To catch and reverse a quantum jump mid-flight
Z.K. Minev,1 S.O. Mundhada,1 S. Shankar,1 P. Reinhold,1 R. Gutiérrez-Jáuregui,2
R.J. Schoelkopf,1 M. Mirrahimi,3, 4 H.J. Carmichael,2 and M.H. Devoret1
arXiv:1803.00545v3 [quant-ph] 12 Feb 2019
2
1
Department of Applied Physics, Yale University, New Haven, Connecticut 06511, USA
The Dodd-Walls Centre for Photonic and Quantum Technologies, Department of Physics,
University of Auckland, Private Bag 92019, Auckland, New Zealand
3
Yale Quantum Institute, Yale University, New Haven, Connecticut 06520, USA
4
QUANTIC team, INRIA de Paris, 2 Rue Simone Iff, 75012 Paris, France
(Dated: February 14, 2019)
Quantum physics was invented to account for two fundamental features of measurement results —
their discreetness and randomness. Emblematic of these features is Bohr’s idea of quantum jumps between two discrete energy levels of an atom1 . Experimentally, quantum jumps were first observed in an
atomic ion driven by a weak deterministic force while under strong continuous energy measurement2–4 .
The times at which the discontinuous jump transitions occur are reputed to be fundamentally unpredictable. Can there be, despite the indeterminism of quantum physics, a possibility to know if a
quantum jump is about to occur or not? Here, we answer this question affirmatively by experimentally demonstrating that the jump from the ground to an excited state of a superconducting artificial
three-level atom can be tracked as it follows a predictable “flight,” by monitoring the population of
an auxiliary energy level coupled to the ground state. The experimental results demonstrate that the
jump evolution when completed is continuous, coherent, and deterministic. Furthermore, exploiting
these features and using real-time monitoring and feedback, we catch and reverse a quantum jump
mid-flight, thus deterministically preventing its completion. Our results, which agree with theoretical predictions essentially without adjustable parameters, support the modern quantum trajectory
theory5–9 and provide new ground for the exploration of real-time intervention techniques in the control
of quantum systems, such as early detection of error syndromes.
Bohr conceived of quantum jumps1 in 1913, and
while Einstein elevated their hypothesis to the level of
a quantitative rule with his AB coefficient theory10,11 ,
Schrödinger strongly objected to their existence12 . The
nature and existence of quantum jumps remained a subject of controversy for seven decades until they were directly observed in a single system2–4 . Since then, quantum jumps have been observed in a variety of atomic13–16
and solid-state17–21 systems. Recently, quantum jumps
have been recognized as an essential phenomenon in
quantum feedback control22,23 , and in particular, for
detecting and correcting decoherence-induced errors in
quantum information systems24,25 .
Here, we focus on the canonical case of quantum jumps
between two levels indirectly monitored by a third — the
case that corresponds to the original observation of quantum jumps in atomic physics2–4 , see the level diagram of
Fig. 1a. A surprising prediction emerges according to
quantum trajectory theory:5,26,27 not only does the state
of the system evolve continuously during the jump between the ground |Gi and excited |Di state, but it is predicted that there is always a latency period prior to the
jump, during which it is possible to acquire a signal that
warns of the imminent occurrence of the jump (see Supplement Sec. IIA). This advance warning signal consists
of a rare, particular lull in the excitation of the ancilla
state |Bi. The acquisition of this signal requires the timeresolved, fully efficient detection of every de-excitation of
|Bi. Exploiting the specific advantages of superconduct-
ing artificial atoms and their quantum-limited readout
chain, we designed an experiment that implements with
maximum fidelity and minimum delay the detection of
the advance warning signal occurring before the quantum jump (see rest of Fig. 1).
First, we developed a superconducting artificial atom
with the necessary V-shape level structure (see Fig. 1a
and Methods). It consists, besides the ground level |Gi,
of one protected, dark level |Di — engineered to not couple to any dissipative environment or any measurement
apparatus — and one ancilla level |Bi, whose occupation is monitored at rate Γ. Quantum jumps between
|Gi and |Di are induced by a weak Rabi drive ΩDG —
although this drive might eventually be turned off during the jump, as explained later. Since a direct measurement of the dark level is not feasible, the jumps are
monitored using the Dehmelt shelving scheme2 . Thus,
the occupation of |Gi is linked to that of |Bi by the
strong Rabi drive ΩBG (ΩDG ≪ ΩBG ≪ Γ). In the
atomic physics shelving scheme2–4 , an excitation to |Bi
is recorded by detecting the emitted photons from |Bi
with a photodetector. From the detection events — referred to in the following as “clicks” — one infers the
occupation of |Gi. On the other hand, from the prolonged absence of clicks (to be defined precisely below;
see also Supplement Sec. II), one infers that a quantum
jump from |Gi to |Di has occurred. Due to the poor
collection efficiency and dead-time of photon counters in
atomic physics28 , it is exceedingly difficult to detect every
2
a
Bright B
Dark D
Γ
ΩBG
300 K
κ
logic &
actuator
ΩDG(t)
b
FPGA controller
real-time filter
demodulator
4K
...
