outside this range are disregarded. A success probability Ps can then be defined as
the probability for s to be outside the range s0 sth . To
quantify the success in generation and distinguishability of the entangled states, we define the average fidelity
F = [h+ |E [J ]|+ i + h+ |E+ [J ]|+ i]/2 and average
concurrence C = [C(E+ [J ])+C(E [J ])]/2. Ec [J ] represents the ensemble average over J for condition c = .
These quantities are illustrated as a function of sth for
the fixed integration time t = 18.5/ in Fig. 4. Even
when keeping all events (sth = 0), F and C are large
with values 0.92 and 0.79 respectively. That is, with this
procedure, it is possible to create and distinguish highly
entangled states with unit probability. If willing to sacrifice some events, this average fidelity and concurrence
is increased to 0.98 and 0.91, respectively. The deviation
from unity in the large sth limit is due to slight corruption
of the state |+ i discussed previously.

In conclusion, we have shown how measurements in

the dispersive regime of two-qubit cavity QED can be
tuned from accessing single to multi-qubit information,
thus allowing for example parity measurements. In addition to allowing complete characterization of the twoqubit states [35] and the implementation of quantum information protocols [6, 11, 12], this allows for generation
of entanglement by measurement with unit probability.
We thank D. Poulin and M. P. da Silva for valuable
discussions. KL was supported by FQRNT and NSERC;
JMG by a CIFAR Junior Fellowship, MITACS, MRI and
NSERC; AB by NSERC, CIFAR and the Alfred P. Sloan
Foundation.

[1] J. S. Lundeen and A. M. Steinberg, Phys. Rev. Lett. 102,

020404 (2009).
[2] G. Kirchmair et al., Nature (London) 460, 494 (2009).
[3] S. Filipp et al., Phys. Rev. Lett. 102, 200402 (2009).
[4] L. DiCarlo et al., Nature (London) 460, 240 (2009).
[5] J. M. Chow et al., arXiv:0908.1955v1 (2009).
[6] S. G. R. Louis et al., Phys. Rev. A 75, 042323 (2007).
[7] W. Mao et al., Phys. Rev. Lett. 93, 056803 (2004).
[8] B. Trauzettel et al., Phys. Rev. B 73, 235331 (2006).
[9] N. S. Williams and A. N. Jordan, Phys. Rev. A 78,
062322 (2008).
[10] C. Hill and J. Ralph, Phys. Rev. A 77, 014305 (2008).
[11] D. Bacon, Phys. Rev. A 73, 012340 (2006).
[12] J. Kerckhoff et al., Phys. Rev. A 79, 024305 (2009).
[13] C. W. J. Beenakker et al., Phys. Rev. Lett. 93, 020501
(2004).
[14] H.-A. Engel and D. Loss, Science 309, 586 (2005).
[15] S. Haroche and J.-M. Raimond, Exploring the Quantum:
Atoms, Cavities, and Photons (Oxford University Press,
Oxford, 2006).
[16] C. L. Hutchison et al., Can. J. Phys. 87, 225 (2009).
[17] D. A. Rodrigues et al., J. Phys.: Condens. Matter 20,
075211 (2008).
[18] F. Helmer and F. Marquardt, Phys. Rev. A 79, 052328
(2009).
[19] L. S. Bishop et al., New J. Phys. 11, 073040 (2009).
[20] A. Blais et al., Phys. Rev. A 75, 032329 (2007).
[21] D. Walls and G. J. Milburn, Quantum Optics (Springer,
Berlin, 2008).
[22] J. Gambetta et al., Phys. Rev. A 77, 012112 (2008).
[23] M. Boissonneault, J. M. Gambetta, and A. Blais, Phys.
Rev. A 79, 013819 (2009).
[24] A. A. Houck et al., Phys. Rev. Lett. 101, 080502 (2008).
[25] H. M. Wiseman and G. J. Milburn, Phys. Rev. A 47, 642
(1993).
[26] M. Devoret, S. M. Girvin, and R. J. Schoelkopf, Ann.
Phys. 16, 767 (2007).
[27] J. Bourassa et al., Phys. Rev. A 80, 032109 (2009).
[28] M. A. Castellanos-Beltran et al., Nat. Phys. 4, 929
(2008).