Documente Academic
Documente Profesional
Documente Cultură
I.
INTRODUCTION
a)
b)
function G().
b) Separation of linear and anharmonic parts
of the Josephson potential.
unclear whether the Markov approximation would be sufficient to describe such losses.
Multimode effects come to the fore in the accurate
computation of the effective Purcell decay of a qubit [14]
or the photon-mediated effective exchange interaction between qubits in the dispersive regime [10], where the perturbation theory is divergent. A phenomenological semiclassical approach to the accurate modeling of Purcell
loss has been suggested [14], based on the availability of
the effective impedance seen by the qubit. A full quantum model that incorporates the effective impedance of
the linear part of the circuit at its core was later presented [25]. This approach correctly recognizes that a
better behaved perturbation theory in the nonlinearity
can be developed if the hybridization of the qubit with
the linear multimode environment is taken into account
at the outset [26]. Incorporating the dressing of the
modes into the basis that is used to expand the nonlinearity gives then rise to self- and cross-Kerr interactions between hybridized modes. This basis however does
not account for the open nature of the resonator. Qubit
loss is then extracted from the poles of the linear circuit
impedance at the qubit port, Z(). This quantity can
in principle be measured or obtained from a simulation
of the classical Maxwell equations. Finding the poles of
2
Z() through Fosters theorem introduces potential numerical complications [27]. Moreover, the interplay of
the qubit nonlinearity and dissipation is not addressed
within Rayleigh-Schr
odinger perturbation theory. An exact treatment of dissipation is important for the calculation of multimode Purcell rates of qubits as well as the
dynamics of driven dissipative qubit networks [28].
The difficulty with incorporating dissipation on equal
footing with energetics in open systems is symptomatic
of more general issues in the quantization of radiation
in finite inhomogeneous media. One of the earliest thorough treatments of this problem [29] proposes to use a
complete set of states in the unbounded space including
the finite body as a scattering object. This modes of
the universe approach [30, 31] is well-defined but has
an impractical aspect: one has to deal with a continuum
of modes, and as a consequence simple properties characterizing the scatterer itself (e.g. its resonance frequencies
and widths) are not effectively utilized. Several methods have been proposed since then to address this shortcoming, which discussed quantization using quasi-modes
(resonances) of the finite-sized open resonator [3235].
Usually, these methods treat the atomic degree of freedom as a two-level system and use the rotating wave and
the Markov approximations.
In the present work, rather than using a Hamiltonian
description, we derive an effective Heisenberg-Langevin
equation to describe the dynamics of a transmon qubit
[36] capacitively coupled to an open multimode resonator
(See Fig. 1a). Our treatment illustrates a general framework that does not rely on the Markov, rotating wave or
two level approximations. We show that the electromagnetic degrees of freedom of the entire circuit can be integrated out and appear in the equation of motion through
the classical electromagnetic Greens function (GF) corresponding to the Maxwell operator and the associated
boundary conditions. A spectral representation of the
GF in terms of a complete set of non-Hermitian modes
[37, 38] accounts for dissipative effects from first principles. This requires the solution of a boundary-value
problem of the Maxwell operator only in the finite domain of the resonator. Our main result is the effective
equation of motion (29), which is a Heisenberg-Langevin
[3941] integro-differential equation for the phase operator of the transmon. Outgoing fields, which may be
desired to calculate the homodyne field at the input of
an amplifier chain, can be conveniently related through
the GF to the qubit phase operator.
As an immediate application, we use the effective
Heisenberg-Langevin equation of motion to study spontaneous emission. The spontaneous emission of a two
level system in a finite polarizable medium was calculated
[42, 43] in the Schr
odinger-picture in the spirit of WignerWeisskopf theory [39]. These calculations are based on
a radiation field quantization procedure which incorporates continuity and boundary conditions corresponding
to the finite dielectric [44, 45], but only focus on separable
geometries where the GF can be calculated semianalyti-
cally. A general approach to calculate spontaneous emission in any geometry [43] uses an expansion of the GF in
terms of a set of non-Hermitian modes for the appropriate boundary value problem [37, 38]. This approach is
able to consistently account for multimode effects where
the atom-field coupling strength is of the order of the
free spectral range of the cavity [22, 43, 46] for which
the atom is found to emit narrow pulses at the cavity
roundtrip period [43]. A drawback of Schrodinger-picture
calculations is their reliance on the two-level and rotating
wave approximations. The Heisenberg-Langevin method
introduced here is valid for arbitrary light-matter coupling, and therefore can access the dynamics accurately
where the rotating-wave approximation is not valid.
In summary, our microscopic treatment of the openness is one essential difference between our study and
previous works on the collective excitations of circuitQED systems with a localized Josephson nonlinearity
[25, 26, 47, 48]. In our work, the lifetime of the collective excitations arises from a proper treatment of the resonator boundary conditions [49]. The harmonic theory
of the coupled transmon-resonator system is exactly solvable via Laplace transform. Transmon qubits typically
operate in a weakly nonlinear regime, where charge dispersion is negligible [36]. We treat the Josephson anharmonicity on top of the non-Hermitian linear theory (See
Fig 1b) using multi-scale perturbation theory (MSPT)
[5052]. First, it resolves the anomaly of secular contributions in conventional time-domain perturbation theories via a resummation [5052]. While this perturbation
theory is equivalent to the Rayleigh-Schrodinger perturbation theory when the electromagnetic environment is
closed, it allows a systematic expansion even when the
environment is open and the dynamics is non-unitary.
Second, we account for the self-Kerr and cross-Kerr interactions [53] between the collective non-Hermitian excitations extending [25, 26]. Third, treating the transmon
qubit as a weakly nonlinear bosonic degree of freedom allows us to include the linear coupling to the environment
non-perturbatively. This is unlike the standard treatment of light-matter coupling as a perturbation. Therefore, the effective equation of motion is valid for all experimentally accessible coupling strengths [1922, 5457].
We finally present a perturbative procedure to reduce
the computational complexity of our main result (29),
when the qubit is weakly anharmonic. Electromagnetic
degrees of freedom are perturbatively traced out resulting in an effective equation of motion (63) in the qubit
Hilbert space only, which is well-suited for efficient numerical simulation.
The paper is organized as follows: In Sec. II, we introduce a toy model to familiarize the reader with the
main ideas and notation. In Sec. III, we present an ab
initio effective Heisenberg picture dynamics for the transmon qubit. The derivation for this effective model has
been discussed in detail in Apps. A and B. In Sec. IV A,
we study linear theory of spontaneous emission. In
Sec. IV B, we employ quantum multi-scale perturbation
3
theory to investigate the effective dynamics beyond linear approximation. The details of multi-scale calculations are presented in App. D. In Sec. IV C we compare
these results with the pure numerical simulation. We
summarize the main results of this paper in Sec. V.
II.
TOY MODEL
an effective dissipation for the cavity. Then, we eliminate the degrees of freedom of the leaky cavity mode to
arrive at an effective equation of motion for the qubit, expressed in terms of the GF of the cavity. The Heisenberg
equations of motion are found as
Xj (t) = j Yj (t) + 2g Yc (t),
n
o
Yj (t) = j Xj (t) + U 0 [Xj (t)] ,
X
Xc (t) = c Yc (t) + 2g Yj (t) +
2gb Yb (t),
Xl (
al + a
l ), Yl i(
al a
l ),
(2)
where a
l represent the boson annihilation operator of sector l j, c, b. Furthermore, g and gb are qubit-cavity
and cavity-bath couplings. U (Xj ) represents the nonlinear part of the potential shown in Fig. 1b with a blue
spider symbol.
The remainder of this section is structured as follows.
In Sec. II A, we eliminate the bosonic degrees of freedom
to obtain an effective Heisenberg-Langevin equation of
motion for the qubit. We dedicate Sec. II B to the resulting dressed spectral function of the qubit.
A.
