Stabilising entanglement by quantum jump-based feedback
A. R. R. Carvalho
Department of Physics, Faculty of Science, The Australian National University, ACT 0200, Australia
J. J. Hope
arXiv:0705.3492v1 [quant-ph] 24 May 2007
Australian Centre for Quantum-Atom Optics, Department of Physics,
Faculty of Science, The Australian National University, ACT 0200, Australia
(Dated: November 26, 2018)
We show that direct feedback based on quantum jump detection can be used to generate entangled
steady states. We present a strategy that is insensitive to detection inefficiencies and robust against
errors in the control Hamiltonian. This feedback procedure is also shown to overcome spontaneous
emission effects by stabilising states with high degree of entanglement.
PACS numbers: 03.67.Mn,42.50.Lc,03.65.Yz
Numerous applications of quantum information theory
require the ability to produce entangled states and perform controlled operations on them. There have been
many successful experiments in this direction [1, 2, 3,
4, 5] but despite the effort to screen the system against
unwanted imperfections and interactions, entanglement
degradation through uncontrolled coupling with the environment remains a major obstacle [6, 7]. Even if experiments were to be performed under perfect conditions, fundamental factors such as spontaneous emission
in atomic qubits [8] would persist, limiting the lifetime
of entangled states and demanding efficient schemes to
protect them.
Recent experimental developments have enabled realtime monitoring and manipulation of individual quantum
systems [9, 10, 11, 12], suggesting that quantum feedback
control [13, 14, 15, 16], may emerge as a natural possible
route to develop strategies to prepare entangled states
and prevent their deterioration. Recent attempts have
been made in this direction with proposals to control
both continuous [17, 18] and discrete [19, 20, 21] variable
entanglement, using either Bayesian [16] or Markovian
(direct) [14, 15] feedback scheme. While in the latter
strategy a simple feedback directly proportional to the
detection signal is used, in the former, control depends
on an estimate of the system state based on the information acquired from the measurement results. Although
this can result in an improvement over the direct feedback scheme, it also comes at the cost of an increasing
complexity in the experimental implementation due to
the (challenging) need for a real time estimation of the
quantum state.
In this Letter we will show that a Markovian feedback scheme based on the continuous monitoring of quantum jumps, together with an appropriate choice for the
feedback Hamiltonian, can lead to an improvement, in
amount and robustness, of the steady state entanglement in a model of two driven and collectively damped
qubits [21, 22]. In the absence of spontaneous emission, a
pure maximally entangled state is dynamically generated
irrespective of detection inefficiencies. Furthermore, this
strategy is also able to cope with spontaneous emission
effects by stabilising highly entangled states.
Our system consists of a pair of two-level atoms
equally, and resonantly, coupled to a single cavity mode,
with a coupling strength g. The atoms can spontaneously
decay with rates γ1 and γ2 , and are simultaneously driven
by a laser field (see Fig. 1). The cavity mode is damped,
and in the limit where its decay rate κ is very large, it
can be adiabatically eliminated leading to the following
master equation for the atomic degrees of freedom [21]:
ρ̇ = −iΩ [(J+ + J− ), ρ] + ΓD[J− ]ρ + γ1 D[σ1 ]ρ + γ2 D[σ2 ]ρ.
(1)
Here, Ω is the effective Rabi frequency of the collective
driving, Γ = g 2 /κ is the collective decay rate of the atoms
and the cavity, and superoperator D acting on an operator c is given by D[c]ρ ≡ cρc† − c† cρ + ρc† c /2. The
Hamiltonian is written in terms of angular momentum
operators J− = σ1 + σ2 and J+ = σ1+ + σ2+ , where
σi = |gi ihei | and σi+ = |ei ihgi | are, respectively, the lowering and raising operators for the i-th two level atom.
In the limit that the collective decay rate is much larger
than the spontaneous emission rates, Γ ≫ γ1 , γ2 , we recover the Dicke model [23]
ρ̇ = Lρ = −iΩ [(J+ + J− ), ρ] + ΓD[J− ]ρ.
