PHYSICAL REVIEW A 69, 062320 (2004)

Cavity quantum electrodynamics for superconducting electrical circuits:

An architecture for quantum computation
Alexandre Blais,1 Ren-Shou Huang,1,2 Andreas Wallraff,1 S. M. Girvin,1 and R. J. Schoelkopf1
1
Departments of Physics and Applied Physics, Yale University, New Haven, Connecticut 06520, USA
2
Department of Physics, Indiana University, Bloomington, Indiana 47405, USA
(Received 7 February 2004; published 29 June 2004; corrected 23 July 2004)

We propose a realizable architecture using one-dimensional transmission line resonators to reach the strong-
coupling limit of cavity quantum electrodynamics in superconducting electrical circuits. The vacuum Rabi
frequency for the coupling of cavity photons to quantized excitations of an adjacent electrical circuit (qubit)
can easily exceed the damping rates of both the cavity and qubit. This architecture is attractive both as a
macroscopic analog of atomic physics experiments and for quantum computing and control, since it provides
strong inhibition of spontaneous emission, potentially leading to greatly enhanced qubit lifetimes, allows
high-fidelity quantum nondemolition measurements of the state of multiple qubits, and has a natural mecha-
nism for entanglement of qubits separated by centimeter distances. In addition it would allow production of
microwave photon states of fundamental importance for quantum communication.

DOI: 10.1103/PhysRevA.69.062320 PACS number(s): 03.67.Lx, 73.23.Hk, 74.50.⫹r, 32.80.⫺t

I. INTRODUCTION leads to lifetime enhancement of the qubit. In Sec. VI, a

quantum nondemolition readout protocol is presented. Real-
Cavity quantum electrodynamics (CQED) studies the ization of one-qubit logical operations is discussed in Sec.
properties of atoms coupled to discrete photon modes in high VII and two-qubit entanglement in Sec. VIII. We show in
Q cavities. Such systems are of great interest in the study of Sec. IX how to take advantage of encoded universality and
the fundamental quantum mechanics of open systems, the decoherence-free subspace in this system.
engineering of quantum states, and measurement-induced de-
coherence [1–3] and have also been proposed as possible
candidates for use in quantum information processing and
transmission [1–3]. Ideas for novel CQED analogs using na- II. BRIEF REVIEW OF CAVITY QED
nomechanical resonators have recently been suggested by
Schwab and collaborators [4,5]. We present here a realistic
Cavity QED studies the interaction between atoms and the
proposal for CQED via Cooper pair boxes coupled to a one-
quantized electromagnetic modes inside a cavity. In the op-
dimensional (1D) transmission line resonator, within a
simple circuit that can be fabricated on a single microelec- tical version of CQED [2], schematically shown in Fig. 1(a),
tronic chip. As we discuss, 1D cavities offer a number of one drives the cavity with a laser and monitors changes in
practical advantages in reaching the strong-coupling limit of the cavity transmission resulting from coupling to atoms fall-
CQED over previous proposals using discrete LC circuits ing through the cavity. One can also monitor the spontaneous
[6,7], large Josephson junctions [8–10], or 3D cavities emission of the atoms into transverse modes not confined by
[11–13]. Besides the potential for entangling qubits to realize the cavity. It is not generally possible to directly determine
two-qubit gates addressed in those works, in the present the state of the atoms after they have passed through the
work we show that the CQED approach also gives strong cavity because the spontaneous emission lifetime is on the
and controllable isolation of the qubits from the electromag- scale of nanoseconds. One can, however, infer information
netic environment, permits high-fidelity quantum nondemo- about the state of the atoms inside the cavity from real-time
lition (QND) readout of multiple qubits, and can produce monitoring of the cavity optical transmission.
states of microwave photon fields suitable for quantum com- In the microwave version of CQED [3], one uses a very-
munication. The proposed circuits therefore provide a simple high-Q superconducting 3D resonator to couple photons to
and efficient architecture for solid-state quantum computa- transitions in Rydberg atoms. Here one does not directly
tion, in addition to opening up a new avenue for the study of monitor the state of the photons, but is able to determine
entanglement and quantum measurement physics with mac- with high efficiency the state of the atoms after they have
roscopic objects. We will frame our discussion in a way that passed through the cavity (since the excited state lifetime is
makes contact between the language of atomic physics and of the order of 30 ms). From this state-selective detection
that of electrical engineering. one can infer information about the state of the photons in
We begin in Sec. II with a brief general overview of the cavity.
CQED before turning to a discussion of our proposed solid- The key parameters describing a CQED system (see Table
state realization of cavity QED in Sec. III. We then discuss in I) are the cavity resonance frequency ␻r, the atomic transi-
Sec. IV the case where the cavity and qubit are tuned in tion frequency ⍀, and the strength of the atom-photon cou-
resonance and in Sec. V the case of large detuning which pling g appearing in the Jaynes-Cummings Hamiltonian [14]

1050-2947/2004/69(6)/062320(14)/$22.50 69 062320-1 ©2004 The American Physical Society

BLAIS et al. PHYSICAL REVIEW A 69, 062320 (2004)

transit time ttransit of the atom through the cavity.

In the absence of damping, exact diagonalization of the
Jaynes-Cumming Hamiltonian yields the excited eigenstates
(dressed states) [15]

兩+ ,n典 = cos ␪n兩↓,n典 + sin ␪n兩↑,n + 1典, 共2兲

兩− ,n典 = − sin ␪n兩↓,n典 + cos ␪n兩↑,n + 1典, 共3兲

and ground state 兩↑ , 0典 with corresponding eigenenergies

ប
E ±,n = 共n + 1兲ប␻r ± 冑4g2共n + 1兲 + ⌬2 , 共4兲
2

ប⌬
E↑,0 = − . 共5兲
2
FIG. 1. (Color online) (a) Standard representation of a cavity In these expressions,
quantum electrodynamic system, comprising a single mode of the
electromagnetic field in a cavity with decay rate ␬ coupled with a
coupling strength g = Ermsd / ប to a two-level system with spontane-
ous decay rate ␥ and cavity transit time ttransit. (b) Energy spectrum
1
␪n = tan−1
2
冉
2g冑n + 1
⌬
, 冊 共6兲

of the uncoupled (left and right) and dressed (center) atom-photon

and ⌬ ⬅ ⍀ − ␻r the atom-cavity detuning.
states in the case of zero detuning. The degeneracy of the two-
dimensional manifolds of states with n − 1 quanta is lifted by
Figure 1(b) shows the spectrum of these dressed states for
2g冑n + 1. (c) Energy spectrum in the dispersive regime (long- the case of zero detuning, ⌬ = 0, between the atom and cavity.
dashed lines). To second order in g, the level separation is indepen- In this situation, degeneracy of the pair of states with n + 1
dent of n, but depends on the state of the atom. quanta is lifted by 2g冑n + 1 due to the atom-photon interac-
tion. In the manifold with a single excitation, Eqs. (2) and (3)

冉 冊
reduce to the maximally entangled atom-field states 兩± , 0典
1 ប⍀ z = 共兩↑ , 1典 ± 兩↓ , 0典兲 / 冑2. An initial state with an excited atom and
H = ប ␻ r a †a + + ␴ + បg共a†␴− + ␴+a兲 + H␬ + H␥ .
2 2 zero photons 兩↑ , 0典 will therefore flop into a photon 兩↓ , 1典 and
共1兲 back again at the vacuum Rabi frequency g / ␲. Since the
excitation is half atom and half photon, the decay rate of
Here H␬ describes the coupling of the cavity to the con- 兩± , 0典 is 共␬ + ␥兲 / 2. The pair of states 兩± , 0典 will be resolved in
tinuum which produces the cavity decay rate ␬ = ␻r / Q, while a transmission experiment if the splitting 2g is larger than
H␥ describes the coupling of the atom to modes other than this linewidth. The value of g = Ermsd / ប is determined by the
the cavity mode which cause the excited state to decay at rate transition dipole moment d and the rms zero-point electric
␥ (and possibly also produce additional dephasing effects). field of the cavity mode. Strong coupling is achieved when
An additional important parameter in the atomic case is the g Ⰷ ␬ , ␥ [15].

TABLE I. Key rates and CQED parameters for optical [2] and microwave [3] atomic systems using 3D cavities, compared against the
proposed approach using superconducting circuits, showing the possibility for attaining the strong cavity QED limit 共nRabi Ⰷ 1兲. For the 1D
superconducting system, a full-wave 共L = ␭兲 resonator, ␻r / 2␲ = 10 GHz, a relatively low Q of 104, and coupling ␤ = Cg / C⌺ = 0.1 are assumed.
For the 3D microwave case, the number of Rabi flops is limited by the transit time. For the 1D circuit case, the intrinsic Cooper-pair box
decay rate is unknown; a conservative value equal to the current experimental upper bound ␥ 艋 1 / 共2 ␮s兲 is assumed.