(3b)
(3c)
Yc (t) = c Xc (t),
Xb (t) = b Yb (t) + 2gb Yc (t)
Y (t) = X (t),
b
(3d)
(3e)
(3f)
Eliminating Yj,c,b (t) using Eqs. (3b), (3d) and (3f) first,
and integrating out the bath degree of freedom via
Markov approximation [39, 60] we obtain effective equations for the qubit and cavity as
n
o
Xj (t) + j2 Xj (t) + U 0 [Xj (t)] = 2gc Xc (t), (4a)
Xc (t) + 2c Xc (t) + c2 Xc (t)
n
o
= 2gj Xj (t) + U 0 [Xj (t)] fB (t),
(4b)
(5)
(3a)
Z
0
d
Gc ()ei(tt ) ,
Gc (t, t0 )
2
(6)
(7a)
(7b)
c () as
we obtain an algebraic solution for G
1
(8)
),
( C )( + C
p
with C c ic and c c2 2c . Taking the
inverse Fourier transform of Eq. (8) we find the single
mode GF of the cavity oscillator
c () =
G
Gc (t, t0 ) =
0
1
sin [c (t t0 )] ec (tt ) (t t0 ),
c
(9)
4
c () reside in the lower-half
where since the poles of G
of the complex -plane, Gc (t, t0 ) is retarded (causal) and
(t) stands for the Heaviside step function [62].
Then, the general solution to Eq. (4b) can be expressed
in terms of Gc (t, t0 ) as [63]
Z t
n
o
Xc (t) = 2gj
dt0 Gc (t, t0 ) Xj (t0 ) + U 0 [Xj (t0 )]
0
(10)
Substituting Eq. (10) into the RHS of Eq. (4a) and defining
c
(11a)
K(t) 4g 2 Gc (t, 0),
j
D(t) 2gc Gc (t, 0),
(11b)
Linear theory
f (s)
dtest f (t),
(13)
(15)
s2 + 2c s + c2
(16)
= s2 + j2
Dj (s) s2 + j2 1 + K(s)
(17)
4g 2 j c
2
,
s + 2c s + c2
which is the denominator of the algebraic Laplace solution (15). Therefore, its roots determine the complex
resonances of the coupled system. The poles of Dj (s)
are, on the other hand, the bare complex frequencies of
the dissipative cavity oscillator found before, zc iC .
Therefore, Dj (s) can always be represented formally as
Dj (s) = (s pj )(s pj )
(s pc )(s pc )
,
(s zc )(s zc )
(18)
(19)
since the memory integral on the RHS appears as a convolution between the kernel K(t) and earlier values of
Xj (t0 ) for 0 < t0 < t. Employing the convolution identity
Z t
0
0
0
X
j (s),
L
dt K(t t )Xj (t ) = K(s)
(14)
0
Xj (t) + j2 + g 2 Xj (t) = g(j + c )Xc (t),
(20a)
Xc (t) + 2c Xc (t) + c2 + g 2 Xc (t)
= g(j + c )Xj (t) fB (t).
(20b)
pj
pc
pRW
j
pRW
c
{s}
1.2
0.8
-0.1
-0.08
-0.06
-0.04
-0.02
{s}
a)
{pj }
{pc }
0
-0.01
-0.02
-0.01
-0.02
-0.03
-0.04
10
{pc }10-5
b)
-10
-5
{pj }10-5
c)
2
g 2 (j + c )2
.
s2 + 2c s + (c2 + g 2 )
(21)
rates to half of the bare decay rate of the dissipative cavity oscillator.
In summary, we have obtained the effective equation
of motion (12) for the quadrature Xj (t) of the nonlinear oscillator. This equation incorporates the effects of
memory, initial conditions of the cavity and drive. It
admits an exact solution via Laplace transform in the
absence of nonlinearity. To lowest order, the Josephson
nonlinearity is a time-domain perturbation Xj3 (t) in
Eq. (12). This amounts to a quantum Duffing oscillator
[65] coupled to a linear environment. Time-domain perturbation theory consists of an order by order solution
of Eq. (12). A naive application leads to the appearance
of resonant coupling between the solutions at successive
orders. The resulting solution contains secular contributions, i.e. terms that grow unbounded in time. We
present the resolution of this problem using multi-scale
perturbation theory (MSPT) [5052] in Sec. IV B.
III.
EFFECTIVE DYNAMICS OF A
TRANSMON QUBIT
6
Definition
Physical Meaning
Notation
C/cL
unitless capacitance
s
g j /(g + j )
series capacitance
g /(g + j )
capacitive ratio
(x, x0 )
1 + s (x x0 )
capacitance per length
Ej,c
lcLEj,c /~
unitless energy
p
j
8Ec Ej
bare transmon frequency
(Ec /Ej )1/2
nonlinearity measure
1/2
2
(t)
dt V (t)
flux
0
0
(t)
2 /
phase
j (t)
Trph {
ph (0)
j (t)}
reduced phase
X (t)
(t)/
unitless quadrature
zpf
j (t)
X
j (t)/zpf
reduced unitless quadrature
TABLE I. Definitions of some parameters and variables. Operators are denoted by a hat notation.
,
(22)
0
t
t,
lcL
lcL , 2
Ci
, i = R, L, j, g, s
cL
Ej,c
.
lcL
~
(23)
Ej,c
(24)
(25)
where
g /(g + j ) is a capacitive ratio, j
p
8Ec Ej is the unitless bare transmon frequency and x0
is the location of transmon. The phase field (x,
t) of the
resonator satisfies an inhomogeneous wave equation
2
x (x, x0 )t2 (x,
t) = s j2 sin [j (t)](x x0 ),
(26)
where (x, x0 ) = 1 + s (x x0 ) is the unitless capacitance per unit length modified due to coupling to the
transmon qubit, and s g j /(g + j ) is the unitless
(28a)
x |
x=0+ = x L |x=0
= L t2 L (0 , t) (0
+ , t) .
(28b)
Equations (25-28b) completely describe the dynamics of a transmon qubit coupled to an open resonator.
Note that according to Eq. (25) the bare dynamics of the
0 , t).
transmon is modified due to the force term t2 (x
Therefore, in order to find the effective dynamics for the
transmon, we need to solve for (x,
t) first and evaluate it at the point of connection x = x0 . This can be
done using the classical electromagnetic GF by virtue of
the homogeneous part of Eqs. (26,27) being linear in the
quantum fields (see App. B 1). Substituting it into the
LHS of Eq. (25) and further simplifying leads to the effective dynamics for the transmon phase operator
j (t) + (1 )j2 sin [j (t)] =
Z t
d2
dt0 K0 (t t0 )j2 sin [j (t0 )]
+ 2
dt 0
Z +
d
+
it
+
DR ()inc
R (1 , )e
2
Z +
d
it
+
DL ()inc
L (0 , )e
2
Z 1+
Z +
h
i
d
0 , 0) eit .
+
dx0
I(x0 , ) i (x
0 , 0) (x
0
2
(29)
The electromagnetic GF is the basic object that appears
in the various kernels constituting the above integrodifferential equation:
Z
Kn ( ) s
d n
G(x0 , x0 , )ei ,
2
(30a)
DR () 2i G(x0 , 1+ , ),
0 , 0 , ),
DL () 2i 3 G(x
(30b)
0 , x0 , ).
I(x , ) (x , x0 )G(x
(30d)
(30c)
x0 , ) =
G(x,
X
(x0 , )
n (x, )
n
n
n2 ()
(31)
n (x, ) and
n (x, ) are the right and left eigenwhere
functions of the Helmholtz eigenvalue problem with outgoing BC and hence carry a constant flux when x .
Note that in this representation, both the CF frequencies
n (x, ) parametrically depend
n () and the CF modes
on the source frequency . The expressions for n () and
n (x, ) are given in App. B 3.
10-6
s = 0.0001
s = 0.0010
6
s = 0.0050
s = 0.0100
s = 0.0500
4
0.4
0.3
n /
n /
0.2
0.1
50
100
50
100
n /
n /
a) R = L = 105
b) R = L = 5 103
1.5
0.4
n /
0.6
n /
0.2
0.5
50
n /
c) R = L = 102
100
50
100
n /
d) R = L = 101
FIG. 4. (Color online) Decay rate n versus oscillation frequency n for the first 100 non-Hermitian modes for x0 = 0
and different values of R = L and s .