(2)
Before investigating the influence of feedback in this
equation, let us first briefly analyse the entanglement
properties of its steady state solutions [22]. The first
important feature of Eq. (2) is that it is symmetric with
respect to exchange of the atoms. This suggests that, instead of using the two qubit basis {|ggi, |gei, |egi, |eei},
one should use angular momentum states, and analyse
the system in terms of the symmetric (j = 1)
|1i = |ggi, |2i =
|gei + |egi
√
, |3i = |eei,
2
(3)
and anti-symmetric (j = 0)
|4i =
subspaces.
|gei − |egi
√
2
(4)
2
FIG. 1: Schematic view of the model. The system consists of a
pair of two-level simultaneously driven by a laser and coupled
to a damped cavity. Conditioned on the measurement of the
output of the leaky cavity, a Hamiltonian is applied to the
atoms, completing the feedback scheme.
A simple inspection shows that |4i is a stationary state
solution of Eq. (2). Despite the triviality of the dynamics in this subspace, it shouldn’t be regarded as a totally
uninteresting case as far as entanglement production is
concerned, since its asymptotic state is a pure, maximally
entangled one. In fact, this situation was explored in a
recent proposal for producing Werner states in a system
of atoms inside a cavity [24]. On the other hand, in the
symmetric subspace, entanglement is dynamically generated from any initial condition, even from initially separable states. However, even for optimal parameters, the
amount of entanglement (given by the concurrence [25])
in this case is only about 10% of the Bell state’s value [22].
Now we can complete the scenario by introducing the
description of the measurement scheme and feedback.
The idea is depicted in Fig. 1: the cavity output is measured by a photo-detector D whose signal provides the
input to the application of the control Hamiltonian F .
Note that, in this kind of monitoring, the absence of signal predominates and the control is only triggered after
a detection click, i.e. a quantum jump, occurs. The unconditioned master equation for this case was derived by
Wiseman in [15] and, for our system, it reads
ρ̇ = −iΩ [(J+ + J− ), ρ] + ΓD[Ufb J− ]ρ.
(5)
The jump feedback is easier to interpret when one
write the last term of Eq. 5 explicitly: D[Ufb J− ] =
†
Ufb J− ρJ+ Ufb
− (J+ J− ρ + ρJ+ J− )/2. The unitary transformation Ufb = exp [−iF δt/~], representing the finite
amount of evolution imposed by the control Hamiltonian
on the system, is only applied immediately after a detection (or jump) event, which is described by the action of
J− in the first term of the superoperator D[J− ]. Note
that the anti-symmetric Bell state |4i remains a stationary solution independently of the form of Ufb .
Once the measurement prescription has been chosen,
the freedom to design a feedback to produce a steady
state with the desired properties lies in the different
choices for the feedback operator Ufb . Although this
represents an enormous range of possibilities, even when
considering the limitations imposed by experimental constraints, here we will restrict to two different cases.
FIG. 2: (Color online) Steady state concurrence as a function
of feedback and driving strengths for Jx (left) and σx (right)
control, with (bottom) or without (top) spontaneous emission
effects. While the influence of decay is pronounced in the Jx
control, the local strategy is left basically unaltered. In this
case, and with γ/Γ = 0.01, a highly entangled state (c = 0.95)
is stabilised for almost all parameters.
Our first choice is the feedback Ufb = exp[−iλ̃Jx ] that
preserves the symmetry properties with respect to exchange of atoms. This coincides with the driving Hamiltonian in Eq. (2), and was used in [21] to show that a feedback control based on homodyne detection can increase
the asymptotic entanglement by 3 times as compared to
the non-controlled case. Although this is a substantial
augment, it corresponds to only 30% of the maximum
possible value for the entanglement. In the jump-based
feedback, the steady state solutions in the symmetric subspace can be analytically calculated from Eq. (5) and
the corresponding entanglement are shown as a function
of the driving and feedback strengths in Fig. 2a. The
maximum concurrence, c ≈ 0.45, occurs at Ω/Γ ≈ 0.08
and λ̃ ≈ ± − 1.49, exceeding the value obtained via
homodyne-based feedback. Despite this further improvement, the absolute value of the entanglement provided
by this feedback strategy is still far from the maximum
value c = 1. Moreover, its performance will worsen considerably when spontaneous emission is added to Eq. (2),
as shown in Fig. 2a for γ1 = γ2 ≡ γ = 0.01 Γ. Even
with such a small value, atomic decay plays a crucial role
here: it breaks the symmetry of the system, opening the
way for interference between different subspaces to occur. This is the explanation of the disappearance of the
higher peaks from Fig. 2a to Fig. 2b: the corresponding
stationary solution shows a large component of the state
|2i that interferes destructively with the state |4i, which
is driven by the anti-symmetric part of the dynamics.