Parameter Symbol 3D optical 3D microwave 1D circuit

Resonance or transition frequency ␻r / 2␲, ⍀ / 2␲ 350 THz 51 GHz 10 GHz

Vacuum Rabi frequency g / ␲, g / ␻r 220 MHz, 3 ⫻ 10−7 47 kHz, 1 ⫻ 10−7 100 MHz, 5 ⫻ 10−3
Transition dipole d / ea0 ⬃1 1 ⫻ 103 2 ⫻ 104
Cavity lifetime 1/␬,Q 10 ns, 3 ⫻ 107 1 ms, 3 ⫻ 108 160 ns, 104
Atom lifetime 1/␥ 61 ns 30 ms 2 ␮s
Atom transit time ttransit 艌50 ␮s 100 ␮s ⬁
Critical atom number N0 = 2␥␬ / g2 6 ⫻ 10−3 3 ⫻ 10−6 艋6 ⫻ 10−5
Critical photon number m0 = ␥2 / 2g2 3 ⫻ 10−4 3 ⫻ 10−8 艋1 ⫻ 10−6
Number of vacuum Rabi flops nRabi = 2g / 共␬ + ␥兲 ⬃10 ⬃5 ⬃102

062320-2
CAVITY QUANTUM ELECTRODYNAMICS FOR… PHYSICAL REVIEW A 69, 062320 (2004)

For large detuning, g / ⌬ Ⰶ 1, expansion of Eq. (4) yields

the dispersive spectrum shown in Fig. 1(c). In this situation,
the eigenstates of the one excitation manifold take the form
[15]
兩− ,0典 ⬃ − 共g/⌬兲兩↓,0典 + 兩↑,1典, 共7兲

兩+ ,0典 ⬃ 兩↓,0典 + 共g/⌬兲兩↑,1典. 共8兲

The corresponding decay rates are then simply given by
⌫− ,0 ⯝ 共g/⌬兲2␥ + ␬ , 共9兲

FIG. 2. (Color online). Schematic layout and equivalent lumped

⌫+ ,0 ⯝ ␥ + 共g/⌬兲2␬ . 共10兲 circuit representation of proposed implementation of cavity QED
More insight into the dispersive regime is gained by mak- using superconducting circuits. The 1D transmission line resonator
ing the unitary transformation consists of a full-wave section of superconducting coplanar wave-

冋 册
guide, which may be lithographically fabricated using conventional
g optical lithography. A Cooper-pair box qubit is placed between the
U = exp 共a␴+ − a†␴−兲 共11兲 superconducting lines and is capacitively coupled to the center trace
⌬
at a maximum of the voltage standing wave, yielding a strong elec-
and expanding to second order in g (neglecting damping for tric dipole interaction between the qubit and a single photon in the
the moment) to obtain cavity. The box consists of two small 共⬃100 nm⫻ 100 nm兲 Joseph-

冋 册 冋 册
son junctions, configured in a ⬃1 ␮m loop to permit tuning of the
g2 z † ប g2 z effective Josephson energy by an external flux ⌽ext. Input and out-
UHU† ⬇ ប ␻r + ␴ a a+ ⍀+ ␴. 共12兲
⌬ 2 ⌬ put signals are coupled to the resonator, via the capacitive gaps in
the center line, from 50⍀ transmission lines which allow measure-
As is clear from this expression, the atom transition is ac ments of the amplitude and phase of the cavity transmission, and
Stark/Lamb shifted by 共g2 / ⌬兲共n + 1 / 2兲. Alternatively, one the introduction of dc and rf pulses to manipulate the qubit states.
can interpret the ac Stark shift as a dispersive shift of the Multiple qubits (not shown) can be similarly placed at different
cavity transition by ␴zg2 / ⌬. In other words, the atom pulls antinodes of the standing wave to generate entanglement and two-
the cavity frequency by ±g2 / ␬⌬. bit quantum gates across distances of several millimeters.

In addition to the small effective volume and the fact that

III. CIRCUIT IMPLEMENTATION OF CAVITY QED the on-chip realization of CQED shown in Fig. 2 can be
fabricated with existing lithographic techniques, a
We now consider the proposed realization of cavity QED
transmission-line resonator geometry offers other practical
using the superconducing circuits shown in Fig. 2. A 1D
advantages over lumped LC circuits or current-biased large
transmission line resonator consisting of a full-wave section
Josephson junctions. The qubit can be placed within the cav-
of superconducting coplanar waveguide plays the role of the
ity formed by the transmission line to strongly suppress the
cavity and a superconducting qubit plays the role of the
spontaneous emission, in contrast to a lumped LC circuit,
atom. A number of superconducting quantum circuits could
where without additional special filtering, radiation and para-
function as artificial atom, but for definiteness we focus here
sitic resonances may be induced in the wiring [20]. Since the
on the Cooper-pair box [6,16–18].
resonant frequency of the transmission line is determined
primarily by a fixed geometry, its reproducibility and immu-
A. Cavity: Coplanar stripline resonator nity to 1 / f noise should be superior to Josephson junction
plasma oscillators. Finally, transmission-line resonances in
An important advantage of this approach is that the zero- coplanar waveguides with Q ⬃ 106 have already been dem-
point energy is distributed over a very small effective volume onstrated [21,22], suggesting that the internal losses can be
(⬇10−5 cubic wavelengths) for our choice of a quasi-one- very low. The optimal choice of the resonator Q in this ap-
dimensional transmission line “cavity.” As shown in Appen- proach is strongly dependent on the intrinsic decay rates of
0
dix A, this leads to significant rms voltages Vrms ⬃ 冑ប␻r / cL superconducting qubits which, as described below, are pres-
between the center conductor and the adjacent ground plane ently unknown, but can be determined with the setup pro-
at the antinodal positions, where L is the resonator length and posed here. Here we assume the conservative case of an
c is the capacitance per unit length of the transmission line. overcoupled resonator with a Q ⬃ 104, which is preferable for
At a resonant frequency of 10 GHz 共h␯ / kB ⬃ 0.5 K兲 and for the first experiments.
a 10 ␮m gap between the center conductor and the adjacent
ground plane, Vrms ⬃ 2 ␮V corresponding to electric fields
B. Artificial atom: The Cooper-pair box
Erms ⬃ 0.2 V / m, some 100 times larger than achieved in the
3D cavity described in Ref. [3]. Thus, this geometry might Our choice of “atom,” the Cooper-pair box [6,16], is a
also be useful for coupling to Rydberg atoms [19]. mesoscopic superconducting island. As shown in Fig. 3, the

062320-3
BLAIS et al. PHYSICAL REVIEW A 69, 062320 (2004)

part v. As shown in Appendix A, if the qubit is placed in the

center of the resonator, this latter contribution is given by
0
v = Vrms 共a† + a兲. Taking into account both Vdc g and v in
Eq. (14), we obtain

HQ = − 2EC共1 − 2Ndc ␴z −
g 兲¯
EJ x
2
C
¯␴ − e g
C⌺
冑 ប␻r †
Lc
共a + a兲

⫻共1 − 2Ng − ¯␴z兲. 共15兲

FIG. 3. Circuit diagram of the Cooper-pair box. The gate volt- Working in the eigenbasis 兵兩 ↑ 典 , 兩 ↓ 典其 of the first two terms of
age is connected to the island through an environmental impedance the above expression [23] and adding the Hamiltonian of the
Z共␻兲. oscillator mode coupled to the qubit, the Hamiltonian of the
interacting qubit and resonator system takes the form

冉 冊 冑
island is connected to a large reservoir through a Josephson
junction with Josephson energy EJ and capacitance CJ. It is 1 ប⍀ z Cg ប␻r †
H = ប ␻ r a †a + + ␴ −e 共a + a兲
voltage biased from a lead having capacitance Cg to the is- 2 2 C⌺ Lc
land. If the superconducting gap is larger than both the
⫻关1 − 2Ng − cos共␪兲␴z + sin共␪兲␴x兴. 共16兲
charging energy Ec = e2 / 2C⌺ (where C⌺ = CJ + Cg is the total
box capacitance) and temperature, the only relevant degree Here, ␴x and ␴z are Pauli matrices in the eigenbasis
of freedom is the number of Cooper pairs N on the island. In 兵兩 ↑ 典 , 兩 ↓ 典其, ␪ = arctan关EJ / 4EC共1 − 2Ndc
g 兲兴 is the mixing angle,
this basis, the Hamiltonian describing the superconducting
and the energy splitting of the qubit is ⍀
= 冑E2J + 关4EC共1 − 2Ndc
island takes the form
g 兲兴 / ប [23]. Note that contrary to the
2