Ground G
15 mK
χB χD
ΩBG
ΩDG
...
c
κ
Dark
Bright
ω
Cavity response
d
-1
-1
-1
D
Γ-1 κ-1 Tint ΓBG ∆tmid ΩDG T2R ΓDG
10
0
1
10
10
2
10
3
10
4
10
5
10
6
Cavity
t (ns)
Figure 1.
Principle of the experiment. a, Threelevel atom possessing a hidden transition (shaded region) between its ground |Gi and dark |Di state, driven by Rabi drive
ΩDG (t). Quantum jumps between |Gi and |Di are indirectly
monitored by a stronger Rabi drive ΩBG between |Gi and
the bright state |Bi, whose occupancy is continuously monitored at rate Γ by an auxiliary oscillator (LC circuit on right),
itself measured in reflection by continuous-wave microwave
light (depicted in light blue). When the atom is in |Bi, the
LC circuit resonance frequency shifts to a lower frequency
than when the atom is in |Gi or |Di (effect schematically represented by switch). Therefore, the probe tone performs a
|Bi/not-|Bi measurement on the atom, and is blind to any
superposition of |Gi and |Di. b, The actual atom and LC
oscillator used in the experiment is a superconducting circuit
consisting of two strongly-hybridized transmon qubits placed
inside a readout resonator cavity at 15 mK. Control signals
for the atom and cavity are supplied by a room-temperature
field-programmable gate array (FPGA) controller. This fast
electronics monitors the reflected signal from the cavity, and
after demodulation and filtering, actuates the control signals.
The amplifier chain includes circulators (curved arrows) and
amplifiers (triangles and trapezoids). c, Frequency landscape
of atom and cavity responses, overlaid with the control tones
shown as vertical arrows. The cavity pull χ of the atom is
nearly identical for |Gi and |Di, but markedly distinct for
|Bi. The BG drive is bi-chromatic in order to address the
bright transition independently of the cavity state. d, Hierarchy of timescales involved in the experiment, which are
required to span 5 orders of magnitude. Symbols explained
in text, and summarized in Extended Data Table II.
individual click required to faithfully register the origin in
time of the advance warning signal. However, superconducting systems present the advantage of high collection
efficiencies29–31 , as their microwave photons are emitted
into one-dimensional waveguides and are detected with
the same detection efficiencies as optical photons. Furthermore, rather than monitoring the direct fluorescence
of the |Bi state, we monitor its occupation by dispersively
coupling it to an ancilla readout cavity. This further improves the fidelity of the detection of the de-excitation
from |Bi (effective collection efficiency of photons emitted from |Bi).
The readout cavity, schematically depicted in Fig. 1a
by an LC circuit, is resonant at ωC = 8979.64 MHz and
cooled to 15 mK. Its dispersive coupling to the atom results in a conditional shift of its resonance frequency by
χB /2π = −5.08 ± 0.2 MHz (χD /2π = −0.33 ± 0.08 MHz)
when the atom is in |Bi (|Di), see Fig. 1c. The engineered
large asymmetry between χB and χD together with the
cavity coupling rate to the output waveguide, κ/2π =
3.62 ± 0.05 MHz, renders the cavity response markedly
resolving for |Bi vs. not-|Bi, yet non-resolving29 for |Gi
vs. |Di, thus preventing information about the dark transition from reaching the environment. When probing the
cavity response at ωC − χB , the cavity either remains
empty, when the atom is in |Gi or |Di, or fills with
n̄ = 5±0.2 photons when the atom is in |Bi. This readout
scheme yields a transduction of the |Bi-occupancy signal
with five-fold amplification, which is an important advantage to overcome the noise of the following amplification
stages. To summarize, in this readout scheme, the cavity probe inquires: Is the atom in |Bi or not? The time
needed to arrive at an answer with a confidence level of
68% (signal-to-noise ratio of 1) is Γ−1 ≈ 1/ (κn̄) = 8.8 ns
for an ideal amplifier chain (Supplement Sec. IIIC).
Importantly, the engineered near-zero coupling between the cavity and the |Di state protects the |Di state
from harmful effects, including Purcell relaxation, photon shot-noise dephasing, and the yet essentially unexplained residual measurement-induced relaxation in superconducting qubits (Supplement Sec. I).We have measured the following coherence times for the |Di state:
energy relaxation T1D = 116 ± 5 µs, Ramsey coherence
D
D
T2R
= 120 ± 5 µs, and Hahn echo T2E
= 162 ± 6 µs.
While protected, the |Di state is indirectly quantumnon-demolition (QND) read out by the combination of
the V-structure, the drive between |Gi and |Bi, and the
fast |Bi-state monitoring. In practice, we can access the
population of |Di using an 80 ns unitary rotation followed
by a projective measurement of |Bi (Methods).