An alternative would be to choose a feedback that,
like the spontaneous emission term, does not preserve the
symmetry with respect to exchange of atoms. This may
allow the feedback to move population between the sub-
3
spaces, limiting the possibilities of destructive interference. Fortunately, as we shall show, this hope is realised
with the simplest choice, yet experimentally realistic [26],
of a symmetry-breaking Hamiltonian, where the control
acts on just one of the atoms. We call this local feedback
and represent its action as Ufb = U1 ⊗ I. One can now
replace this form in Eq. (2), project it in the basis defined by Eqs. (3) and (4) using a general unitary U1 , and
then set ρ̇ = 0 to find the stationary states. The system
of 16 equations obtained from this procedure admits a
single solution, namely the anti-symmetric Bell state |4i.
Therefore, opposed to the uncontrolled case where this
state has to be produced beforehand and is then unaffected by the dynamics, now a pure maximally entangled
state is dynamically generated for all initial conditions
and parameters (excluding the trivial cases of absence of
feedback or driving). We illustrate this in Fig. 2c for the
particular choice U1 = exp[−iλ̃σx ], with σx = σ1 + σ1† .
But it is only when spontaneous emission is taken into
account that the advantage of jump-based local feedback
turns to be really remarkable. A comparison between
the time evolution of entanglement for an initial antisymmetric Bell state in the controlled and non-controlled
cases in the presence of atomic decay, and U1 as defined
above, is shown in Fig. 3. Spontaneous emission takes the
system away from the anti-symmetric subspace and the
dynamics without feedback is not able to restore the Bell
state |4i (cases C1 and C2 for γ/Γ = 0.001 and 0.01, respectively). Instead, decay terms increase the symmetric
component as they move the system to its ground state
while the other terms will tend to drive it to the steady
state of Eq. (2). Conversely, in the presence of feedback, the system evolves under two competing dynamics: while the feedback pushes the state |ggi to |4i, spontaneous decay forces the system the other way around.
The final steady state entanglement is, therefore, set by
the balance between those two opposing tendencies, decreasing exponentially with the ratio γ/Γ. Remarkably,
the proposed control also works when different decoherence sources are considered: in case C3 of Fig. 3, besides spontaneous emission, extra local dephasing terms
γdeph D[σi† σi ] were added and a high value of steady state
entanglement is still obtained for γdeph = γ = 0.01 Γ.
Consequently, as soon as the ratio between decoherence
and collective decay rates remain small, the final entangled state will be very close to the anti-symetric Bell
state as shown by the control curves in Fig. 3. For spontaneous emission rates (C1 and C2 ) consistent with recent experimental parameters [27], concurrence remains
above c ≈ 0.95. Note that this performance is only possible with the combination of a local control and jumpbased feedback. If the same control was used under a
homodyne-based feedback, the maximum attainable entanglement would be c ≈ 0.72 (γ = 0) and c ≈ 0.60
(γ = 0.01 Γ), for Ω/Γ ≈ 0.18 and λ̃ ≈ ± − 0.12 [28].