E case of a qubit fabricated outside the cavity where the N2g

HQ = 4Ec 兺N 共N − Ng兲2兩N典具N兩− 2J 兺N 共兩N + 1典具N兩 + H.c.兲, term in Eq. (13) has no effect, here this term slightly renor-
malizes the cavity frequency ␻r and displaces the oscillator
共13兲 coordinate. These effects are implicit in Eq. (16).
where Ng = CgVg / 2e is the dimensionless gate charge repre- At the charge degeneracy point (where Ndc dc
g = CgVg / 2e
senting the total polarization charge injected into the island = 1 / 2 and ␪ = ␲ / 2), neglecting rapidly oscillating terms and
by the voltage source. omitting damping for the moment, Eq. (16) reduces to the
In the charge regime 4Ec Ⰷ EJ and restricting the gate Jaynes-Cummings Hamiltonian (1) with ⍀ = EJ / ប and cou-
charge to the range Ng 苸 关0 , 1兴, only a pair of adjacent charge pling

冑
states on the island are relevant and the Hamiltonian then
reduces to a 2 ⫻ 2 matrix ␤e ប␻r
g= , 共17兲
ប cL
Eel z EJ x
HQ = − ¯␴ − ¯␴ , 共14兲
2 2 where ␤ ⬅ Cg / C⌺. The quantum electrical circuit of Fig. 2 is
therefore mapped to the problem of a two-level atom inside a
with Eel = 4EC共1 − 2Ng兲. The Cooper-pair box can in this case cavity. Away from the degeneracy point, this mapping can
be mapped to a pseudospin-1 / 2 particle, with effective fields still be performed, but with a coupling strength reduced by
in the x and z directions. sin共␪兲 and an additional term proportional to 共a† + a兲.
Replacing the Josephson junction by a pair of junctions in In this circuit, the “atom” is highly polarizable at the
parallel, each with energy EJ / 2, the effective field in the x charge degeneracy point, having transition dipole moment
direction becomes EJcos共␲⌽ext / ⌽0兲 / 2. By threading a flux d ⬅ បg / Erms ⬃ 2 ⫻ 104 atomic units 共ea0兲, or more than an
⌽ext in the loop formed by the pair of junctions and changing order of magnitude larger than even a typical Rydberg atom
the gate voltage Vg, it is possible to control the effective [15]. An experimentally realistic [18] coupling ␤ ⬃ 0.1 leads
fields acting on the qubit. In the setup of Fig. 2, application to a vacuum Rabi rate g / ␲ ⬃ 100 MHz, which is three orders
of dc gate voltage on the island can be conveniently achieved of magnitude larger than in corresponding atomic microwave
by applying a bias voltage to the center conductor of the CQED experiments [3] or approximately 1% of the transition
transmission line. The resonator coupling capacitance C0, the frequency. Unlike the usual CQED case, these artificial “at-
gate capacitance Cg (the capacitance between the center con- oms” remain at fixed positions indefinitely and so do not
ductor of the resonator and the island), and the capacitance to suffer from the problem that the coupling g varies with po-
ground of the resonator then act as a voltage divider. sition in the cavity.
A comparison of the experimental parameters for imple-
mentations of cavity QED with optical and microwave
C. Combined system: Superconducting cavity QED
atomic systems and for the proposed implementation with
For a superconducting island fabricated inside a resonator, superconducting circuits is presented in Table I. We assume
in addition to a dc part Vdc
g , the gate voltage has a quantum here a relatively low Q = 104 and a worst case estimate, con-

062320-4
CAVITY QUANTUM ELECTRODYNAMICS FOR… PHYSICAL REVIEW A 69, 062320 (2004)

A conservative estimate of the noise energy for a 10 GHz

cryogenic high-electron-mobility (HEMT) amplifier is namp
= kBTN / ប␻r ⬃ 100 photons, where TN is the noise tempera-
ture of the amplification circuit. As a result, these spectral
features should be readily observable in a measurement time
tmeas = 2namp / 具n典␬ or only ⬃32 ␮s for 具n典 ⬃ 1.

V. LARGE DETUNING: LIFETIME ENHANCEMENT

FIG. 4. Expected transmission spectrum of the resonator in the
absence (dashed line) and presence (solid line) of a superconducting For qubits not inside a cavity, fluctuation of the gate volt-
qubit biased at its degeneracy point. Parameters are those presented age acting on the qubit is an important source of relaxation
in Table I. The splitting exceeds the line width by two orders of and dephasing. As shown in Fig. 3, in practice the qubit’s
magnitude. gate is connected to the voltage source through external wir-
ing having, at the typical microwave transition frequency of
sistent with the bound set by previous experiments with su- the qubit, a real impedance of value close to the impedance
perconducting qubits (discussed further below), for the in- of free space 共⬃50 ⍀兲. The relaxation rate expected from
trinsic qubit lifetime of 1 / ␥ 艌 2 ␮s. purely quantum fluctuations across this impedance (sponta-
The standard figures of merit [24] for strong coupling are neous emission) is [18,23]

冉冊
the critical photon number needed to saturate the atom on
resonance, m0 = ␥2 / 2g2 艋 1 ⫻ 10−6, and the minimum atom 1 E2 e 2
= 2 J 2 ␤2SV共+ ⍀兲, 共18兲
number detectable by measurement of the cavity output, N0 T1 EJ + Eel ប
= 2␥␬ / g2 艋 6 ⫻ 10−5. These remarkably low values are
where SV共+⍀兲 = 2ប⍀ Re关Z共⍀兲兴 is the spectral density of
clearly very favorable and show that superconducting cir-
voltage fluctuations across the environmental impedance (in
cuits could access the interesting regime of very strong cou-
the quantum limit). It is difficult in most experiments to pre-
pling.
cisely determine the real part of the high-frequency environ-
mental impedance presented by the leads connected to the
IV. ZERO DETUNING qubit, but reasonable estimates [18] yield values of T1 in the
range of 1 ␮s.
In the case of a low-Q cavity 共g ⬍ ␬兲 and zero detuning, For qubits fabricated inside a cavity, the noise across the
the radiative decay rate of the qubit into the transmission line environmental impedance does not couple directly to the qu-
becomes strongly enhanced by a factor of Q relative to the bit, but only indirectly through the cavity. For the case of
rate in the absence of the cavity [15]. This is due to the strong detuning, coupling of the qubit to the continuum is
resonant enhancement of the density of states at the atomic therefore substantially reduced. One can view the effect of
transition frequency. In electrical engineering language, the the detuned resonator as filtering out the vacuum noise at the
⬃50⍀ external transmission-line impedance is transformed qubit transition frequency or, in electrical engineering terms,
on resonance to a high value which is better matched to as providing an impedance transformation which strongly
extract energy from the qubit. reduces the real part of the environmental impedance seen by
For strong coupling g ⬎ ␬ , ␥, the first excited state be- the qubit.
comes a doublet with linewidth 共␬ + ␥兲 / 2, as explained in Solving for the normal modes of the resonator and trans-
Sec. II. As can be seen from Table I, the coupling in the mission lines, including an input impedance R at each end of
proposed superconducting implementation is so strong that, the resonator, the spectrum of voltage fluctuations as seen by
even for the low Q = 104 we have assumed, 2g / 共␬ + ␥兲 the qubit fabricated in the center of the resonator can be
shown to be well approximated by
⬃ 100 vacuum Rabi oscillations are possible. Moreover, as
shown in Fig. 4, the frequency splitting 共g / ␲ ⬃ 100 MHz兲 2ប␻r ␬/2
will be readily resolvable in the transmission spectrum of the SV共⍀兲 = . 共19兲
Lc ⌬ + 共␬/2兲2
2
resonator. This spectrum, calculated here following Ref.
[25], can be observed in the same manner as employed in Using this transformed spectral density in Eq. (18) and as-
optical atomic experiments, with a continuous-wave mea- suming a large detuning between the cavity and qubit, the
surement at low drive, and will be of practical use to find the relaxation rate due to vacuum fluctuations takes a form that
dc gate voltage needed to tune the box into resonance with reduces to 1 / T1 ⬅ ␥␬ = 共g / ⌬兲2␬ ⬃ 1 / 共64 ␮s兲, at the qubit’s
the cavity. degeneracy point. This is the result already obtained in Eq.
Of more fundamental importance than this simple avoided (10) using the dressed-state picture for the coupled atom and
level crossing, however, is the fact that the Rabi splitting cavity, except for the additional factor ␥ reflecting a loss of
scales with the square root of the photon number, making the energy to modes outside of the cavity. For large detuning,
level spacing anharmonic. This should cause a number of damping due to spontaneous emission can be much less
novel nonlinear effects [14] to appear in the spectrum at than ␬.
higher drive powers when the average photon number in the One of the important motivations for this CQED experi-
cavity is large 共具n典 ⬎ 1兲. ment is to determine the various contributions to the qubit