IV.
(33)
such that the initial excitation exists in the transmon sector of Hilbert space with zero photons in the resonator
and waveguides. The spontaneous emission was conventionally studied through the Markov approximation of
the memory term which results only in a modification of
the qubit-like pole. This is the Purcell modified spontaneous decay where, depending on the density of the states
of the environment, the emission rate can be suppressed
or enhanced [7276]. We extract the spontaneous decay
as the real part of transmon-like pole in a full multimode
calculation that is accurate for any qubit-resonator coupling strength.
A product initial density matrix like Eq. (33) allows
us to reduce the generic dynamics significantly, since the
8
and we obtain the effective dynamics
(t) + 2 [1 + iK (0)] X
j (t)
X
j
1
j
Z t
j (t0 ).
=
dt0 K2 (t t0 )j2 X
(35)
(39)
(40)
(36)
The derivation of Eq. (36) can be found in Apps. B 5 and
B 6.
Note that, due to the sine nonlinearity, Eq. (36) is not
closed in terms of j (t). However, in the transmon regime
[36], where Ej Ec , the nonlinearity in the spectrum of
transmon is weak. This becomes apparent when we work
with the unitless quadratures
j (t) j (t) ,
X
zpf
j (t)
Xj (t)
,
zpf
(37)
+O
zpf
zpf
3!zpf
zpf
(38)
2
= Xj (t)
X 3 (t) + O 2 ,
6 j
where (Ec /Ej )1/2 appears as a measure for the
strength of the nonlinearity. In experiment, the Josephson energy Ej can be tuned through the FBL while the
charging energy Ec ispfixed. Therefore, a higher transmon frequency j = 8Ec Ej is generally associated with
a smaller and hence weaker nonlinearity.
The remainder of this section is organized as follows.
In Sec. IV A we study the linear theory. In Sec. IV B we
develop a perturbation expansion up to leading order in
.
A.
Linear theory
In this subsection, we solve the linear effective dynamics and discuss hybridization of the transmon and the
resonator resonances. We emphasize the importance of
off-resonant modes as the coupling g is increased. We
next investigate the spontaneous decay rate as a function of transmon frequency j and coupling g and find
an asymmetric dependence on j in agreement with a
previous experiment [14].
Neglecting the cubic term in Eq. (38), the partial trace
with respect to the resonator modes can be taken directly
(41)
Equations (40) and (41) contain the solution for the reduced quadrature operator of the transmon qubit in the
Laplace domain.
In order to find the time domain solution, it is necessary to study the poles of Eq. (40) and consequently the
roots of Dj (s). The characteristic function Dj (s) can be
expressed as (see App. C)
Dj (s) = s2 + j2 +
(
)
X
s{cos
[2
(x
)]s
+
sin
[2
(x
)]
}
n
0
n
0
n
j2 +
Mn
,
(s + n )2 + n2
n
(42)
where n (x) is the phase of the non-Hermitian eigenfunc n (x) = |
n (x)|ein (x) . We identify the
tion such that
term
n (x0 )|2
Mn s |
(43)
Y (s pm )(s p )
m
.
)
(s
z
)(s
z
m
m
m
(45)
10
3.3
pj
p1
3.2
3.08
pj
p1
3.1
3.07
a)
10-4 -15
-10 -5
{s}
-1
-4
-2
{s} 10-3
{s} 10-3
a) 1 mode
b) 5 modes
b)
3.2
3.2
{s}
-0.04 -0.02
{s}
2.9
2.9
-2
{s}
0
-0.06
3.1
3
pj
p1
3.2
{s}
{s}
15
{s}
3.09
pj
pn
{s}
20
3.1
3
pj
p1
2.9
-10
-5
{s} 10
-3
zm pm
(s zm )(s zm
)
zm
pm
= 1
1
(s pm )(s pm )
s pm
s pm
zm pm
.
= 1 + O
s pm
(46)
The consequence of a small shift |pm zm | as compared
to the strongly hybridized resonant mode |p1 z1 | is that
it can be neglected in the expansion for 1/Dj (s). The
relative size of these contributions is controlled by the
coupling g . As rule of thumb, the less hybridized a resonator pole is, the less it contributes to qubit dynamics.
Ultimately, the truncation in this work is established by
imposing the convergence of the numerics.
A numerical solution for the roots of Eq. (42) at
weak coupling g reveals that the mode resonant with
the transmon is significantly shifted, with comparatively
small shifts |pm zm | in the other resonator modes (See
Fig. 5). At weak coupling, the hybridization of pj and
p1 is captured by a single resonator mode. Next, we plot
in Fig. 6 the effect of truncation on the response of the
multimode system in a band around s = pj . As the cou-
c) 10 modes
3.1
3
2.9
-0.02
pj
p1
-0.01
{s}
d) 20 modes
10
15
10
5
0
100
10-1
10-2
10-3
10-4
10-5
10-3
j {pj }
j {pj }
10-4
3
2
1
0
a) g = 103
|Axj |
|Ax1 |
|Ax2 |
|Ax3 |
|Ax4 |
|Ax5 |
Aj AX
j Xj (0) + Aj Yj (0),
Y
An AX
n Xj (0) + An Yj (0),
(49a)
(49b)
(s pj,n )
s
,
Dj (s) s=pj,n
j
(s pj,n )
.
Dj (s) s=pj,n
(50a)
(50b)
Y
The dependence of AX
j,n and Aj,n on coupling g has been
studied in Fig. 8. The transmon-like amplitude (blue) is
always dominant, and further off-resonant modes have
smaller amplitudes. By increasing g , the resonator-like
amplitude grow significantly first and reach an asymptote
as predicted by Eq. (44).
X
|AX
j |, |An |
b)
|AY
j
|, |AY
n|
B.
Xj (s) =
+ H.c.,
(47)
+
s pj
s pn
n
|Ayj |
|Ay1 |
|Ay2 |
|Ay3 |
|Ay4 |
|Ay5 |
g
a)
b) g = 5 103
100
10-1
10-2
10-3
10-4
10-5
Perturbative corrections
(52a)
(52b)
11
1
0.8
|uj |
|u1 |
|u2 |
|u3 |
|u4 |
|u5 |
0.6
0.4
0.2
0
g
FIG. 9. (Color online) Hybridization coefficients uj and
un of the first five modes for the case where the transmon
is infinitesimally detuned below the fundamental mode, i.e.
j = 1 as a function of g [0, 0.5]. Other parameters are
set as R = L = 0 and j = 0.05. The black vertical dotted
line shows the value of j .
1 |a(0)|2 ,
(53b)
2
where a(0) = (X0 + iY0 )/2. One may wonder how this
leading-order correction is modified in the presence of dis
sipation. Adding a small damping term X(t)
to Eq. (51)
such that requires a new time scale t leading
to
X (0) (t) = e 2 t a(0)eit + c.c. ,
(54a)
3
1 |a(0)|2 et .
(54b)
2
Equations (54a-54b) illustrate a more general fact that
the dissipation modifies the frequency renormalization by
a decaying envelope. This approach can be extended by
introducing higher order (slower) time scales 2 t, 2 t, t
etc. The lowest order calculation above is valid for times
short enough such that t 2 , 2 , 1 1 .
Besides the extra complexity due to non-commuting
algebra of quantum mechanics, the principles of MSPT
remain the same in the case of a free quantum Duffing oscillator [77]. The Heisenberg equation of motion is iden
tical to Eq. (51) where we promote X(t) X(t).
We
obtain the O(1) solution (see App. D 2) as
"
#
i
t
i
t
(0)e
+
e
a
(0)
(0)
t
(t) = e 2
X
+ H.c. (55a)
t
2 cos 3
4 te
with an operator-valued renormalization of the frequency
3
t
= 1 H(0)e
,
(55b)
2
1
H(0)
a
(0)
a(0) + a
(0)
a (0) .