The behaviour as a function of λ̃ and Ω for this choice
of feedback Hamiltonian remains basically unchanged
(Fig. 2d), as compared with Fig. 2c, with only the val-
FIG. 3: Time evolution of concurrence for Ω = 3Γ and initial
state |4i in the presence of spontaneous emission effects for
the cases where γ = 0.001 Γ (C1 ) and γ = 0.01 Γ (C2 ). Remarkably, the local σx feedback control with λ̃ = π/2 is able
to stabilise the state with large amount of entanglement even
when an extra dephasing is added (γdeph = γ = 0.01 Γ in C3 ).
ues of stationary entanglement slightly reduced. This is
again remarkable, as errors or fluctuations in the feedback and driving strengths would not affect significantly
the final entanglement. In the absence of decay, the
steady state was independent of the form of U1 , provided
it was not the identity, which corresponds to the absence
of feedback. However, the rate at which feedback induced
that steady state was affected by the choice of control. As
the high entanglement in the presence of decay is due to
a competition between the decay and the feedback rate,
it is not surprising that different choices of U1 lead to different steady state values. Fortunately, these variations
are very small, particularly for perturbations around our
choice of σx . Thus our scheme will be robust against
imperfections in the feedback Hamiltonian.
Up to now, our analysis has neglected effects of inefficiencies in the detection process, which may be important, as feedback relies on the manipulation of the system
based on information gained by the measurement. In the
jump feedback case, the extension of Eq. (5) to allow
for a inefficient detection can be done by identifying two
distinct situations when a jump occurs: in the first the
detector clicks and the feedback transformation Ufb is applied, in the second the detector fails to click and there
is no control action. The corresponding equation reads
i
ρ̇ = − Ω [(J+ + J− ), ρ]+ΓηD[Ufb J− ]ρ+Γ (1 − η) D[J− ]ρ.
~
(6)
When the detector efficiency η is zero, no information
is extracted from the measurement and the equation reduces to Eq. (2) where no feedback is applied. Evidently,
in the limit of perfect detection Eq. (5) is regained, and,
for a local control, and without spontaneous decay, a
maximally entangled steady state is reached. In the intermediate case where 0 < η < 1, one would expect that
imperfect knowledge gain should lead to a worse control.
However, the anti-symmetric Bell state is a steady state
of both Eqs. (2) and (5), and one can show, proceeding
4
less efficient detection will represent an effectively weaker
control and therefore a change in the balance between decay and feedback dynamics. However, in the limit considered here (Γ/γ ≫ 1) this decrease is small (see Fig. 4).
FIG. 4: Effect of different detection efficiencies on entanglement evolution for a σx control with Ω = 0.4Γ, λ̃ = π/2,
and both atoms initially in the ground state. Without decay
the system always reaches the anti-symmetric Bell state but
at times depending on the efficiency, while with spontaneous
emission there is a small decrease in the final entanglement.
In conclusion, we have proposed a quantum jump
based feedback scheme to prepare and stabilise highly
entangled states in the presence of spontaneous emission
and decoherence. By monitoring the environment and
feeding back with a suitable interaction, in our case a
local one, we were able to modify the master equation in
such a way that its steady state coincides with the target
state. The strategy works when this engineered dynamics
possesses a single stationary state and is much stronger
than the undesired decay effects [29]. The scheme performs well in the presence of detection inefficiencies and
is also robust against imperfections in the preparation of
the feedback Hamiltonian.
exactly as in the unit-efficiency case, that this also holds
true for Eq. (6) for any η > 0. The effect is illustrated in
Fig. 4 where the time evolution of concurrence is shown
for a fixed driving frequency Ω = 0.4Γ, and for the σx
feedback Hamiltonian with λ̃ = π/2. Without atomic
decay, a non-unit detection efficiency (η = 0.5) simply
delays the time at which stationarity is achieved as compared to the η = 1 case. In the presence of atomic decay
there will be also a decrease in the asymptotic entanglement as η decreases. This should be also expected as
Evidently, many different aspects of the problem remain open. A comparison between feedback schemes using different monitoring methods and controls is under
investigation and will be presented elsewhere [28]. Also
relevant, from the point of view of scalability requirements in quantum information, would be the extension
to higher number of atoms. Would this strategy also be
efficient for a larger number of them? Even in the case
of two atoms the possibilities are vast and the question
whether the present results can be further improved by
designing better feedback schemes is open.