062320-5
BLAIS et al. PHYSICAL REVIEW A 69, 062320 (2004)

decay rate so that we can understand their fundamental

physical origins as well as engineer improvements. Besides
␥␬ evaluated above, there are two additional contributions to
the total damping rate ␥ = ␥␬ + ␥⬜ + ␥NR. Here ␥⬜ is the decay
rate into photon modes other than the cavity mode and ␥NR is
the rate of other (possibly nonradiative) decays. Optical cavi-
ties are relatively open and ␥⬜ is significant, but for 1D
microwave cavities, ␥⬜ is expected to be negligible (despite
the very large transition dipole). For Rydberg atoms the two
qubit states are both highly excited levels and ␥NR represents
(radiative) decay out of the two-level subspace. For Cooper-
pair boxes, ␥NR is completely unknown at the present time, FIG. 5. (Color online) Transmission spectrum of the cavity,
but could have contributions from phonons, two-level sys- which is “pulled” by an amount ±g2 / ⌬ = ± 2.5␻r ⫻ 10−4, depending
tems in insulating [20] barriers and substrates, or thermally on the state of the qubit (red for the excited state, blue for the
excited quasiparticles. ground state). To perform a measurement of the qubit, a pulse of
For Cooper box qubits not inside a cavity, recent experi- microwave photons, at a probe frequency ␻␮w = ␻r or ␻r ± g2 / ⌬, is
ments [18] have determined a relaxation time 1 / ␥ = T1 sent through the cavity. Additional peaks near ⍀ corresponding to
⬃ 1.3 ␮s despite the backaction of continuous measurement qubit flips are suppressed by g / ⌬.
by a SET electrometer. Vion et al. [17] found T1 ⬃ 1.84 ␮s
(without measurement backaction) for their charge-phase qu- depending on the qubit being in its ground or excited states,
bit. Thus, in these experiments, if there are nonradiative de- the transmission spectrum will present a peak of width ␬ at
cay channels, they are at most comparable to the vacuum ␻r − g2 / ⌬ or ␻r + g2 / ⌬. With the parameters of Table I, this
dispersive pull of the cavity frequency is ±g2 / ␬⌬ = ± 2.5 line-
radiative decay rate (and may well be much less) estimated
widths for a 10% detuning. Exact diagonalization (4) shows
using Eq. (18). Experiments with a cavity will present the
that the pull is power dependent and decreases in magnitude
qubit with a simple and well-controlled electromagnetic en- for cavity photon numbers on the scale n = ncrit ⬅ ⌬2 / 4g2. In
vironment, in which the radiative lifetime can be enhanced the regime of nonlinear response, single-atom optical bista-
with detuning to 1 / ␥␬ ⬎ 64 ␮s, allowing ␥NR to dominate bility [14] can be expected when the drive frequency is off
and yielding valuable information about any nonradiative resonance at low power but on resonance at high power [29].
processes. The state-dependent pull of the cavity frequency by the
qubit can be used to entangle the state of the qubit with that
VI. DISPERSIVE QND READOUT OF QUBITS of the photons transmitted or reflected by the resonator. For
g2 / ␬⌬ ⬎ 1, as in Fig. 5, the pull is greater than the linewidth,
In addition to lifetime enhancement, the dispersive regime and irradiating the cavity at one of the pulled frequencies
is advantageous for readout of the qubit. This can be realized ␻r ± g2 / ⌬, the transmission of the cavity will be close to
by microwave irradiation of the cavity and then probing the unity for one state of the qubit and close to zero for the other
transmitted or reflected photons [26]. [30].
Choosing the drive to be instead at the bare cavity fre-
A. Measurement protocol quency ␻r, the state of the qubit is encoded in the phase of
A drive of frequency ␻␮w on the resonator can be mod- the reflected and transmitted microwaves. An initial qubit
eled by [15] state 兩␹典 = ␣兩 ↑ 典 + ␤兩 ↓ 典 evolves under microwave irradiation
into the entangled state 兩␺典 = ␣兩↑ , ␪典 + ␤兩↓ , −␪典, where tan ␪
H␮w共t兲 = ប␧共t兲共a†e−i␻␮wt + ae+i␻␮wt兲, 共20兲 = 2g2 / ␬⌬ and 兩±␪典 are (interaction representation) coherent
states with the appropriate mean photon number and oppo-
where ␧共t兲 is a measure of the drive amplitude. In the dis-
site phases. In the situation where g2 / ␬⌬ Ⰶ 1, this is the most
persive limit, one expects from Fig. 1(c) peaks in the trans-
appropriate strategy.
mission spectrum at ␻r − g2 / ⌬ and ⍀ + 2g2 / ⌬ if the qubit is
It is interesting to note that such an entangled state can be
initially in its ground state. In a frame rotating at the drive
used to couple qubits in distant resonators and allow quan-
frequency, the matrix elements for these transitions are, re-
tum communication [31]. Moreover, if an independent mea-
spectively,
surement of the qubit state can be made, such states can be
具↑,0兩H␮w兩− ,n典 ⬃ ␧, turned into photon Schrödinger cats [15].
To characterize these two measurement schemes corre-
␧g sponding to two different choices of the drive frequency, we
具↑,0兩H␮w兩+ ,n典 ⬃ . 共21兲 compute the average photon number inside the resonator n̄
⌬ and the homodyne voltage on the 50⍀ impedance at the
In the large detuning case, the peak at ⍀ + 2g2 / ⌬, corre- output of the resonator. Since the power coupled to the out-
sponding approximatively to a qubit flip, is highly sup- side of the resonator is P = 具n典ប␻r␬ / 2 = 具Vout典2 / R, the homo-
pressed. dyne voltage can be expressed as 具Vout典 = 冑Rប␻r␬具a + a†典 / 2
The matrix element corresponding to a qubit flip from the and is proportional to the real part of the field inside the
excited state is also suppressed and, as shown in Fig. 5, cavity.

062320-6
CAVITY QUANTUM ELECTRODYNAMICS FOR… PHYSICAL REVIEW A 69, 062320 (2004)

FIG. 6. (Color online) Results of numerical simulations using FIG. 7. (Color online) Same as Fig. 6 for the drive at the bare
the quantum-state diffusion method. A microwave pulse of duration cavity frequency ␻r. Depending on the qubit’s state, the pulse is
⬃15/ ␬ and centered at the pulled frequency ␻r + g2 / ⌬ drives the either above or below the combined cavity-qubit resonance and so
cavity. (a) The occupation probability of the excited state (right is partly transmitted and reflected but with a large relative phase
axis, solid lines), for the case in which the qubit is initially in the shift that can be detected with homodyne detection. In (b), the op-
ground (blue) or excited (red) state and intracavity photon number posing phase shifts cause a change in sign of the output, which can
(left axis, dash lines), are shown as a function of time. Though the be measured with high signal to noise to realize a single-shot, QND
qubit states are temporarily coherently mixed during the pulse, the measurement of the qubit.
probability of real transitions is seen to be small. Depending on the
qubit’s state, the pulse is either on or away from the combined
tons in the cavity (left axis). Figure 6(b) shows, in a frame
cavity-qubit resonance and therefore is mostly transmitted or mostly
rotating at the drive frequency, the real part of the cavity
reflected. (b) The real component of the cavity electric field ampli-
electric field amplitude (left axis) and transmitted voltage
tude (left axis) and the transmitted voltage phasor (right axis) in the
output transmission line for the two possible initial qubit states. The
phase (right axis) in the output transmission line, again for
parameters used for the simulation are presented in Table I. the two possible initial qubit states. These quantities are
shown in Fig. 7 for a drive at the bare frequency ␻r.
In the absence of dissipation, the time dependence of the As expected, for the first choice of drive frequency, the
field inside the cavity can be obtained in the Heisenberg information about the state of the qubit is mostly stored in
picture from Eqs. (12) and (20). This leads to a closed set of the number of transmitted photons. When the drive is at the
differential equations for a, ␴z, and a␴z which is easily bare frequency, however, there is very little information in
solved. In the presence of dissipation, however [i.e., per- the photon number, with most of the information being
forming the transformation (11) on H␬ and H␥, and adding stored in the phase of the transmitted and reflected signal.
the resulting terms to Eqs. (12) and (20)], the set is no longer This phase shift can be measured using standard heterodyne
closed and we resort to numerical stochastic wave function techniques. As also discussed in Appendix C, both ap-
calculations [32]. See Appendix B for a brief presentation of proaches can serve as a high-efficiency quantum nondemoli-
this numerical method. tion dispersive readout of the state of the qubit.
Figures 6 and 7 show the numerical results for the two
choices of drive frequency and using the parameters of Table
B. Measurement time and backaction
I. For these calculations, a pulse of duration ⬃15/ ␬ with a
hyperbolic tangent rise and fall is used to excite the cavity. As seen from Eq. (12), the backaction of the dispersive
Figure 6 corresponds to a drive at the pulled frequency ␻r CQED measurement is due to quantum fluctuations of the
+ g2 / ⌬. In Fig. 6(a) the probability P↓ to find the qubit in its number of photons n within the cavity. These fluctuations
excited state (right axis) is plotted as a function of time for cause variations in the ac Stark shift 共g2 / ⌬兲n␴z, which in
the qubit initially in the ground (blue) or excited state (red). turn dephase the qubit. It is useful to compute the corre-
The dashed lines represent the corresponding number of pho- sponding dephasing rate and compare it with the measure-