Once the state of the readout cavity is imprinted with
information about the occupation of |Bi, photons leak
through the cavity output port into a superconducting waveguide, which is connected to the amplification
chain, see Fig. 1b, where they are amplified by a factor
of 1012 . The first stage of amplification is a quantumlimited Josephson parametric converter (JPC), which is
followed by a high-electron-mobility transistor (HEMT)
amplifier at 4 K. The overall efficiency of the amplification chain is η = 0.33 ± 0.03, which includes all possible
loss of information, such as due to photon loss, thermal
photons, jitter, etc. (see Methods). At room temperature, the heterodyne signal is demodulated by a homebuilt field-programmable gate array (FPGA) controller,
with a 4 ns clock period for logic operations. The measurement record consists of a time series of two quadrature outcomes, Irec and Qrec , every 260 ns, which is the
integration time Tint , from which the FPGA controller
3
dN
b
τnot-B
∆tcatch
∆tcatch
B
G
D
Time-record position (µs)
Normalized counts
Qrec Irec (photon)1/2
a
ΓB G
ΓGD
τnot-B (µs)
Figure 2. Unconditioned monitoring of quantum jumps in the 3-level system. a, Typical measurement of integrated,
with duration Tint , quadratures Irec and Qrec of signal reflected from readout cavity as a function of time. The color of the dots
(see legend) denotes the state of the atom estimated by a real-time filter implemented with the FPGAs (Methods). On top,
the vertical arrows indicate “click” events (dN ) corresponding to the inferred state changing from |Bi to not-|Bi. The symbol
τnot-B corresponds to the time spent in not-|Bi, which is the time between two clicks minus the last duration spent in |Bi. An
advance warning that a jump to |Di is occurring is triggered when no click has been observed for a duration ∆tcatch , which is
chosen between 1 and 12 µs at the start of the experiment. b, Log-log plot of the histogram of τnot-B (shaded green) for 3.2 s
of continuous data of the type of panel (a). Solid line is a bi-exponential fit defining jump rates ΓBG = (0.99 ± 0.06 µs)−1 and
ΓGD = (30.8 ± 0.4 µs)−1 .
estimates the state of the atom in real time. To reduce
the influence of noise, the controller applies a real-time,
hysteretic IQ filter (see Methods), and then, from the estimated atom state, the control drives of the atom and
readout cavity are actuated, realizing feedback control.
Having described the setup of the experiment, we proceed to report its results. The field reflected out of the
cavity is monitored in a free-running protocol, for which
the atom is subject to the continuous Rabi drives ΩBG
and ΩDG , as depicted in Fig. 1. Figure 2a shows a typical
trace of the measurement record, displaying the quantum
jumps of our three-level artificial atom. For most of the
displayed duration of the record, Irec switches rapidly
between a low and high value, corresponding to approximately 0 (|Gi or |Di) and 5 (|Bi) photons in the cavity,
respectively. Spikes in Qrec , such as the one at t = 210 µs
are recognized by the FPGA logic as a short-lived excursion of the atom to a higher excited state (Methods). The
corresponding state of the atom, estimated by the FPGA
controller, is depicted by the color of the dots. A change
from |Bi to not-|Bi is equivalent to a “click” event, in
that it corresponds to the emission of a photon from |Bi
to |Gi, whose occurrence time is shown by the vertical
arrows in the inferred record dN (t) (top). We could also
indicate upward transitions from |Gi to |Bi, corresponding to photon absorption events (not emphasized here),
which would not be detectable in the atomic case.
In the example record, the detection of clicks stops
completely at t = 45 µs, which reveals a quantum jump
from |Gi to |Di, see Methods for operational definition of
quantum jumps. The state |Di survives for 90 µs before
the atom returns to |Gi at t = 135 µs, when the rapid
switching between |Gi and |Bi resumes until a second
quantum jump to the dark state occurs at t = 350 µs.
Thus, the record presents jumps from |Gi to |Di in the
form of click interruptions. These “outer” jumps occur
on a much longer time scale than the “inner” jumps from
|Gi to |Bi.
In Fig. 2b, which is based on the continuous tracking
of the quantum jumps for 3.2 s, a histogram of the time
spent in not-|Bi, τnot-B , is shown (See Extended Data
Fig. 2 for the time spent in |Bi). The panel further shows
a fit of the histogram by a bi-exponential curve that models two interleaved Poisson processes. This yields the
average time the atom rests in |Gi before an excitation
to |Bi, Γ−1
BG = 0.99 ± 0.06 µs, and the average time the
atom stays up in |Di before returning to |Gi and being
detected, Γ−1
GD = 30.8±0.4 µs. The average time between
two consecutive |Gi to |Di jumps is Γ−1
DG = 220 ± 5 µs.
The corresponding rates depend on the atom drive amplitudes (ΩDG and ΩBG ) and the measurement rate Γ
(Supplement Sec. II). Crucially, all the rates in the system must be distributed over a minimum of 5 orders of
magnitude, as shown in Fig. 1d.