(55c)
2
t
)t . (56)
(0) (t) |ni = ne 2 t ei(1 3n
2 e
hn 1| X
Having learned from these toy problems, we return
to the problem of spontaneous emission which can be
mapped
into a quantum Duffing oscillator with =
1/2
2
, up to leading order in perturbation, cou6 (Ec /Ej )
pled to multiple leaky quantum harmonic oscillators (see
Eq. (38)). We are interested in finding an analytic expression for the shift of the hybridized poles, pj and pn ,
that appear in the reduced dynamics of the transmon.
The hybridized poles pj and pn are the roots of Dj (s)
and they are associated with the modal decomposition of
the linear theory in Sec. IV A. The modal decomposition
can be found from the linear solution Xj (t) that belongs
to the full Hilbert space as
!
X
pn t
pj t
An e
+ H.c.
Xj (t) = Aj e +
n
(57)
j e
uj a
pj t
n e
un a
pn t
+ H.c.
This is the full-Hilbert space version of Eq. (48). It represents the unperturbed solution upon which we are building our perturbation theory. We have used bar-notation
to distinguish the creation and annihilation operators in
the hybridized mode basis. Furthermore, uj and un represent the hybridization coefficients, where they determine how much the original transmon operator Xj (t), is
transmon-like and resonator-like. They can be obtained
from a diagonalization of the linear Heisenberg-Langevin
equations of motion (see App. D 3). The dependence of
uj and un on coupling g is shown in Fig. (9) for the case
where the transmon is infinitesimally detuned below the
fundamental mode of the resonator. For g = 0, uj = 1
and un = 0 as expected. As g reaches j , u1 is substantially increased and becomes comparable to uj . By
increasing g further, un for the off-resonant modes start
to grow as well.
The nonlinearity acting on the transmon mixes all the
unperturbed resonances through self- and cross-Kerr contributions [25, 26, 53]. Kerr shifts can be measured in a
multimode cQED system [78, 79]. We therefore solve for
the equations of motion of each mode. These are (see
App. D 3)
Xl (t) + 2l Xl (t)
"
#3
X
+ l2 Xl (t) l uj Xj (t) +
un Xn (t)
= 0,
(58)
a
where X
l + a
l is the quadrature of the lth mode,
l
and l and l are the decay rate and the oscillation frequency, respectively. Equation (58) is the leading order
12
SHO
M SP T
- D
E j (t) --F X
0.02
1
0.5
0
0.7
0.01 g
0.8
0.9
1.1
a)
SHO
M SP T
- D
E j (t) --F X
0.2
0.15
0.1 g
1
0.5
0
0.05
0
b)
j (t)i from
FIG. 10. (Color online) Fourier transform of hX
linear solution (red dashed) and MSPT (blue solid) for
j = 0.05, R = L = 0.001, Ej = 50Ec and initial state
|0i +|1i
|j (0)i = j2 j as a function of g . The maximum value
j (t)i at each g is set to 1. a) g [0, 0.02],
of F hX
g = 0.001. b) g [0, 0.2], g = 0.02.
j
n
,
p
j = pj + i j uj Hj (0)e
+
uj un Hn (0)e
2
n
(59a)
while pn = n in is corrected as
p
n = pn + i
3 h 4
2j t
n (0)e2n t + u2 u2 H
j un H
n j j (0)e
2
m (0)e2m t ,
u2n u2m H
m6=n
(59b)
(0) and H
(0) represent the Hamiltonians of
where H
j
n
(59c)
These are the generalizations of the single quantum Duffing results (55b) and (55c) and reduce to them as g 0
where uj = 1 and un = 0. Each hybdridized mode is corrected due to a self-Kerr term proportional to u4l , and
cross-Kerr terms proportional to u2l u2l0 . Contributions of
the form u2l ul0 ul00 [26] do not appear up to the lowest
order in MSPT.
The consequence of the hybdridized modes being
mixed by the Kerr contribution is that the effect of the
Josephson nonlinearity on transmon dynamics is comparatively stronger at weaker couplings g . In terms of
Eqs. (59a-59b), the MSPT solution reads
Aj (0)epj t + epj t Aj (0)
(0)
+ H.c.
Xj (t) =
3
2 cos 4 j u4j te2j t
(60)
X An (0)epn t + epn t An (0)
+ H.c. ,
+
3
2 cos 4 j u4n te2n t
n
where Aj,n is defined in Eq. (57). In Fig. 10a, we have
compared the Fourier transform of hXj (t)i calculated
both for the MSPT solution (60) and the linear solu|0i +|1i
X
(t)
Tr
{
(0)
X
(t)}
j
1
j
ph
ph
j
j
Z t
j (t0 ) Trph {
=
dt0 j2 K2 (t t0 )[X
ph (0)Xj3 (t0 )}],
0
(61)
13
j (t)
X
with 62 . If we are interested in the numerical results only up to O(2 ) then the cubic term can be replaced as
i3
h
(62)
+ O 2 .
Xj3 (t) = Xj (t)
1
0
-1
V.
CONCLUSION
In this paper, we introduced a new approach for studying the effective non-Markovian Heisenberg equation of
motion of a transmon qubit coupled to an open multimode resonator beyond rotating wave and two level approximations. The main motivation to go beyond a two
level representation lies in the fact that a transmon is
a weakly nonlinear oscillator. Furthermore, the information regarding the electromagnetic environment is encoded in a single function, i.e. the electromagnetic GF.
E
j (t)
X
D
10
15
20
15
20
15
20
j (t)
X
b) g = 0.01
1
0
-1
10
t
c) g = 0.1
1
0
-1
j (t)
X
(63)
20
a) g = 0
15
=0
Since we know the linear solution (57) for Xj (t) analytically, the trace can be performed directly (see App. E).
We obtain the reduced equation in the Hilbert of transmon as
h
i
10
1
0
-1
10
t
d) g = 0.2
FIG. 11. (Color online) Comparison of short-time dynamics between the results from linear theory (black dash-dot),
j (t)i for
MSPT (red dotted) and numerical (blue solid) of hX
the same parameters as in Fig. (10). The oscillation frequency
and decay rate of the most dominant pole (transmon-like) are
controlled by hybridization strength. For a) where g = 0,
there is no dissipation and the transmon is isolated. The
decay rate increases with g such that the Q-factor for the
transmon-like resonance reaches Qj j /j 625.3 in Fig.
d).
As a result, the opening of the resonator is taken into account analytically, in contrast to the Lindblad formalism
where the decay rates enter only phenomenologically.
We applied this theory to the problem of spontaneous
emission as the simplest possible example. The weak
nonlinearity of the transmon allowed us to solve for the
dynamics perturbatively in terms of (Ec /Ej )1/2 which appears as a measure of nonlinearity. Neglecting the nonlinearity, the transmon acts as a simple harmonic oscillator and the resulting linear theory is exactly solvable
via Laplace transform. By employing Laplace transform, we avoided Markov approximation and therefore
accounted for the exact hybridization of transmon and
resonator resonances. Up to leading nonzero order, the
transmon acts as a quantum Duffing oscillator. Due to
the hybridization, the nonlinearity of the transmon introduces both self-Kerr and cross-Kerr corrections to all
hybdridized modes of the linear theory. Using MSPT,
we were able to obtain closed form solutions in Heisenberg picture that do not suffer from secular behavior.
14
A direct numerical solution confirmed the improvement
provided by the perturbation theory over the harmonic
theory. Surprisingly, we also learned that the linear theory becomes more accurate for stronger coupling since
the nonlinearity is suppressed in the qubit-like resonance
due to being shared between many hybdridized modes.
The theory developed here illustrates how far one can
go without the concept of photons. Many phenomena in
the domain of quantum electrodynamics, such as spontaneous or stimulated emission and resonance fluorescence,
have accurate semiclassical explanations in which the
electric field is treated classically while the atoms obey
the laws of quantum mechanics. For instance, the rate
of spontaneous emission can be related to the local density of electromagnetic modes in the weak coupling limit.
While it is now well understood that the electromagnetic
fluctuations are necessary to start the spontaneous emission process [80], it is important to ask to what extent a
quantized electromagnetic field effects the qubit dynamics [81]. We find here that although the electromagnetic
degrees of freedom are integrated out and the dynamics can systematically be reduced to the Hilbert space
of the transmon, the quantum state of the electromagnetic environment reappears in the initial and boundary
conditions when computing observables.