062320-7
BLAIS et al. PHYSICAL REVIEW A 69, 062320 (2004)

ment rate—i.e., the rate at which information about the state Tm⌫␸ = 1. 共28兲
of the qubit can be acquired.
To determine the dephasing rate, we assume that the cav- This exceeds the quantum limit [33] Tm⌫␸ = 1 / 2 by a factor
ity is driven at the bare cavity resonance frequency and that of 2. Equivalently, in the language of Ref. [34] (which uses a
the pull of the resonance is small compared to the linewidth definition of the measurement time twice as large as that
␬. The relative phase accumulated between the ground and above) the efficiency ratio is ␹ ⬅ 1 / 共Tm⌫␸兲 = 0.5.
excited states of the qubit is The failure to reach the quantum limit can be traced [35]
to the fact that that the coupling of the photons to the qubit is

␸共t兲 = 2
g2
⌬
冕
0
t
dt⬘n共t⬘兲, 共22兲
not adiabatic. A small fraction R ⬇ ␪20 of the photons incident
on the resonator are reflected rather than transmitted. Be-
cause the phase shift of the reflected wave [14] differs by ␲
which yields a mean phase advance 具␸典 = 2␪0N with ␪0 between the two states of the qubit, it turns out that, despite
= 2g2 / ␬⌬ and N = ␬n̄t / 2 the total number of transmitted pho- its weak intensity, the reflected wave contains precisely the
tons [14]. For weak coupling, the dephasing time will greatly same amount of information about the state of the qubit as
exceed 1 / ␬ and, in the long-time limit, the noise in ␸ in- the transmitted wave which is more intense but has a smaller
duced by the ac Stark shift will be Gaussian. Dephasing can phase shift. In the language of Ref. [34], this “wasted” infor-
then be evaluated by computing the long-time decay of the mation accounts for the excess dephasing relative to the mea-
correlator surement rate. By measuring also the phase shift of the re-

冓 冉冕 冊冔
flected photons, it could be possible to reach the quantum
t limit.
具␴+共t兲␴−共0兲典 = exp i dt⬘␸共t⬘兲 Another form of possible backaction is mixing transitions

冋 冉 冊冕冕
0

册
between the two qubit states induced by the microwaves.
2 t t
1 g2 First, as seen from Fig. 6(a) and 7(a), increasing the average
⯝ exp − 2 dt1dt2具n共t1兲n共t2兲典 . number of photons in the cavity induces mixing. This is sim-
2 ⌬ 0 0
ply caused by dressing of the qubit by the cavity photons.
共23兲 Using the dressed states (2) and (3), the level of this coherent
mixing can be estimated as
To evaluate this correlator in the presence of a continuous-
wave (cw) drive on the cavity, we first perform a canonical 1 ¯ ¯ 典
transformation on the cavity operators a共†兲 by writing them in P↓,↑ = 具±,n 兩1 ± ␴z兩±,n 共29兲
2
terms of a classical ␣共*兲 and a quantum part d共†兲:

a共t兲 = ␣共t兲 + d共t兲. 共24兲 =

1
2
冉1±
⌬
冑4g 共n + 1兲 + ⌬2 .
2 冊 共30兲
Under this transformation, the coherent state obeying a兩␣典
= ␣兩␣典 is simply the vacuum for the operator d. It is then easy Exciting the cavity to n = ncrit yields P↓ ⬃ 0.85. As is clear
to verify that from the numerical results, this process is completely revers-
ible and does not lead to errors in the readout.
具关n共t兲 − n̄兴关n共0兲 − n̄兴典 = ␣2具d共t兲d†共0兲典 = n̄e−␬兩t兩/2 . 共25兲 The drive can also lead to real transitions between the
qubit states. However, since the coupling is so strong, large
It is interesting to note that the factor of 1 / 2 in the exponent detuning ⌬ = 0.1 ␻r can be chosen, making the mixing rate
is due to the presence of the coherent drive. If the resonator limited not by the frequency spread of the drive pulse, but
is not driven, the photon number correlator rather decays at a rather by the width of the qubit excited state itself. The rate
rate ␬. Using this result in Eq. (23) yields the dephasing rate of driving the qubit from ground to excited state when n
photons are in the cavity is R ⬇ n共g / ⌬兲2␥. If the measure-
␬
⌫␸ = 4␪20 n̄. 共26兲 ment pulse excites the cavity to n = ncrit, we see that the ex-
2 citation rate is still only 1/4 of the relaxation rate. As a result,
the main limitation on the fidelity of this QND readout is the
Since the rate of transmission on resonance is ␬n̄ / 2, this
decay of the excited state of the qubit during the course of
means that the dephasing per transmitted photon is 4␪20.
the readout. This occurs (for small ␥) with probability
To compare this result to the measurement time Tmeas, we
Prelax ⬃ ␥tmeas ⬃ 15␥ / ␬ ⬃ 3.75% and leads to a small error
imagine a homodyne measurement to determine the transmit-
Perr ⬃ 5␥ / ␬ ⬃ 1.5% in the measurement, where we have
ted phase. Standard analysis of such an interferometric setup
taken ␥ = ␥␬. As confirmed by the numerical calculations of
[14] shows that the minimum phase change which can be
resolved using N photons is ␦␪ = 1 / 冑N. Hence the measure-
Fig. 6 and 7, this dispersive measurement is therefore highly
nondemolition.
ment time to resolve the phase change ␦␪ = 2␪0 is

1 C. Signal to noise
Tm = 2, 共27兲
2␬n̄␪0 For homodyne detection in the case where the cavity pull
g2 / ⌬␬ is larger than 1, the signal-to-noise ratio (SNR) is
which yields given by the ratio of the number of photons, nsig = n␬⌬t / 2,

062320-8
CAVITY QUANTUM ELECTRODYNAMICS FOR… PHYSICAL REVIEW A 69, 062320 (2004)

TABLE II. Figures of merit for readout and multiqubit entangle- distinguish the two states and thus cannot dephase the qubit.
ment of superconducting qubits using dispersive (off-resonant) cou- This also implies that a charge measurement cannot be used
pling to a 1D transmission-line resonator. The same parameters as to determine the state of the system [4,5]. While the first
Table I and a detuning of the Cooper-pair box from the resonator of derivative of the energy splitting with respect to gate voltage
10% 共⌬ = 0.1␻r兲 are assumed. Quantities involving the qubit decay vanishes at the degeneracy point, the second derivative, cor-
␥ are computed both for the theoretical lower bound ␥ = ␥␬ for responding to the difference in charge polarizability of the
spontaneous emission via the cavity and (in parentheses) for the two quantum states, is maximal. One can think of the qubit
current experimental upper bound 1 / ␥ 艌 2 ␮s. Though the signal to as a nonlinear quantum system having a state-dependent ca-
noise of the readout is very high in either case, the estimate of the
pacitance (or in general, an admittance) which changes sign
readout error rate is dominated by the probability of qubit relaxation
between the ground and excited states [36]. It is this change
during the measurement, which has a duration of a few cavity life-
in polarizability which is measured in the dispersive QND
times 关⬃共1 – 10兲␬−1兴. If the qubit nonradiative decay is low, both
high-efficiency readout and more than 103 two-bit operations could
measurement.
be attained. In contrast, standard charge measurement schemes
[37,18] require moving away from the optimal point. Sim-
Parameter Symbol 1D circuit monds et al. [20] have recently raised the possibility that
there are numerous parasitic environmental resonances
Dimensionless cavity pull g2 / ␬⌬ 2.5 which can relax the qubit when its frequency ⍀ is changed
Cavity-enhanced lifetime ␥␬ = 共⌬ / g兲2␬−1
−1
64 ␮s during the course of moving the operating point. The disper-
Readout SNR SNR ⫽ 共ncrit / namp兲␬ / 2␥ 200 (6) sive CQED measurement is therefore highly advantageous
Readout error Perr ⬃ 5 ⫻ ␥ / ␬ 1.5% 共14% 兲
since it operates best at the charge degeneracy point. In gen-
eral, such a measurement of an ac property of the qubit is
One-bit operation time T␲ ⬎ 1 / ⌬ ⬎0.16 ns
strongly desirable in the usual case where dephasing is domi-
Entanglement time t冑iSWAP = ␲⌬ / 4g2 ⬃0.05 ␮s nated by low-frequency 共1 / f兲 noise. Notice also that the pro-
Two-bit operations Nop = 1 / 关␥ t冑iSWAP兴 ⬎1200共40兲 posed quantum nondemolition measurement would be the
inverse of the atomic microwave CQED measurement in
accumulated over an integration period ⌬t, divided by the which the state of the photon field is inferred nondestruc-
detector noise namp = kBTN / ប␻r. Assuming the integration tively from the phase shift in the state of atoms sent through
time to be limited by the qubit’s decay time 1 / ␥ and exciting the cavity [3].
the cavity to a maximal amplitude ncrit = 100⬃ namp, we ob-
tain SNR ⫽ 共ncrit / namp兲共␬ / 2␥兲. If the qubit lifetime is longer VII. COHERENT CONTROL
than a few cavity decay times 共1 / ␬ = 160 ns兲, this SNR can
be very large. In the most optimistic situation where ␥ = ␥␬, While microwave irradiation of the cavity at its resonance
the signal-to-noise ratio is SNR= 200. frequency constitutes a measurement, irradiation close to the
When taking into account the fact that the qubit has a qubit’s frequency can be used to coherently control the state
finite probability to decay during the measurement, a better of the qubit. In the former case, the phase shift of the trans-
strategy than integrating the signal for a long time is to take mitted wave is strongly dependent on the state of the qubit
advantage of the large SNR to measure quickly. Simulations and hence the photons become entangled with the qubit, as
have shown that in the situation where ␥ = ␥␬, the optimum shown in Fig. 8. In the latter case, however, driving is not a
integration time is roughly 15 cavity lifetimes. This is the measurement because, for large detuning, the photons are
pulse length used for the stochastic numerical simulations largely reflected with a phase shift which is independent of
shown above. The readout fidelity, including the effects of the state of the qubit. There is therefore little entanglement
this stochastic decay, and related figures of merit of the between the field and qubit in this situation and the rotation
single-shot high efficiency QND readout are summarized in fidelity is high.
Table II. To model the effect of the drive on the qubit, we add the
This scheme has other interesting features that are worth microwave drive of Eq. (20) to the Jaynes-Cumming Hamil-
mentioning here. First, since nearly all the energy used in tonian (1) and apply the transformation (11) (again neglect-
this dispersive measurement scheme is dissipated in the re- ing damping) to obtain the effective one-qubit Hamiltonian
mote terminations of the input and output transmission lines,
it has the practical advantage of avoiding quasiparticle gen-
eration in the qubit.
H1q =
ប
2
冋⍀+2
g2 †
⌬
冉
a a+
1
2
冊 册
− ␻ ␮w ␴ z + ប
g␧共t兲 x
⌬
␴