Having observed the quantum jumps in the freerunning protocol, we proceed to conditionally actuate the
system control tones in order to tomographically reconstruct the time dynamics of the quantum jump from |Gi
to |Di, see Fig. 3a. Like previously, after initiating the
atom in |Bi, the FPGA controller continuously subjects
the system to the atom drives (ΩBG and ΩDG ) and to
the readout tone (R). However, in the event that the
controller detects a single click followed by the complete
absence of clicks for a total time ∆tcatch , the controller
suspends all system drives, thus freezing the system evolution, and performs tomography, as explained in Methods. Note that in each realization, the tomography measurement yields a single +1 or -1 outcome, one bit of
information for a single density matrix component. We
also introduce a division of the duration ∆tcatch into two
phases, one lasting ∆ton during which ΩDG is left on and
one lasting ∆toff = ∆tcatch − ∆ton during which ΩDG is
turned off. As we explain below, this has the purpose
of demonstrating that the evolution of the jump is not
simply due to the Rabi drive between |Gi and |Di.
In Fig. 3b, we show the dynamics of the jump mapped
out in the full presence of the Rabi drive, ΩGD , by set-
a
R
ΩBG
ΩDG
Prep |B>
4
dN(t)
...
...
Tomo.
...
...
b
∆ton
∆toff
∆tcatch
ZGD
XGD
experiment
theory
YGD
∆tmid
c
∆toff = 0
∆ton
XGD
ZGD
YGD
∆tmid
’
∆toff = ∆tcatch- ∆ton
Catch time ∆tcatch (µs)
Figure 3. Catching the quantum jump mid-flight. a,
The atom is initially prepared in |Bi. The readout tone (R)
and atom Rabi drive ΩBG are turned on until the catch condition is fulfilled, consisting of the detection of a click followed
by the absence of click detections for a total time ∆tcatch .
The Rabi drive ΩDG starts with ΩBG , but can be shut off
prematurely, prior to the end of ∆tcatch . A tomography measurement is performed after ∆tcatch . b & c, Conditional
tomography revealing the continuous, coherent, and, surprisingly, deterministic flight (when completed) of the quantum
jump from |Gi to |Di. The error bars are smaller than the
size of the dots. The mid-flight time ∆tmid is defined by
ZGD = 0. The jump proceeds even when ΩDG is turned off
at the beginning of the flight (panel c), ∆ton = 2 µs. Data
obtained from 6.8 × 106 experimental realizations. Solid lines:
theoretical prediction (Supplement Sec. IIIA). Dashed lines in
panel c: theory curves for the ∆ton interval, reproduced from
panel b. The data suggests that an advance-warning signal of
the jump can be provided by a no-click period for catch time
∆tcatch = ∆tmid , at which half of the jumps will complete.
ting ∆toff = 0. From 3.4 × 106 experimental realizations
we reconstruct, as a function of ∆tcatch , the quantum
state, and present the evolution of the jump from |Gi to
|Di as the normalized, conditional GD tomogram (Methods). For ∆tcatch < 2 µs, the atom is predominantly detected in |Gi (ZGD = −1), whereas for ∆tcatch > 10 µs,
it is predominantly detected in |Di (ZGD = +1). Imperfections, mostly excitations to higher levels, reduce
the maximum observed value to ZGD = +0.9 (Supplement Sec. IIIB2). For intermediate no-click times, between ∆tcatch = 2 µs and ∆tcatch = 10 µs, the state
of the atom evolves continuously and coherently from
|Gi to |Di — the flight of the quantum jump. The
time of mid flight, ∆tmid ≡ 3.95 µs, is markedly shorter
than the Rabi period 2π/ΩDG = 50 µs, and is given by
2 −1 2
ΩBG
ΩBG
the function ∆tmid = 2Γ
ln ΩDG
+
1
, in which
Γ
ΩDG enters logarithmically (Supplement Sec. IIA). The
maximum
coherence of the superposition, corresponding
p
2 + Y 2 , during the flight is 0.71 ± 0.005, see
to XGD
GD
also Extended Data Fig. 3, quantitatively understood to
be limited by several small imperfections (Supplement
Sec. IIIB2).
Motivated by the quantum trajectory analysis,
we fit the experimental data with ZGD (∆tcatch ) =
a + b tanh(∆tcatch /τ + c), XGD (∆tcatch ) = a′ +
b′ sech(∆tcatch /τ ′ + c′ ), and YGD (∆tcatch ) = 0. We compare the fitted jump parameters (a, a′ , b, b′ , c, c′ , τ, τ ′ ) to
those calculated from the theory and numerical simulations using independently measured system characteristics, and find agreement at the percent level (Supplement
Sec. IIIA).
By repeating the experiment with ∆ton = 2 µs, in
Fig. 3c, we show that the jump proceeds even if the
GD drive is shut off at the beginning of the no-click period. The jump remains coherent and only differs from
the previous case in a minor renormalization of the overall amplitude and timescale. The mid-flight time of the
jump, ∆t′mid , is given by a modified formula (Supplement
Sec. IIA3). The results demonstrate that the role of the
Rabi drive ΩDG is to initiate the jump and provide a
reference for the phase of its evolution32 . Note that the
∆tcatch ≫ ∆tmid non-zero steady state value of XGD in
Fig. 3b is the result of the competition between the Rabi
drive ΩDG and the effect of the measurement of |Bi (Supplement Sec. IIA2). This is confirmed in Fig. 3c, where
ΩDG = 0, and where there is no offset in the steady state
value.