Although we studied only the spontaneous emission
problem in terms of quadratures, our theory can be applied to a driven-dissipative problem as well and all the
mathematical machinery developed in this work can be
used in more generic situations. In order to maintain a
reasonable amount of material in this paper, we postpone
the results of the driven-dissipative problem, as well as
the study of correlation functions to future work.
VI.
1
Uj (j )
= t2 (x0 , t),
Cg + Cj j
(A2)
Uj (j )
.
j
(A3)
(A5)
The boundary conditions (BC) are derived from continuity of current at each end as
1
1
x |x=L = x R |x=L+
l
l h
i (A6a)
R (L+ , t) ,
= CR t2 (L , t)
ACKNOWLEDGEMENTS
1
1
x |x=0+ = x L |x=0
l
l
= CL t2 L (0 , t) (0+ , t) ,
(A6c)
continuity of flux at x = x0
+
(x = x
0 , t) = (x = x0 , t),
(A6b)
(A7)
Uj (j )
.
j
(A8)
L
n
15
3) Find the Hamiltonian operator by promoting the
classical conjugate variables to quantum operators
m, Q
n } = mn [
m, Q
n ] = i~mn .
such that {
We use a hat-notation to distinguish operators from
classical variables.
In Eqs. (A12a) and (A12b), we have defined the unitless oscillation frequency
j as
2
2
2
2 Ej
= 8Ec Ej ,
(A14)
j lcL
Cj 0
(x, t) and R,L (x, t) have the exact same form as the
classical Euler-Lagrange equations of motion.
Next, we define unitless parameters and variables as
Ej,c
Ej,c lcL
,
(A15)
~
x
x
,
L
t
t L ,
vp
L,
vp
(A9)
2 , n
0
2e
Ci
,
cL
i = R, L, j, g, s
(A10)
e
with Ec 2C
.
j
In what follows, we work with the unitless Eqs. (A12aA12c) and BCs (A13a-A13d) and drop the bars.
(A11)
In order to find the effective dynamics of the transmon qubit, one has to solve for the flux field (x,
t) and
substitute the result back into the RHS of time evolution of the qubit given by Eq. (A12a). It is possible to
perform this procedure in terms of the resonator GF. In
Sec. B 1 we define the resonator GF. In Sec. B 3 we study
the spectral representation of the GF in terms of a suitable set of non-Hermitian modes. In Sec. B 4, we discuss
the derivation of the effective dynamics of transmon in
terms of the resonator GF. Finally, in Secs. B 5 and B 6
we discuss how the generic dynamics is reduced for the
problem of spontaneous emission.
j (t) + (1 )
j2 sin [j (t)] = t2 (
x0 , t),
(A12a)
2
x (
x, x
0 )t2 (
x, t) = s
j2 sin [j (t)](
xx
0 ),
(A12b)
x2 R,L (
x, t) t2 R,L (
x, t) = 0,
(A12c)
(A13a)
x |
x=0+ = x L |x=0
= L t2 L (0 , t) (0
+ , t) ,
(A13b)
(
x=
0 , t)
= (
x=
x
+
0 , t),
x=x s t2 (
x0 , t)
x |
x=x+ x |
0
= s
j2 sin [j (t)].
(A13c)
(A13d)
1.
Z
d
G(x, t|x0 , t0 ) =
G(x, x0 , )ei(tt0 ) , (B2b)
2
Eq. (B1) transforms into a Helmholtz equation
2
x0 , ) = (x x0 ).
x + 2 (x, x0 ) G(x,
(B3)
G
+
G
= 0, (B4b)
x
s
+
x=x0
x=x0
x=x0
16
x G
x=1
= x G
x=1+
x=1
x=0
= R 2 G
x G
x=0
= x G
x=1+
(B4c)
,
x=0+
= L 2 G
x=0+
(B4d)
x=0,1
n6=0
(B10)
X
n (x)
n (x0 ) X 1
n (x)
n (x0 )
=
,
2
2
n
2
n
nZ
nN
(B7)
x0 , ) =
G(x,
Note that BCs (B4a-B4d) do not specify what hap x0 , ) at x . We model the baths by
pens to G(x,
imposing outgoing BCs at infinity as
, x0 , ), (B5)
x0 , )
= i G(x
x G(x,
2.
Note that eigenfunctions of a Hermitian differential operator form a complete orthonormal basis. This allows
x0 , )
us to deduce the spectral representation of G(x,
[63, 82, 83] as
(B11)
but with open BCs the same as Eqs. (B4a-B5). Note that
n (x, ) and eigenfrequencies
the resulting CF modes
n () parametrically depend on the source frequency .
Considering only an outgoing plane wave solution for
the left and right waveguides based on (B5), the general
n (x, ) reads
solution for
< i ()x
+ Bn< ein ()x , 0 < x < x0
An e n
ix
C
e
,
x
>1
n
ix
Dn e
,
x<0
(B12)
Applying BCs (B4a-B4d) leads to a characteristic equation
17
n ()
sin[n ()]
sin [n ()] + (R + L )n () cos[n ()]
2
n ()
()
sin[n ()]
R L n2 () 2i
cos[n ()] + 1 + n 2
n ()
+ s n () cos[n ()x0 ] L
{in () cos[n ()x0 ] + sin[n ()x0 ]}
n ()
cos[n ()(1 x0 )] R
{in () cos[n ()(1 x0 )] + sin[n ()(1 x0 )]} = 0,
(B13)
+2i ()x n
+ix
n (x, )
in ()(1+x0 )
n
0
e
+ (1 2iL n ()) e
,
x>1
2iR n ()e
(x, )
n (x, ) = mn ,
(B15)
dx(x, x0 )
m
0
n (x, ) =
n (x, ). The normalization of Eq. (B14) is
X
n (x0 , )
n (x, )
x0 , ) =
G(x,
.
(B16)
2
n2 ()
n
Examining Eq. (B16), we realize that there are two sets
x0 , ) in the complex plane. First, from
of poles of G(x,
setting the denominator of Eq. (B16) to zero which gives
= n (). These are the quasi-bound eigenfrequencies
that satisfy the transcendental characteristic equation
2in
e
(1 2iL n )(1 2iR n )
i
(B17)
+ s n [e2in x0 + (1 2iL n )]
2
[e2in (1x0 ) + (1 2iR n )] = 0.
The quasi bound solutions n to Eq. (B17) reside in the
lower half of complex -plane and come in symmetric
pairs with respect to the ={} axis, i.e. both n and
n satisfy the transcendental Eq. (B17). Therefore, we
can label the eigenfrequencies as
n=0
i0 ,
n = +n in ,
(B18)
n +N
n in ,
n N
(B19)
(B21)
t t0 .
(B22)
18
Multiplying Eq. (B20a) by G(x, t|x0 , t0 ) and Eq. (B20b)
by (x
0 , t0 ) and integrating over the dummy variable x0
in the interval [0 , 1+ ] and over t0 in the interval [0, t+ ]
and finally taking the difference gives
Z t+
Z 1+
0
0
dt
dx
Gx20
x20 G
{z
}
0
0
|
1+
(b)
t0 )(x x0 ) = 0,
GS(j )(x0 x0 ) + (t
{z
}
{z
} |
|
G(x, t|x0 , t+ ) = 0,
Z
s
t+
(B24)
Z
(x,
t) =
|0
Z
t+
0
t0 )t20 G(x, t|x, t0 )
dt0 (x,
0
(B27)
G(x, t|x0 , t
)t20 (x
0 , t0 )
x0 =1+
dt0 (Gx0
x0 G)|x0 =0
t+
(B23)
where we have used the shorthand notation G
G(x, t|x0 , t0 ) and (x
0 , t0 ).
The term labeled as (a) can be simplified further
through integration by parts in x0 as
(B26)
hence the upper limit t0 = t+ vanishes. The second contribution comes from the Dirac -functions in (x, x0 )
and (x, x0 ) which gives
(d)
(c)
(B25)
(a)
+ (x, x0 )
t20 G (x0 , x0 )Gt20
|
{z
}
dx0 (
t0 G Gt0 )|
t0 =0 ,
and (x,
t), respectively.