[1] D. Bouwmeester, J.-W. Pan, M. Daniell, H. Weinfurter,
and A. Zeilinger, Phys. Rev. Lett. 82, 1345 (1999).
[2] A. Rauschenbeutel, G. Nogues, S. Osnaghi, P. Bertet,
M. Brune, J.-M. Raimond, and S. Haroche, Science 288,
2024 (2000).
[3] C. A. Sackett, D. Kielpinski, B. E. King, C. Langer,
V. Meyer, C. J. Myatt, M. Rowe, Q. A. Turchette, W. M.
Itano, D. J. Wineland, et al., Nature 404, 256 (2000).
[4] J.-W. Pan, M. Daniell, S. Gasparoni, G. Weihs, and
A. Zeilinger, Phys. Rev. Lett. 86, 4435 (2001).
[5] C. F. Roos, M. Riebe, H. Häffner, W. Hänsel, J. B. elm,
G. P. T. Lancaster, C. Becher, F. Schmidt-Kaler, and
R. Blatt, Science 304, 1478 (2004).
[6] T. Yu and J. H. Eberly, Phys. Rev. Lett. 93, 140404
(2004).
[7] A. R. R. Carvalho, F. Mintert, and A. Buchleitner, Phys.
Rev. Lett. 93, 230501 (2004).
[8] C. F. Roos, G. P. T. Lancaster, M. Riebe, H. Häffner,
W. Hänsel, S. Gulde, C. Becher, J. Eschner, F. SchmidtKaler, and R. Blatt, Phys. Rev. Lett. 92, 220402 (2004).
[9] J. M. Geremia, J. K. Stockton, and H. Mabuchi, Science
304, 270 (2004).
[10] J. E. Reiner, W. P. Smith, L. A. Orozco, H. M. Wiseman,
and J. Gambetta, Phys. Rev. A 70, 023819 (2004).
[11] N. V. Morrow, S. K. Dutta, and G. Raithel, Phys. Rev.
Lett. 88, 093003 (2002).
[12] P. Bushev, D. Rotter, A. Wilson, F. Dubin, C. Becher,
J. Eschner, R. Blatt, V. Steixner, P. Rabl, and P. Zoller,
Phys. Rev. Lett. 96, 043003 (2006).
V. P. Belavkin, Rep. Math. Phys. 43, 405 (1999).
H. M. Wiseman and G. J. Milburn, Phys. Rev. Lett. 70,
548 (1993).
H. M. Wiseman, Phys. Rev. A 49, 2133 (1994).
A. C. Doherty and K. Jacobs, Phys. Rev. A 60, 2700
(1999).
S. Mancini, Phys. Rev. A 73, 010304(R) (2006).
S. Mancini and H. M. Wiseman (2006), preprint arXiv
quant-ph/0610006.
J. K. Stockton, R. van Handel, and H. Mabuchi, Phys.
Rev. A 70, 022106 (2004).
S. Mancini and J. Wang, Eur. Phys. J. D 32, 257 (2005).
J. Wang, H. M. Wiseman, and G. J. Milburn, Phys. Rev.
A 71, 042309 (2005).
S. Schneider and G. J. Milburn, Phys. Rev. A 65, 042107
(2002).
G. S. Agarwal, Springer Tracs Mod. Phys. 70, 1 (1974).
G. S. Agarwal and K. T. Kapale, Phys. Rev. A 73, 022315
(2006).
W. K. Wootters, Phys. Rev. Lett. 80, 2245 (1998).
S. Nußmann, M. Hijlkema, B. Weber, F. Rohde,
G. Rempe, and A. Kuhn, Phys. Rev. Lett. 95, 173602
(2005).
A. D. Boozer, A. Boca, R. Miller, T. E. Northup, and
H. J. Kimble, Phys. Rev. Lett. 97, 083602 (2006).
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
5
[28] A. R. R. Carvalho and J. J. Hope (2006), in preparation.
[29] A. R. R. Carvalho, P. Milman, R. L. de Matos Filho, and
L. Davidovich, Phys. Rev. Lett. 86, 4988 (2001).