The results of Fig. 3 demonstrate that despite the unpredictability of the jumps from |Gi to |Di, they are preceded by an identical no-click record. While the jump
starts at a random time and can be prematurely interrupted by a click, the deterministic nature of the uninterrupted flight comes as a surprise given the quantum fluctuations in the heterodyne record Irec during the jump
— an island of predictability in a sea of uncertainty.
In Fig. 4b, we show that by choosing ∆tcatch = ∆tmid
for the no-click period to serve as an advance warning
signal, we reverse the quantum jump33 in the presence of
ΩDG , confirming its coherence; the same result is found
when ΩDG is off, see Extended Data Fig. 4. The reverse
pulse characteristics are defined in Fig. 4a. For ϕI = π/2,
our feedback protocol succeeds in reversing the jump to
|Gi with 83.1% ± 0.3% fidelity, while for ϕI = 3π/2, the
protocol completes the jump to |Di, with 82.0% ± 0.3%
fidelity. In a control experiment, we repeat the protocol
by applying the reverse pulse at random times, rather
than those determined by the advance warning signal.
Without the advance warning signal, the measured pop-
5
a
b
Z
Population
θI
ϕI
ΩDG on
PG
PD
control
expt.
Y
X
X’
Reverse pulse angle φI
c
PG
Reversed
Reversed
Caught
Caught
ΩDG on
ΩDG off
ΩDG off
ΩDG on
Catch and reverse duration ∆tcatch (µs)
Figure 4.
Reversing the quantum jump mid-flight.
a, Bloch sphere of the GD manifold, showing the axis X’ for
the jump reversal, defined by the azimuthal angle ϕI . The
angle of the intervention pulse is θI . b, Success probabilities
PG (purple) and PD (orange) to reverse to |Gi and complete
to |Di the quantum jump mid-flight at ∆tcatch = ∆tmid , with
θI = π/2, in the presence of the Rabi drive ΩDG . The error bars are smaller than the size of the dots. Black dots:
success probability for |Gi (closed dots) and |Di (open dots)
in a control experiment where intervention is applied at random times along the record, rather than at ∆tcatch . c, Optimal success of reverse protocol (purple) as a function of
∆tcatch . The FPGA controller is programmed with the optimal {θI (∆tcatch ) , ϕI (∆tcatch )}. Closed and open dots correspond to ∆ton = ∆tcatch and ∆ton = 2 µs, respectively. Red
points show the corresponding open-loop (no intervention) results from Fig. 3b and c.
ulations only reflect those of the ensemble average.
In a final experiment, we programmed the controller with the optimal reverse pulse parameters
{θI (∆tcatch ) , ϕI (∆tcatch )}, and as shown in Fig. 4c, we
measured the success of the reverse protocol as a function of the catch time, ∆tcatch . The closed/open dots indicate the results for ΩDG on/off, while the solid curves
are theory fits motivated by the exact analytic expressions (Supplement Sec. IIIA). The complementary red
dots and curves reproduce the open-loop results of Fig. 3
for comparison.
From the experimental results of Fig. 2a one can infer, consistent with Bohr’s initial intuition and the original ion experiments, that quantum jumps are random
and discrete. Yet, the results of Fig. 3 support a contrary view, consistent with that of Schrödinger: the evolution of the jump is coherent and continuous. Noting
the difference in time scales in the two figures, we interpret the coexistence of these seemingly opposed point
of views as a unification of the discreteness of countable
events like jumps with the continuity of the deterministic Schrödinger’s equation. Furthermore, although all
6.8×106 recorded jumps (Fig. 3) are entirely independent
of one another and stochastic in their initiation and termination, the tomographic measurements as a function of
∆tcatch explicitly show that all jump evolutions follow an
essentially identical, predetermined path in Hilbert space
— not a randomly chosen one — and, in this sense, they
are deterministic. These results are further corroborated
by the reversal experiments shown in Fig. 4, which exploit
the continuous, coherent, and deterministic nature of the
jump evolution and critically hinge on priori knowledge
of the Hilbert space path. With this knowledge ignored
in the control experiment of Fig. 4b, failure of the reversal is observed.
In conclusion, these experiments revealing the coherence of the jump, promote the view that a single quantum
system under efficient, continuous observation is characterized by a time-dependent state vector inferred from
the record of previous measurement outcomes, and whose
meaning is that of an objective, generalized degree of freedom. The knowledge of the system on short timescales is
not incompatible with an unpredictable switching behavior on long time scales. The excellent agreement between
experiment and theory including known experimental imperfections (Supplement Sec. IIIA) thus provides support
to the modern quantum trajectory theory and its reliability for predicting the performance of real-time intervention techniques in the control of single quantum systems.
Acknowledgments Z.K.M. acknowledges fruitful discussion
with S.M. Girvin, H.M. Wiseman, K. Mølmer, N. Ofek, V.V. Albert, and M.P. Silveri. V.V. Albert addressed one aspect of the
Lindblad theoretical modeling regarding the waiting-time distribution. Facilities use was supported by the Yale Institute for
Nanoscience and Quantum Engineering (YINQE), the Yale SEAS
cleanroom, and NSF MRSEC DMR 1119826. This research was
supported by ARO under Grant No. W911NF-14-1-0011. R.G.J.
and H.J.C. acknowledge the support of the Marsden Fund Council
from Government funding, administered by the Royal Society of
New Zealand under Contract No UOA1328.