At the end, we find a generic solution for the flux field
(x,
t) in the domain [0 , 1+ ] as
t+
x0 =1+
dt0 [(x
0 , t0 )x0 G(x, t|x0 , t0 ) G(x, t|x0 , t0 )x0 (x
0 , t0 )]|x0 =0
{z
}
Boundary Contribution
1+
dx0 [(x
0 , t0 )t0 G(x, t|x0 , t0 ) G(x, t|x0 , t0 )t0 (x
0 , t0 )]|t0 =0
{z
}
+
0
Z
+ s
|
(B29)
t+
dt0 (x,
t0 )t20 G(x, t|x, t0 ) G(x, t|x0 , t0 )t20 (x
0 , t0 ) .
{z
}
F eedback induced by transmon
(B30)
The boundary terms consist of two separate contribu-
(B31a)
(B31b)
19
F (t, t0 ), we have
t+
Z
(B32)
x2
dx0
x1
+
+
R (1+ , 1 ) =
inc
out
R (1 , 1 ) +
R (1 , 1 ),
0 , 1+ , ),
DR () 2i 3 G(x
0 , 0 , ),
DL () 2i 3 G(x
(B41b)
0 , x0 , ),
I(x , ) (x , x0 )G(x
(B41d)
(B41c)
Z +
d
it
+
DL ()inc
L (0 , )e
2
Z 1+
Z +
h
i
d
0
0 , 0) eit .
+
I(x0 , ) i (x
0 , 0) (x
dx
0
2
(B42)
This is Eq. (29) in Sec. III.
(B37)
R
0
By taking the integral in t0 as dt0 ei(2 1 )t =
2(1 2 ), Eq. (B33) can be simplified as
i
d h
it
0 , x0 = 1+ , )
2i G(x
inc
,
R (0 , ) e
2
(B38)
(B41a)
0
+
0
+
x0
out
out
R (x = 1 , 1 ) = +i1
R (x = 1 , 1 ), (B35a)
0
+
0
+
inc
x0
inc
R (x = 1 , 1 ) = i1
R (x = 1 , 1 ). (B35b)
d n
G(x0 , x0 , )ei ,
2
(B34)
defined as
+
0 , 1+ , 2 )
i(1 + 2 )G(x
inc
R (1 , 1 )
+
0 , 1+ , 2 )
+ i(2 1 )G(x
out
R (1 , 1 )
(B40)
Next, we write
R (x0 , ) as the sum of incoming and
outgoing parts
(B33)
d n
0 , x0 , )
(x0 , x0 )G(x
2
h
io
0 , 0) i (x
(x
0 , 0) eit .
x =1
Kn ( ) s
d2 h
d1
0 , x0 , 2 )
R (x0 , 1 )x0 G(x
dt0
2
2
i
0
0
0 , x0 , 2 )x0
G(x
R (x0 , 1 )
ei1 t ei2 (tt ) .
0
+
i
d h
it
0 , x0 = 0 , )
2i G(x
inc
.
L (0 , ) e
2
(B39)
5.
(B43)
20
Let us first calculate K2 ( ). By choosing an integration
contour
in the complex -plane shown in Fig. 12a and
1/2
X
applying
Cauchys residue theorem [83, 90] we find
~
(H)
j
(x,
0) = 1
a
n (0) + a
n (0)
n (x), I
(H)
2n cL
n
0 , x0 , )ei
d 2 G(x
(B44a)
C
!
Z
Z
(H) 1/2
X
2
i
~
0 , x0 , )ei
n
(H)
=
d
G(x
,
x
,
)e
+
d 2 G(x
0
0
a
n (0) a
n (0) n (x),
(x,
0) = 1j
i
I
II
2cL
n
i
X
1h
(B44b)
2 +in
n [
= 2i
n n (x0 )] e
2
n=0
where we have used superscript notation (H) to distin
X
guish Hermitian from non-Hermitian modes. By taking
n (x0 )|2 sin [n + n 2n (x0 )]en ,
= 2
|n ||
the partial trace over the photonic sector we find
n=0
(B49)
Trph ph a
n (0) a
n (0)
(B45)
where due to nonzero opening of the resonator, both n
= h0|ph a
n (0) a
n (0) |0iph = 0.
n (x) are in general complex valued. Therefore, we
and
inc
have
defined
arctan
,
(B50)
soning as Eq. (B45). Therefore, the effective dynamics
n
n
for the spontaneous emission problem reduces to
!
n (x)]
=[
n (x) arctan
.
(B51)
n (x)]
j (t) + (1 )j2 Trph {
ph (0) sin [j (t)]}
<[
Z t
d2
dt0 K0 (t t0 )j2 Trph {
ph (0) sin [j (t0 )]} .
= 2
As the radius
of the half-circle in Fig. 12a is taken to
R
dt 0
infinity, II d 2 G(x0 , x0 , ) approaches zero. This can
(B46)
be checked by a change of variables
[24]
= RII ei ,
(B47)
X
n=0
Z
where we have used Eq. (B41a) to rewrite timederivatives of K0 ( ) in terms of Kn ( ).
6.
X 1
n (x)
n (x0 )
,
2
n
(B48)
nZ
n (x)
n (x, = n ) is the quasi-bound eigenwhere
function.
II
Z
lim
RII
d
II
n (x0 )]2
( + in )[
ei
( n )( + n )
RII
(B53)
R
On the other hand, I in this limit reads
Z
0 , x0 , )ei
lim
d 2 G(x
RII I
Z
0 , x0 , )ei ,
=
d 2 G(x
(B54)
= 2
n=0
(B55)
21
where the first sum comes from the residues at = n
and = n , while the last sum is the residue at = 0
and they completely cancel each other and we get
K0 (0) = 0.
(B60)
b)
X
K2 ( ) =
An sin [n + n 2n (x0 )]en , (B56)
n=0
with An s
(B61)
2
(x
)
n2 + 2n
n 0 .
0 , x0 , )
= d G(x0 , x0 , ) +
d G(x
I
II
"
#
(B57)
X
n (x0 )]2
n (x0 )]2
[
[
= 2i
+
2
2
n=0
= 2i
(t) + 2 [1 + iK (0)] X
j (t)
X
j
1
j
Z t
j (t0 ).
=
dt0 K2 (t t0 ) 2 X
(C1)
f (s)
dtest f (t).
(C2)
0
n=0
R
It can be shown again that II 0 as RII from
which we find that
X
n (x0 )|2 cos [2n (x0 )]
iK1 (0) = s
|
n=0
(B58)
A
p n
=
cos [2n (x0 )].
n2 + 2n
n=0
K0 (0) has an extra pole at = 0, so the previous
contour is not well defined. Therefore, we shift the integration contour as shown in Fig. 12. Then, we have
I
0 , x0 , )
d G(x
C
Z
Z
0 , x0 , )
= d G(x0 , x0 , ) +
d G(x
I
II
"
#
(B59)
X
n (x0 )]2
(x0 )]2
1 [
[
n
= 2i
2
n
n
n=0
"
#
X
n (x0 )]2
n (x0 )]2
1 [
[
2i
+
= 0,
2
n
n
n=0
Z
=L
dt0 f (t t0 )g(t0 )
N
n1
X
dN
N
N n d
f (t) = s f (s)
s
f (t)
,
N
n1
dt
dt
t=0
n=1
(C4)
(C5)
22
where we have defined
Dj (s) s2 + 2 (s),
h
i
2 (s) .
2 (s) j2 1 + iK1 (0) + K
(C6a)
(C6b)
X
p
nN
X
nN
An
An
cos [2n (x0 )]
n2 + 2n
1.
X
A
( 2 + 2n ) cos [2n (x0 )]
p n
cos [2n (x0 )] n
(s + n )2 + n2
n2 + 2n
n=0
{n cos [2n (x0 )] n sin [2n (x0 )]} s
(s + n )2 + n2
X
s{cos [2n (x0 )]s + sin [2n (x0 )]n }
=
Mn
,
(s + n )2 + n2
n=0
(C9)
An
n (x0 )|2 .