Another key feature of the cavity QED readout is that it + ប共␻r − ␻␮w兲a†a + ប␧共t兲共a† + a兲 共31兲
lends itself naturally to operation of the box at the charge
degeneracy point 共Ng = 1 / 2兲, where it has been shown that T2 in a frame rotating at the drive frequency ␻␮w. Choosing
can be enormously enhanced [17] because the energy split- ␻␮w = ⍀ + 共2n + 1兲g2 / ⌬, H1q generates rotations of the qubit
ting has an extremum with respect to gate voltage and isola- about the x axis with Rabi frequency g␧ / ⌬. Different drive
tion of the qubit from 1 / f dephasing is optimal. The deriva- frequencies can be chosen to realize rotations around arbi-
tive of the energy splitting with respect to gate voltage is the trary axes in the x − z plane. In particular, choosing ␻␮w = ⍀
charge difference in the two qubit states. At the degeneracy + 共2n + 1兲g2 / ⌬ − 2g␧ / ⌬ and t = ␲⌬ / 2冑2g␧ generates the Had-
point this derivative vanishes and the environment cannot amard transformation H. Since H␴xH = ␴z, these two choices

062320-9
BLAIS et al. PHYSICAL REVIEW A 69, 062320 (2004)

number n̄ ⬇ ␧2 / ⌬2 ⬃ 0.1. Virtual population and depopulation

of the cavity can be realized much faster than the cavity
lifetime 1 / ␬ and, as a result, the qubit feels the effect of the
drive rapidly after the drive has been turned on. The limit on
the speed of turn on and off of the drive is set by the detun-
ing ⌬. If the drive is turned on faster than 1 / ⌬, the frequency
spread of the drive is such that part of the drive’s photons
will pick up phase information (see Fig. 8) and dephase the
qubit. As a result, for large detuning, this approach leads to a
fast and accurate way to coherently control the state of the
qubit.
To model the effect of the drive on the resonator an alter-
FIG. 8. (Color online) Phase shift of the cavity field for the two
native model is to use the cavity-modified Maxwell-Bloch
states of the qubit as a function of detuning between the driving and
resonator frequencies. Obtained from the steady-state solution of
equations [25]. As expected, numerical integration of the
the equation of motion for a共t兲 while only taking into account Maxwell-Bloch equations reproduce very well the stochastic
damping on the cavity and using the parameters of Table I. Readout numerical results when the drive is at the qubit’s frequency
of the qubit is realized at, or close to, zero detuning between the but do not reproduce these numerical results when the drive
drive and resonator frequencies where the dependence of the phase is close to the bare resonator frequency (Figs. 6 and 7)—i.e.,
shift on the qubit state is largest. Coherent manipulations of the when entanglement between the qubit and photons cannot be
qubit are realized close to the qubit frequency which is 10% de- neglected.
tuned from the cavity (not shown on this scale). At such large de-
tunings, there is little dependence of the phase shift on the qubit’s VIII. RESONATOR AS QUANTUM BUS: ENTANGLEMENT
state. OF MULTIPLE QUBITS

The transmission-line resonator has the advantage that it

of frequency are sufficient to realize any one-qubit logical
should be possible to place multiple qubits along its length
operation.
共⬃1 cm兲 and entangle them together, which is an essential
Assuming that we can take full advantage of lifetime en-
requirement for quantum computation. For the case of two
hancement inside the cavity (i.e., that ␥ = ␥␬), the number of
qubits, they can be placed closer to the ends of the resonator
␲ rotations about the x axis which can be carried out is N␲
but still well isolated from the environment and can be sepa-
= 2␧⌬ / ␲g␬ ⬃ 105␧ for the experimental parameters assumed
rately dc biased by capacitive coupling to the left and right
in Table I. For large ␧, the choice of drive frequency must
center conductors of the transmission line. Additional qubits
take into account the power dependence of the cavity fre-
would have to have separate gate bias lines installed.
quency pulling.
For the pair of qubits labeled i and j, both coupled with
Numerical simulation shown in Fig. 9 confirms this
strength g to the cavity and detuned from the resonator but in
simple picture and that single-bit rotations can be performed
resonance with each other, the transformation (11) yields the
with very high fidelity. It is interesting to note that since
effective two-qubit Hamiltonian [3,38,39]

冋 册 冋 册
detuning between the resonator and the drive is large, the
cavity is only virtually populated, with an average photon g2 z 1 g2
H2q ⬇ ប ␻r + 共␴i + ␴zj 兲 a†a + ប ⍀ + 共␴zi + ␴zj 兲
⌬ 2 ⌬
g2 + −
+ប 共␴ ␴ + ␴−i ␴+j 兲. 共32兲
⌬ i j
In addition to ac Stark and Lamb shifts, the last term couples
the qubits through virtual excitations of the resonator.
In a frame rotating at the qubit’s frequency ⍀, H2q gen-
erates the evolution

冋 冉
U2q共t兲 = exp − i
g2 †
⌬
1
冊
t a a + 共␴zi + ␴zj 兲
2
册

冢 冣
1
FIG. 9. (Color online) Numerical stochastic wave function g2 g2
simulation showing coherent control of a qubit by microwave irra- cos t i sin t
⌬ ⌬
diation of the cavity at the ac Stark- and Lamb-shifted qubit fre- ⫻ 丢 1r , 共33兲
quency. The qubit (red line) is first left to evolve freely for about g2 g2
i sin t cos t
40 ns. The drive is turned on for t = 7␲⌬ / 2g␧ ⬃ 115 ns, correspond- ⌬ ⌬
ing to 7␲ pulses, and then turned off. Since the drive is tuned far
1
away from the cavity, the cavity photon number (black line) is small
even for the moderately large drive amplitude ␧ = 0.03 ␻r used here. where 1r is the identity operator in resonator space. Up to

062320-10
CAVITY QUANTUM ELECTRODYNAMICS FOR… PHYSICAL REVIEW A 69, 062320 (2004)

phase factors, this corresponds at t = ␲⌬ / 4g2 ⬃ 50 ns to a X. SUMMARY AND CONCLUSIONS