Author Contributions Z.K.M. initiated and performed the
experiment, designed the sample, analyzed the data, and carried
out the initial theoretical and numerical modeling of the experiment. Z.K.M. conceived the experiment based on theoretical predictions by H.J.C. H.J.C. and R.G.J. performed the presented theoretical modeling and numerical simulations. S.O.M. contributed
to the experimental setup and design of the device, and with
S.S. to its fabrication. P.R. and R.J.S. assisted with the FPGA.
M.M. contributed theoretical support, and M.H.D. supervised the
project. Z.K.M. and M.H.D. wrote the manuscript, and H.J.C.
contributed the theoretical supplement. All authors provided suggestions for the experiment, discussed the results and contributed
to the manuscript.
Correspondence Correspondence
and
requests
for
materials should be addressed to Z.K. Minev (email:
zlatko.minev@aya.yale.edu)
and
M.H.
Devoret
(email:
michel.devoret@yale.edu)
6
METHODS
Monitoring quantum jumps
Here, we aim to briefly explain how the GD dynamics
is monitored and when do we conclude that a quantum
jump has occurred.
Monitoring the GD manifold through B de-excitations.
The state of the atom within the GD manifold is monitored indirectly, by measuring the rate of de-excitations
from the ancillary state |Bi, while the G to B excitation
tone ΩBG is applied. As explained below, the monitoring
scheme is such that when the atom is in the dark state,
|Di, the rate of de-excitations from |Bi to |Gi is zero.
Conversely, when the atom is in |Gi, the rate is nonzero. Henceforth, we will refer to a de-excitation from
|Bi to |Gi simply as a de-excitation. In summary, when
the rate of de-excitations for a measurement segment is
zero, |Di is assigned to it; otherwise, |Gi or |Bi is assigned (see IQ filter section of Methods). The rate can
be monitored by either a direct or indirect method, as
explained further below.
Quantum jumps. Sections of the (continuous) measurement record are converted into state assignments, as
discussed above, such as B, G, or D. In the experiment,
long sequences of such measurements yield the same result, i.e., GGG. . . or DDD. . . When the string of results
suddenly switches its value, we say that a quantum jump
has occurred.34
Source of the difference for the de-excitation rates. The
rate of de-excitations is zero when the atom is in |Di
because the V-shape level structure forbids any direct
DB transitions; hence, |Bi cannot be excited from |Di.
Conversely, when the atom is in |Gi, the Rabi drive ΩBG
can excite the atom to |Bi. Since this ancillary state is
effectively short-lived, it almost immediately de-excites
back to |Gi. Note that in this explanation we neglect
parasitic transitions to higher excited states, which are
considered in the Supplement.
De-excitation detection: direct or indirect. A deexcitation can be detected by a direct or, alternatively,
indirect method. For atomic experiments, direct detection is a natural choice. The photon emitted by the atom
during the de-excitation, carrying away the energy once
stored in |Bi, is collected and destructively absorbed in
the sensor of a photodetecting measurement apparatus,
which produces a “click” signal (in practice, a current or
voltage pulse). Unfortunately, unavoidable imperfections
and detector non-idealities prohibit this method for the
continuous detection of nearly every single de-excitation
(Supplement Sec. III). Alternatively, one can use an indirect monitoring method. In our experiment, instead of
detecting the emitted photon, we detect the de-excitation
by monitoring the |Bi population through an ancillary
degree of freedom, the readout cavity, coupled to the
atom.
Indirect (dispersive) detection. The readout cavity frequency depends on the state of the atom. When the atom
is in |Bi, the readout cavity frequency shifts down by
more than a cavity linewidth. The cavity frequency, and
hence the |Bi population of the atom, is probed by a continuous readout tone applied at the |Bi-cavity frequency.
When the atom is in |Bi, the probe tone is resonant and
fills the cavity with a large number of photons, n̄. Otherwise, when the atom is not-in-|Bi, the probe tone is far
off resonant and the cavity is empty of photons. Choosing
n̄ ≫ 1 makes a change in the |Bi occupancy conspicuous, and hence a de-excitation, |Bi to not-|Bi, is readily
observed, even in the presence of measurement inefficiencies and imperfections. As explained in Sec. IIIC of the
Supplement, this indirect dispersive method in effect increases the signal-to-noise ratio (SNR) and de-excitation
detection efficiency. Another notable difference between
the direct and indirect method is that in the indirect
method the atom fully but shortly occupies |Bi before
de-exciting to |Gi, while in the direct scheme the probability amplitude to be in |Bi is never appreciable before a de-excitation, see Sec. II of the Supplementary
Information. In other words, in the direct monitoring
scheme, there are explicitly two sets of quantum jumps:
the BG and DG ones. The BG ones occurs much faster
and are nested within the DG jumps. The fast dynamics
of these “inner” jumps is used to interrogate the dynamics
of “outer,” DG jumps.