= s |
2
2
n + n
X(t)
+ 2 X(t) X 3 (t) = 0,
(C10)
(D1)
(D2)
(D3a)
(D3b)
(D4a)
X
s{cos
[2
(x
)]s
+
sin
[2
(x
)]
}
n
0
n
0
n
.
j2 +
Mn
(s + n )2 + n2
n=0
|
{z
}
(D5)
(C11)
(2 + )a(, ) = 0,
(D6)
23
0.5
0.5
X(t)
X(t)
2.
1
-1
0
-0.5
-0.5
0
10
-1
20
10
X(0),
20
X(0)
= Y (0)
(D15)
t
a) = 0.01, = 0.1
b) = 0.01, = 0.2
X(t)
=x
(0) (t, ) +
x(1) (t, ) + O(2 ),
2
dt = t + + O( ).
which gives the -dependence of a(, ) as
a(, ) = ( )e 2 .
(D7)
|a(, )|2 = 0,
(D10)
O(1) : t2 x
(0) + 2 x
(0) = 0,
O() : t2 x
(1) + 2 x
(1)
(D17a)
h
i3
= 2 x
(0) 2t x
(0) .
(D17b)
(D9)
(D12)
where we have defined the decay rate . and a
normalized frequency
(t) as
3
(t) 1 |(0)|2 et .
(D13)
2
Furthermore, (0) is determined based on initial conditions as (0) = (X0 + iY0 )/2.
A comparison between the numerical solution (blue),
O(1) MSPT solution (D12) (red) and linear solution
(black) is made in Fig. 13 for the first ten oscillation
periods. The MSPT solution captures the true oscillation frequency better than the linear solution. However,
it is only valid for t 2 , 2 , 1 1 up to this order
in perturbation theory.
(D18)
Furthermore,
from
the
commutation
relation
we find that [
[
x(t, ), y(t, )] = 2i1
a( ), a
( )] = 1.
Substituting Eq. (D17a) into the RHS of Eq. (D17b)
and setting the secular term oscillating at to zero we
obtain
d
a( )
+ 2 a
( )
a( )
a ( )
d
+
a( )
a ( )
a( ) + a
( )
a( )
a( ) = 0,
(D11)
(D16b)
(D8)
Multipliying Eq. (D8) by a (, ) and its complex conjugate by a(, ) and taking the difference gives
(D16a)
2i
(D19)
(D20)
where
) 1 a
H(
( )
a( ) + a
( )
a ( ) .
2
(D21)
(D22)
24
) = H(0).
1. Expand f H(0)
as a Taylor series in powers of
operator H(0)
,
2. Weyl-order
the n
hseries in o
term
as
W a
(0) H(0)
h
i
h
inm
n
m
P n
1
H(0)
a
(0)
H(0)
.
n
2
m
term-by
m=0
a
(0)ei
3
2 H(0)
+ ei 2 H(0) a
(0)
.
2 cos 3
4
(D24)
+i
t
+i
t
(D25)
a
(0)e
+e
a
(0)
,
3
2 cos 4 t
where
[1 3
2 H(0)] appears as a renormalized
frequency operator.
The physical quantity of interest is the expectation
(0) (t) with respect to the initial density matrix
value of X
(0). The number basis of the simple harmonic oscillator
is a complete basis for the Hilbert space of the Duffing
oscillator such that
X
(0) =
cmn |mi hn| .
(D26)
3.
uj Xj + uc Xc ,
8
j,c and
where uj,c are real hybridization coefficients and X
Yj,c represent j-like and c-like canonical operators. For
g = 0, uj 1, uc 0, Xj,c Xj,c and Yj,c Yj,c . To
find uj,c consider the Heisenberg equations of motion
Xj (t) + j2 Xj (t) = 2gc Xc (t),
Xc (t) + 2 Xc (t) = 2gj Xj (t).
c
(D30a)
(D30b)
Expressing X~ (Xj Xc )T , the system above can be written as X~ + V X~ = 0, where V is a 2 2 matrix. Plugging
an Ansatz X~ = X~0 eit leads to an eigensystem whose
eigenvalues are j,c and whose eigenvectors give the hybridization coefficients uj,c .
The Heisenberg equations of motion for the hybdridized modes Xl (t), l j, c, reads
h
i3
mn
hn 1| a
(0) |ni ei 2 hn|H(0)|ni
hn 1| a
(t) |ni =
2 cos 3
4 t
+
i 3
2 hn1|H(0)|n1i
2 cos
= hn 1| a
(0) |ni e
(0)
(1)
Xl (t) = x
l (t, j , c ) + l x
l (t, j , c )
(1)
hn 1| a
(0) |ni
3
4 t
3n
i 2 t
,
(D27)
+ l0 yl (t, j , c ) + O(2j , 2c , j c ),
dt = t + j j + c c + O(2j , 2c , j c ).
(D32a)
(D32b)
l
O(1) : t2 x
(0)
l
+ l2 x
= 0,
(D33)
25
whose general solution reads
(0)
x
l (t, j , c ) = a
l (j , c )eil t
+a
l (j , c )e+il t .
(D34)
l (l ) =
a
where
3l
2
l (0)l
h
2 cos
[a
[a
l1 , a
l2 ] = l1 l2 1,
l1 , a
l2 ] = [a
l1 , a
l2 ] = 0.
(D35)
(0)
l = 2t l x
+ l2 x
l
h
i3
(0)
2
(0)
l u j x
= 0,
j + uc x
c
(1)
l
O(l ) of l :t2 x
(1)
l
O(l0 ) of l : t2 y
(1)
l
+ l2 y
(0)
= 2t l0 x
l .
(D36a)
+ ei
3
3l
2
3ul l l
4
l (0)l
h
a
l (0)
(D43)
(0)
At last, Xl (t) is found by replacing l = l t as
l (0)eil t + eil t a
a
(0)
(0)
(0)
l
3
Xl (t) = x
(t,
t)
=
l
l
3ul l l
t
2 cos
4
(0)
a
(0)e+il t + e+il t a
3
+ l
,
3ul l l
t
2 cos
4
(D36b)
(D44)
h
i
(0) . Plugging the expressions
where l l 1 32 l h
l
(0), we find the explicit operator renormalfor and h
da
l
3l n 3 h
la
l
i
ul H
l + a
l H
dl
4
h
io
l0 a
l0
+ 2ul u2l0 H
l + a
l H
= 0,
(D37)
where
h
i
l (l ) 1 a
l (l )a
l (l ) + a
l (l )a
l (l ) .
H
2
(D38)
Hermitian conjugate by a
l (l ) and adding them we find
that
l (l )
dH
= 0,
dl
(D39)
l (0) u3 H
h
l l (0) + 2ul ul0 Hl0 (0),
(D40)
(0)
+
u
u
H
(0)
,
j = j j u4j H
j
j c c
2
h
i
3
(0) + u2 u2 H
c = c j u4c H
c
c j j (0) .
2
(D41)
Equation (D41) has the same form as Eq. (D20) while the
Hamiltonian H(0) is replaced by an effective Hamiltonian
hl (0)l .
(D42)
a
l ( ) = W a
l (0) exp +i
2
(D45b)
Equations. (D45a-D45b) are symmetric under j c, implying that in the normal mode picture all modes are
renormalized in the same manner. The terms proportional to u4j,c and u2j,c u2c,j are the self-Kerr and cross-Kerr
contributions, respectively.
This analysis can be extended to the case of a Duffing
oscillator coupled to multiple harmonic oscillators without further complexity, since the Hilbert spaces of the distinct normal modes do not mix to lowest order in MSPT.
Consider the full Hamiltonian of our cQED system as
j X 2 + Y 2 X 4
H
j
j
j
4
2
X n
X
(D46)
2
2
Xn + Yn +
+
gn Yj Yn ,
4
n
n
da
l
3l h
l (0)a
l (0) = 0.
i
h
l + a
l h
dl
4
(D45a)
1
n (x0 ).
j j n
2
(D47)
uj Xj +
un Xn .