冑iSWAP logical operation. Up to one-qubit gates, this opera-
In summary, we propose that the combination of one-
tion is equivalent to the controlled-NOT gate. Together with
dimensional superconducting transmission-line resonators,
one-qubit gates, the interaction H2q is therefore sufficient for
which confine their zero-point energy to extremely small vol-
universal quantum computation [40]. Assuming again that
umes, and superconducting charge qubits, which are electri-
we can take full advantage of the lifetime enhancement in-
cally controllable qubits with large electric dipole moments,
side the cavity, the number of 冑iSWAP operations which can
constitutes an interesting system to access the strong-
be carried out is N2q = 4⌬ / ␲␬ ⬃ 1200 for the parameters as-
coupling regime of cavity quantum electrodynamics. This
sumed above. This can be further improved if the qubit’s
combined system is an advantageous architecture for the co-
nonradiative decay is sufficiently small and higher Q cavities
herent control, entanglement, and readout of quantum bits
are employed.
for quantum computation and communication. Among the
When the qubits are detuned from each other, the off-
practical benefits of this approach are the ability to suppress
diagonal coupling provided by H2q is only weakly effective
radiative decay of the qubit while still allowing one-bit op-
and the coupling is for all practical purposes turned off. Two-
erations, a simple and minimally disruptive method for read-
qubit logical gates in this setup can therefore be controlled
out of single and multiple qubits, and the ability to generate
by individually tuning the qubits. Moreover, single-qubit and
tunable two-qubit entanglement over centimeter-scale dis-
two-qubit logical operations on different qubits and pairs of
tances. We also note that in the structures described here, the
qubits can both be realized simultaneously, a requirement to
emission or absorption of a single photon by the qubit is
reach presently known thresholds for fault-tolerant quantum
tagged by a sudden large change in the resonator transmis-
computation [41].
sion properties [29], making them potentially useful as
It is interesting to point out that the dispersive QND read-
single-photon sources and detectors.
out presented in Sec. VI may be able to determine the state
of multiple qubits in a single shot without the need for addi-
ACKNOWLEDGEMENTS
tional signal ports. For example, for the case of two qubits
with different detunings, the cavity pull will take four differ- We are grateful to David DeMille, Michel Devoret, Clif-
ent values ±g21 / ⌬1 ± g22 / ⌬2, allowing single-shot readout of ford Cheung, and Florian Marquardt for useful conversa-
the coupled system. This can in principle be extended to N tions. We also thank André-Marie Tremblay and the Cana-
qubits provided that the range of individual cavity pulls can dian Foundation for Innovation for access to computing
be made large enough to distinguish all the combinations. facilities. This work was supported in part by the National
Alternatively, one could read them out in small groups at the Security Agency (NSA) and Advanced Research and Devel-
expense of having to electrically vary the detuning of each opment Activity (ARDA) under Army Research Office
group to bring them into strong coupling with the resonator. (ARO) Contract No. DAAD19-02-1-0045, NSF DMR-
0196503, NSF DMR-0342157, the NSF ITR program under
IX. ENCODED UNIVERSALITY
Grant No. DMR-0325580, the David and Lucile Packard
AND DECOHERENCE-FREE SUBSPACE
Foundation, the W.M. Keck Foundation, and NSERC.

Universal quantum computation can also be realized in APPENDIX A: QUANTIZATION OF THE 1D

this architecture under the encoding L = 兵兩↑↓典 , 兩↓↑典其 by con- TRANSMISSION-LINE RESONATOR
trolling only the qubit’s detuning and, therefore, by turning
A transmission line of length L, whose cross-section di-
on and off the interaction term in H2q [42].
mension is much less then the wavelength of the transmitted
An alternative encoded two-qubit logical operation to the
signal, can be approximated by a 1D model. For relatively
one suggested in Ref. [42] can be realized here by tuning the
low frequencies it is well described by an infinite series of
four qubits forming the pair of encoded qubits in resonance
inductors with each node capacitively connected to ground,
for a time t = ␲⌬ / 3g2. The resulting effective evolution op-
as shown in Fig. 2. Denoting the inductance per unit length l
erator can be written as Û2q = exp关−i共␲⌬ / 3g2兲␴ˆ x1␴ˆ x2兴, where and the capacitance per unit length c, the Lagrangian of the
␴ˆ xi is a Pauli operator acting on the ith encoded qubit. To- circuit is

冕 冉 冊
gether with encoded one-qubit operations, Û2q is sufficient L/2
for universal quantum computation using the encoding L. l 2 1 2
L= dx j − q , 共A1兲
We point out that the subspace L is a decoherence-free −L/2 2 2c
subspace with respect to global dephasing [43] and use of
this encoding will provide some protection against noise. where j共x , t兲 and q共x , t兲 are the local current and charge den-
sity, respectively. We have ignored for the moment the two
The application of Û2q on the encoded subspace L, however,
semi-infinite transmission lines capacitively coupled to the
causes temporary leakage out of this protected subspace.
resonator. Defining the variable ␪共x , t兲,
This is also the case with the approach of Ref. [42]. In the
present situation, however, since the Hamiltonian generating
Û2q commutes with the generator of global dephasing, this ␪共x,t兲 ⬅ 冕 x

−L/2
dx⬘q共x⬘,t兲, 共A2兲
temporary excursion out of the protected subspace does not
induce noise on the encoded qubit. the Lagrangian can be rewritten as

062320-11
BLAIS et al. PHYSICAL REVIEW A 69, 062320 (2004)

L= 冕 冉L/2

−L/2
dx
l ˙2 1
2
␪ − 共ⵜ ␪兲2 .
2c
冊 共A3兲
now extend outside of the central segment which causes a
slight redshift, of order C0 / Lc, of the cavity resonant fre-
quency.
The corresponding Euler-Lagrange equation is a wave equa- As shown in Fig. 2, we assume the qubit to be fabricated
tion with the speed v = 冑1 / lc. Using the boundary conditions at the center of the resonator. As a result, at low tempera-
due to charge neutrality, tures, the qubit is coupled to the mode k = 2 of the resonator,
which as an antinode of the voltage in its center. The rms
␪共− L/2,t兲 = ␪共L/2,t兲 = 0, 共A4兲 voltage between the center conductor and the ground plane is
we obtain
0
then Vrms = 冑ប␻r / cL with ␻r = ␻2 and the voltage felt by the
0
qubit is V共0 , t兲 = Vrms 关a2共t兲 + a†2共t兲兴. In the main body of this

␪共x,t兲 = 冑 兺 2
L
ko,cutoff

ko=1
␾ko共t兲cos
k o␲ x
L
paper, we work only with this second harmonic and drop the
mode index on the resonator operators.

+冑 兺 2
L
ke,cutoff

ke=2
␾ke共t兲sin
k e␲ x
L
, 共A5兲
APPENDIX B: TREATMENT OF DISSIPATION

The evolution of the total density matrix, including the

for odd and even modes, respectively. For finite length L, the qubit, cavity mode, and baths, is described by the von Neu-
transmission line acts as a resonator with resonant frequen- man equation
cies ␻k = k␲v / L. The cutoff is determined by the fact that the i
resonator is not strictly one dimensional. ␳˙ tot = − 关Hsys + H␬ + H␥, ␳tot兴, 共B1兲
Using the normal-mode expansion (A5) in (A3), one ob- ប
tains, after spatial integration, the Lagrangian in the form of where Hsys stands for the first three terms of Eq. (1) plus the
a set of harmonic oscillators: drive Hamiltonian of Eq. (20). An explicit expression for H␬

L= 兺k
l ˙ 2 1 k␲
␾ −
2 k 2c L
冉 冊 2
␾2k . 共A6兲
can be found in Ref. [14]. When the coupling between the
system (qubit plus cavity mode) and the baths is weak, the
reduced density operator for the system can be shown to
Promoting the variable ␾k and its canonically conjugated obey the master equation [14]
momentum ␲k = l␾ ˙ k to conjugate operators and introducing
i 1
the boson creation and annihilation operators a†k and ak sat- ␳˙ = − 关Hsys, ␳兴 −
ប
兺 共L† Lm␳ + ␳Lm† Lm − 2Lm␳Lm† 兲
2 m=兵␬,␥其 m
isfying 关ak , a†k 兴 = ␦kk⬘, we obtain the usual relations diagonal-
⬘
izing the Hamiltonian obtained from the Lagrangian (A6): 共B2兲

␾ˆ k共t兲 = 冑 ប ␻ kc L
2 k␲
关ak共t兲 + a†k 共t兲兴, 共A7兲
in the Markov approximation. Here, Lm are Lindblad opera-
tors describing the effect of the baths on the system and can
be expressed as L␬ = 冑␬a and L␥ = 冑␥␴−. The effect of finite

冑
temperature and pure dephasing, for example, can also be
ប ␻ kl taken into account easily by introducing additional Lindblad
␲ˆ k共t兲 = − i 关ak共t兲 − a†k 共t兲兴. 共A8兲
2 operators.
The master equation is solved numerically by truncating
From these relations, the voltage on the resonator can be the cavity Hilbert space to N photons. This leads to 共2N兲2
expressed as coupled differential equations which, for large N, can be dif-
1 ⳵ ␪共x,t兲 ficult to solve in practice. An alternative approach is to write
V共x,t兲 = an equivalent stochastic differential equation for the wave
c ⳵x function [32,44]. There exist different such “unravelings” of

=− 兺冑
⬁

k =1
o
ប ␻ ko
Lc
sin 冉 冊k o␲ x
L
关ako共t兲 + ak† 共t兲兴
o
the master equation and here we use the quantum state dif-
fusion equation [32,44]