Setup of the Experiment
Setup and signals. Our experiments were carried out in
a cryogen-free dilution refrigerator (Oxford Triton 200 ).
The cavity and JPC35 were shielded from stray magnetic
fields by a cryogenic µ-metal (Amumetal A4K ) shield.
Our input-output cryogenic setup is nearly identical to
that described in Ref. 25, aside from the differences evident in the schematic of our setup (see Fig. 1b and Methods) or described in the following.
The control tones depicted in Fig. 1 were each generated from individual microwave generators (ΩDG and
ΩB0 : Agilent N5183A; readout cavity tone R and ΩB1 :
Vaunix LabBrick LMS-103-13 and LMS-802-13, respectively). To achieve IQ control, the generated tones were
mixed (Marki Microwave Mixers IQ-0618LXP for the
cavity and IQ-0307LXP for ΩB0 , ΩB1 , and ΩDG ) with
intermediate-frequency (IF) signals synthesized by the
16 bit digital-to-analog converters (DACs) of the integrated FPGA controller system (Innovative Integration
VPXI-ePC ). Prior to mixing, each analog output was
filtered by a 50 Ω low pass filter (Mini-Circuits BLP300+) and attenuated by a minimum of 10 dB. The radiofrequency (RF) output was amplified at room temperature (MiniCircuits ZVA-183-S+) and filtered by MiniCircuits coaxial bandpass filters. The output signal was
further pulse modulated by the FPGA with high isolation
SPST switches (Analog Device HMC-C019 ), which provided additional 80 dB isolation when the control drives
were turned off. The signals were subsequently routed
to the input lines of the refrigerator, whose details were
described in Refs. 25 and 36.
7
At room temperature, following the cryogenic highelectron mobility amplifier (HEMT; Low Noise Factory
LNF-LNC7_10A), the signal were amplified by 28 dB
(Miteq AFS3-00101200-35-ULN ) before being mixed
down (Marki image reject double-balanced mixer IRW0618 ) to an intermediate frequency (IF) of 50 MHz,
where they were band-pass filtered (Mini-Circuits SIF50+) and further amplified by a cascaded preamplifier (Stanford Research Systems SR445A), before finally
digitization by the FPGA analog-to-digital converters
(ADC).
Atom-cavity implementation
The superconducting artificial atom consisted of two
coupled transmon qubits fabricated on a 2.9 mm-by7 mm double-side-polished c-plane sapphire wafer with
the Al/AlOx /Al bridge-free electron-beam lithography
technique37,38 . The first transmon (B) was aligned with
the electric field of the fundamental TE101 mode of an
aluminum rectangular cavity (alloy 6061; dimensions:
5.08 mm by 35.5 mm by 17.8 mm), while the second transmon (D) was oriented perpendicular to the
first and positioned 170 µm adjacent to it. The inductance of the Josephson junction of each transmon (9 nH
for both B and D), the placement and dimensions of
each transmon, and the geometry of the cavity were
designed and optimized using finite-element electromagnetic analysis and the energy-participation-ratio (EPR)
method39 . The analysis also verified that the coupling
between the two qubits is described by the Hamiltonian
Ĥint = −χDB n̂B ⊗ n̂D , where n̂B/D is the photon number
operator of the B/D qubit, and χDB is the cross-Kerr
frequency.
The measured frequency and anharmonicity of the
D qubit were ωD /2π = 4845.255 MHz and αDG /2π =
152 MHz, respectively, while those of the B qubit were
ωB /2π = 5570.349 MHz and αBG /2π = 195 MHz, respectively. The cross-Kerr coupling was χDB /2π = 61 MHz.
The relaxation time of |Bi was T1B = 28±2 µs, limited by
the Purcell effect by design, while its Ramsey coherence
B
time was T2R
= 18 ± 1 µs. The remaining parameters of
the system are provided in the main text.
Atom and cavity drives
In all experiments, the following drive parameters were
used: The DG Rabi drive, ΩDG , was applied 275 kHz
below ωD , to account for the Stark shift of the cavity. The BG drive, ΩBG , was realized as a bi-chromatic
tone in order to unselectively address the BG transition,
which was broadened and Stark shifted due to the coupling between |Bi and the readout cavity. Specifically,
we address transitions from |Gi to |Bi with a Rabi drive
ΩB0 /2π = 1.20 ± 0.01 MHz at frequency ωBG , whereas
transitions from |Bi to |Gi are addressed with a Rabi
drive ΩB1 /2π = 0.60 ± 0.01 MHz tuned 30 MHz below
ωBG . This bi-chromatic scheme provided the ability to
tune the up-click and down-click rates independently, but
otherwise essentially functioned as an incoherent broadband source.
IQ filter
To mitigate the effects of imperfections in the atom
readout scheme in extracting a |Bi/not-|Bi result, we
applied a two-point, hysteretic IQ filter, implemented
on the FPGA controller in real time. The filter is realized by comparing the present quadrature record values {Irec , Qrec }, with three thresholds (IB , IB̄ , and QB )
as summarized in Extended Data Table I.