8
n
(D48)
26
The procedure to arrive at uj,c and j,c is a generalization
of the one presented under Eqs. (D30a-D30b).
The Heisenberg dynamics of each normal mode is then
obtained as
#3
"
l (t) + 2 X
l (t) l uj X
j (t) +
n (t)
= 0,
X
un X
l
n
(D49)
where l
j
l ul
j (0) +
n (0) , (D50a)
j = j j uj H
uj un H
2
n
as
and
n
h
2 2
n = n 3 j u4 H
n n (0) + un uj Hj (0)
2
m (0) .
u2n u2m H
X
(t)
Tr
{
(0)
X
(t)}
j
1
j
ph
ph
j
j
Z t
j (t0 ) Trph {
dt0 j2 K2 (t t0 )[X
ph (0)Xj3 (t0 )}].
=
0
(E1)
If we are only interested in the numerical results up to
linear order in then we can write
(0)
(1)
Xj (t) = Xj (t) + Xj (t) + O(2 ),
(D50b)
m6=n
(E2)
(0)
Note that in this appendix Xj (t) differs the MSPT notation in the main body and represents the linear solu(0)
tion. We know the exact solution for Xj (t) via Laplace
transform as
)
( (0)
"
#
(0)
sXj (0) + j Yj (0)
X
(0)
3
1
4
2
t
2
2
2
t
j
n
j (0)e
n (0)e
+
uj un H
, Xj (t) = L
p
j = pj + i j uj H
Dj (s)
2
n
i
P h
(0)
(0)
(D51a)
n
and resonator-like mode pn = n in as
+ L1
Dj (s)
3 h 4
2n t
2 2
2j t
+ un uj Hj (0)e
pn = pn + i j un Hn (0)e
( (0)
)
2
(0) + j Y (0) (0)
sX
j
j
1
ph
=L
1
X
Dj (s)
2 2
2m t
+
un um Hm (0)e
.
i
P h
m6=n
n(0) (0) + bn (s)Yn(0) (0)
an (s)X
n
(D51b)
j L1
+1
,
Dj (s)
(0)
a
j (0)epj t + epj t a
j (0)
(0)
+ H.c.
Xj (t) = uj
where we have employed the fact that at t = 0, the
3
2 cos 4 j u4j te2j t
Heisenberg and Schrodinger operators are the same and
n (0)epn t + epn t a
n (0)
un a
+ H.c. .
+
(0)
3
(0) (0) 1
ph ,
Xj (0) = X
(E5a)
2 cos 4 j u4n te2n t
n
j
(D52)
Note that if there is no coupling, uj = 1 and un = 0 and
we retrieve the MSPT solution of a free Duffing oscillator
given in Eq. (D25).
(0)
(0)
ph ,
Yj (0) = Yj (0) 1
(E5b)
j Yn(0) (0),
Yn(0) (0) = 1
n(0) (0).
j X
Xn(0) (0) = 1
(E5c)
(E5d)
27
The coefficients an (s) and bn (s) can be found from the
circuit elements and are proportional to light-matter coupling gn . However, for the argument that we are are
trying to make, it is sufficient to keep them in general
form.
Note that equation (E4) can be written formally as
(0)
(0) (t) 1
j,ph (t).
ph + 1
j X
Xj (t) = X
j
(E6)
h
i3
(0)
Therefore, Xj (t) is found as
Trph
(E10)
3 (t) + O(2 ),
=X
j
h
i3 h
i3
(0)
3
(0) (t) 1
j,ph
ph + 1
j X
Xj (t) = X
(t)
j
h
i2
(0) (t) X
j,ph (t) + X
(0) (t) X
2 (t) .
+3 X
j
i3 h
i3
h
(0)
(0) (t)
ph (0) Xj (t)
X
j
j,ph
(E7)
Finally, we have to take the partial trace with respect
to the photonic sector. For the initial density matrix
ph (0) = |0iph h0|ph
j,ph (t)i = hX
3 (t)i = 0.
hX
j,ph
ph
ph
(E8)
2
The only nonzero expectation values in hXj,ph
(t)iph are
2
2
i3
h
n
(0) (t).
(0) (t) + 3L1
X
X
j
j
Dj (s)
h
i3
(t) + 2 [1 + iK (0)] X
X
(t)
X
(t)
j
1
j
j
j
Z t
h
i3
0
0
0 2
0
=
dt j K2 (t t ) Xj (t ) Xj (t )
+ O(2 ).
0
(E11)
(E9)
(2011).
[11] A. F. Van Loo, A. Fedorov, K. Lalumi`ere, B. C. Sanders,
A. Blais, and A. Wallraff, Science 342, 1494 (2013).
[12] S. Shankar, M. Hatridge, Z. Leghtas, K. Sliwa, A. Narla,
U. Vool, S. Girvin, L. Frunzio, M. Mirrahimi, and M. Devoret, Nature (2013), 10.1038/nature12802.
[13] M. E. Kimchi-Schwartz, L. Martin, E. Flurin, C. Aron,
M. Kulkarni, H. E. T
ureci, and I. Siddiqi, Phys. Rev.
Lett. 116, 240503 (2016).
[14] A. A. Houck, J. A. Schreier, B. R. Johnson, J. M. Chow,
J. Koch, J. M. Gambetta, D. I. Schuster, L. Frunzio,
M. H. Devoret, S. M. Girvin, and R. J. Schoelkopf, Phys.
Rev. Lett. 101, 080502 (2008).
[15] E. Jeffrey, D. Sank, J. Y. Mutus, T. C. White, J. Kelly,
R. Barends, Y. Chen, Z. Chen, B. Chiaro, A. Dunsworth,
A. Megrant, P. J. J. OMalley, C. Neill, P. Roushan,
A. Vainsencher, J. Wenner, A. N. Cleland, and J. M.
Martinis, Phys. Rev. Lett. 112, 190504 (2014).
[16] N. T. Bronn, Y. Liu, J. B. Hertzberg, A. D. C
orcoles,
A. A. Houck, J. M. Gambetta, and J. M. Chow, Applied
Physics Letters 107, 172601 (2015).
[17] P. Roushan, C. Neill, A. Megrant, Y. Chen, R. Babbush, R. Barends, B. Campbell, Z. Chen, B. Chiaro,
28
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
[40]
[41]
[42]
29
[71]
[72]
[73]
[74]
[75]
[76]
[77]
[78]
[79]
[80]
[81]
[82]
[83]
dia, 2013).
[84] P. T. Leung, S. Y. Liu, S. S. Tong, and K. Young, Phys.
Rev. A 49, 3068 (1994).
[85] P. T. Leung, S. S. Tong, and K. Young, Journal of
Physics A: Mathematical and General 30, 2139 (1997).
[86] S. Severini, A. Settimi, C. Sibilia, M. Bertolotti,
A. Napoli, and A. Messina, Phys. Rev. E 70, 056614
(2004).
[87] E. A. Muljarov, W. Langbein, and R. Zimmermann,
EPL (Europhysics Letters) 92, 50010 (2010).
[88] M. B. Doost, W. Langbein, and E. A. Muljarov, Phys.
Rev. A 87, 043827 (2013).
[89] P. T. Kristensen, R.-C. Ge, and S. Hughes, Phys. Rev.
A 92, 053810 (2015).
[90] D. Mitrinovic and J. D. Keckic, The Cauchy method
of residues: theory and applications, Vol. 259 (Springer
Science & Business Media, 1984).
[91] G. A. Korn and T. M. Korn, Mathematical handbook
for scientists and engineers: Definitions, theorems, and
formulas for reference and review (Dover Publication,
1968).
[92] W. P. Schleich, Quantum optics in phase space (John
Wiley & Sons, 2011).
[93] C. M. Bender, L. R. Mead, and S. S. Pinsky, Phys. Rev.
Lett. 56, 2445 (1986).
[94] C. M. Bender, L. R. Mead, and S. S. Pinsky, Journal of
mathematical physics 28, 509 (1987).
[95] C. M. Bender and G. V. Dunne, Journal of mathematical
physics 29, 1727 (1988).