兺冑 冉 冊
i
⬁
ប ␻ ke k e␲ x 兩d␺典 = − Hsys兩␺典dt +
ប
兺m 共Lm − 具Lm典␺兲兩␺典d␰m
+ cos 关ake共t兲 + ak† 共t兲兴.
ke=1 Lc L e
1
共A9兲 −
2
兺m 共Lm† Lm + 具Lm† 典␺具Lm典␺ − 2具Lm† 典␺Lm兲兩␺典dt.
In the presence of the two semi-infinite transmission lines
共B3兲
coupled to the resonator, the Lagrangian (A3) and the bound-
ary conditions (A4) are modified to take into account the The d␰m are complex independent Wiener processes satisfy-
voltage drop on the coupling capacitors C0. Assuming no ing for their ensemble averages
spatial extent for the capacitors C0, the problem is still solv-
able analytically. Due to this coupling, the wave function can d␰m = d␰md␰n = 0, 共B4兲

062320-12
CAVITY QUANTUM ELECTRODYNAMICS FOR… PHYSICAL REVIEW A 69, 062320 (2004)

d␰m
*
d␰n = ␦mndt. 共B5兲 In the situation of interest in this paper, the operator being
probed is ␴z and, from Eq. (12), the qubit-measurement ap-
An advantage of this approach is that now only 2N paratus interaction Hamiltonian is given for large detuning
coupled differential equations have to be solved. A drawback by Hint = 共g2 / ⌬兲␴za†a, such that 关␴z , Hint兴 = 0. For ␴z to be a
is that the results must be averaged over many realizations of constant of motion also requires that it commute with the
the noise to obtain accurate results. Still, this leads to much
qubit Hamiltonian. This condition is also satisfied in Eq.
less important memory usage and to speedup in the numeri-
(12).
cal calculations [32,45].
That the measured observable is a constant of motion im-
APPENDIX C: QUANTUM NONDEMOLITION plies that repeated observations will yield the same result.
MEASUREMENTS This allows for the measurement result to reach arbitrary
large accuracy by accumulating signal. In practice, however,
Readout of a qubit can lead to both mixing and dephasing there are always environmental dissipation mechanisms act-
[23,33]. While dephasing is unavoidable, mixing of the mea- ing on the qubit independently of the readout. Even in a
sured observable can be eliminated in a QND measurement QND situation, these will lead to a finite mixing rate 1 / T1 of
by choosing the qubit-measurement apparatus interaction the qubit in the course of the measurement. Hence, high fi-
such that the measured observable is a constant of motion. In
delity can only be achieved by a strong measurement com-
that situation, the measurement-induced mixing is rather in-
pleted in a time Tm Ⰶ T1. This simple point is not as widely
troduced in the operator conjugate to the operator being mea-
appreciated as it should be.
sured.

[1] H. Mabuchi and A. Doherty, Science 298, 1372 (2002). Martinis (unpublished).
[2] C. J. Hood, T. W. Lynn, A. C. Doherty, A. S. Parkins, and H. [21] P. K. Day, H. G. LeDuc, B. A. Mazin, A. Vayonakis, and J.
J. Kimble, Science 287, 1447 (2000). Zmuidzinas, Nature (London) 425, 817 (2003).
[3] J. Raimond, M. Brune, and S. Haroche, Rev. Mod. Phys. 73, [22] A. Wallraff and R. Schoelkopf (unpublished).
565 (2001). [23] R. Schoelkopf, A. Clerk, S. Girvin, K. Lehnert, and M. De-
[4] A. Armour, M. Blencowe, and K. C. Schwab, Phys. Rev. Lett. voret, Quantum Noise in Mesoscopic Physics (Kluwer Aca-
88, 148301 (2002). demic, Dordrecht, 2003), Chap. 9, pp. 175–203.
[5] E. K. Irish and K. Schwab, Phys. Rev. B 68, 155311 (2003). [24] H. Kimble, Cavity Quantum Electrodynamics (Academic
[6] Y. Makhlin, G. Schön, and A. Shnirman, Rev. Mod. Phys. 73, Press, San Diego, 1994).
357 (2001). [25] C. Wang and R. Vyas, Phys. Rev. A 55, 823 (1997).
[7] O. Buisson and F. Hekking, in Macroscopic Quantum Coher- [26] A lumped LC circuit was used in Refs. [27,28] to probe flux
ence and Quantum Computing, edited by D. V. Averin, B. qubits in a different way.
Ruggiero, and P. Silvestrini (Kluwer, New York, 2001). [27] E. Il’ichev, N. Oukhanski, A. Izmalkov, T. Wagner, M. Graj-
[8] F. Marquardt and C. Bruder, Phys. Rev. B 63, 054514 (2001). car, H.-G. Meyer, A. Y. Smirnov, A. Maassen van den Brink,
[9] F. Plastina and G. Falci, Phys. Rev. B 67, 224514 (2003). M. Amin, and A. Zagoskin, Phys. Rev. Lett. 91, 097906
[10] A. Blais, A. Maassen van den Brink, and A. Zagoskin, Phys. (2003).
Rev. Lett. 90, 127901 (2003). [28] A. Izmalkov, M. Grajcar, E. Il’ichev, T. Wagner, H.-G. Meyer,
[11] W. Al-Saidi and D. Stroud, Phys. Rev. B 65, 014512 (2001). A. Smirnov, M. Amin, A. Maassen van den Brink, and A.
[12] C.-P. Yang, S.-I. Chu, and S. Han, Phys. Rev. A 67, 042311 Zagoskin, eprint cond-mat/0312332.
(2003). [29] S. Girvin, A. Blais, and R. Huang (unpublished).
[13] J. Q. You and F. Nori, Phys. Rev. B 68, 064509 (2003). [30] We note that for the case of Q = 106, the cavity pull is a re-
[14] D. Walls and G. Milburn, Quantum Optics (Spinger-Verlag, markable ±250 linewidths, but depending on the nonradiative
Berlin, 1994). decay rate of the qubit, this may be in the regime ␬ ⬍ ␥, mak-
[15] S. Haroche, in Fundamental Systems in Quantum Optics , ed- ing the state measurement too slow.
ited by J. Dalibard, J. Raimond, and J. Zinn-Justin (Elsevier, [31] S. van Enk, J. Cirac, and P. Zoller, Science 279, 2059 (1998).
New York, 1992), p. 767. [32] R. Schack and T. A. Brun, Comput. Phys. Commun. 102, 210
[16] V. Bouchiat, D. Vion, P. Joyez, D. Esteve, and M. Devoret, (1997).
Phys. Scr. T76, 165 (1998). [33] M. Devoret and R. Schoelkopf, Nature (London) 406, 1039
[17] D. Vion, A. Aassime, A. Cottet, P. Joyez, H. Pothier, C. Ur- (2000).
bina, D. Esteve, and M. Devoret, Science 296, 886 (2002). [34] A. Clerk, S. Girvin, and A. Stone, Phys. Rev. B 67, 165324
[18] K. Lehnert, K. Bladh, L. Spietz, D. Gunnarsson, D. Schuster, (2003).
P. Delsing, and R. Schoelkopf, Phys. Rev. Lett. 90, 027002 [35] F. Marquardt (unpublished).
(2003). [36] D. Averin and C. Bruder, Phys. Rev. Lett. 91, 057003 (2003).
[19] A. S. Sorensen, C. H. van der Wal, L. Childress, and M. D. [37] Y. Nakamura, Y. Pashkin, and J. Tsai, Nature (London) 398,
Lukin, Phys. Rev. Lett. 92, 063601 (2004). 786 (1999).
[20] R. W. Simmonds, K. M. Lang, D. A. Hite, D. P. Pappas, and J. [38] A. Sørensen and K. Mølmer, Phys. Rev. Lett. 82, 1971 (1999).

062320-13
BLAIS et al. PHYSICAL REVIEW A 69, 062320 (2004)

[39] S.-B. Zheng and G.-C. Guo, Phys. Rev. Lett. 85, 2392 (2000). [42] D. Lidar and L.-A. Wu, Phys. Rev. Lett. 88, 017905 (2002).
[40] A. Barenco, C. Bennett, R. Cleve, D. DiVincenzo, N. Mar- [43] J. Kempe, D. Bacon, D. Lidar, and K. B. Whaley, Phys. Rev. A
golus, S. P, T. Sleator, J. Smolin, and H. Weinfurter, Phys. Rev. 63, 042307 (2001).
A 52, 3457 (1995). [44] C. Gardiner and P. Zoller, Quantum Noise (Springer-Verlarg,
[41] D. Aharonov and M. Ben-Or, in Proceedings of the 37th An- Berlin, 2000).
nual Symposium on Foundations of Computer Science (IEEE [45] K. Molmer, Y. Castin, and J. Dalibard, J. Opt. Soc. Am. B 10,
Computer Society Press, Los Alamitos, CA, 1996), p. 46. 524 (1993).

062320-14