The filter and thresholds were selected to provide a
best estimate of the time of a click, operationally understood as a change in the filter output from |Bi to not|Bi. The IB and IB̄ thresholds were chosen 1.5 standard
deviations away from the I-quadrature mean of the |Bi
and not-|Bi distributions, respectively. The QB threshold was chosen 3 standard deviations away from the Qquadrature mean. Higher excited states of the atom were
selected out by Qrec values exceeding the QB threshold.
Tomography
At the end of each experimental realization, we performed one of 15 rotation sequences on the atom that
transferred information about one component of the density matrix, ρ̂a , to the population of |Bi, which was
measured with a 600 ns square pulse on the readout
cavity. Pulses were calibrated with a combination of
Rabi, derivative removal via adiabatic gate (DRAG)40 ,
All-XY41 , and amplitude pulse train sequences42 . The
readout signal was demodulated with the appropriate
digital filter function required to realize temporal mode
matching43 . To remove the effect of potential systematic offset errors in the readout signal, we subtracted the
measurement results of operator components of ρ̂a and
their opposites. From the measurement results of this
protocol, we reconstructed the density matrix ρ̂a , and
subsequently parametrized it in the useful form
N
N
2 (1 − ZGD )
2 (XGD + iYGD ) RBG + iIBG
N
ρ̂a = N2 (XGD − iYGD )
RBD + iIBD ,
2 (1 + ZGD )
RBG − iIBG
RBD − iIBD
1−N
where XGD , YGD , and ZGD are the Bloch vector components of the GD manifold, N is the total population of the
|Gi and |Di states, while RBG , RBD , IBG and IBD are the
coherences associated with |Bi, relative to the GD manifold. The measured population in |Bi, 1 − N , remains
below 0.03 during the quantum jump, see Extended Data
Fig. 4. Tomographic reconstruction was calibrated and
verified by preparing Clifford states, accounting for the
readout fidelity of 97%.
Control flow of the experiment
A diagrammatic representation of the control flow of
the experiment is illustrated in Extended Data Figure 1a,
whose elements are briefly described in the following.
“Start”: FPGA controller resets its internal memory registers to zero25,44 , including the no-click counter “cnt,”
defined below. “Prepare B”: controller deterministically
prepares the atom in |Bi, a maximally conservative initial
state, with measurement-based feedback45 . “Initialize”:
controller turns on the atom (ΩBG and ΩDG ) and cavity (R) drives and begins demodulation. “Monitor and
catch ∆ton ”: with all drives on (ΩBG , ΩDG , and R), the
8
controller actively monitors the cavity output signal until it detects no-clicks for duration ∆ton , as described in
panel (b), whereafter the controller proceeds to “monitor
and catch ∆toff ” in the case that ∆toff > 0; otherwise,
for ∆toff = 0, the controller proceeds to “tomography”
(“feedback pulse”) for the catch (reverse) protocol. “Monitor and catch ∆toff ”: with the Rabi drive ΩDG off, while
keeping the drives ΩBG and R on, the controller continues to monitor the output signal. The controller exits the
routine only in one of two events: i) if it detects a click, in
which case it proceeds to the “declare B” step of the “monitor and catch ∆ton ” routine, or ii) if no further clicks
are detected for the entirety of the pre-defined duration
∆toff , in which case the controller advances to the “tomography” (“feedback pulse”) routine, when programmed
for the catch (reverse) protocol. “Feedback pulse”: with
all continuous-wave drives turned off, the controller performs a pulse on the DG transition of the atom, defined
by the two angles {θI (∆tcatch ) , ϕI (∆tcatch )}. “Tomography”: controller performs next-in-order tomography sequence (see Tomography section above) while the demodulator finishes processing the final data in its pipeline.
“Advance tomo.”: tomography sequence counter is incremented; after a 50 µs delay, the next realization of the
experiment is started.
The concurrent-programming control flow of
the“monitor and catch ∆ton ” block is illustrated in
Extended Data Fig. 1b; specifically, the master and
demodulator modules of the controller and synchronous
sharing of data between them is depicted. The FPGA
demodulator outputs a pair of 16 bit signed integers,
{Irec , Qrec }, every Tint = 260 ns, which is routed to the
master module, as depicted by the large left-pointing
arrow (top). The master module implements the IQ
filter (see IQ filter section above) and tracks the number
of consecutive not-|Bi measurement results with the
counter “cnt.” The counter thus keeps track of the
no-click time elapsed since the last click, which is
understood as a change in the measurement result from
|Bi to not-|Bi. When the counter reaches the critical
value Non , corresponding to ∆ton , the master and
demodulator modules synchronously exit the current
routine, see the T* branch of the “declare not-B”
decision block. Until this condition is fulfilled (F*),
the two modules proceed within the current routine as
depicted by the black flowlines. To minimize latency
and maximize computation throughput, the master and
demodulator were designed to be independent sequential
processes running concurrently on the FPGA controller,
communicating strictly through synchronous message
passing, which imposed stringent synchronization and
execution time constraints. All master inter-module
logic was constrained to run at a 260 ns cycle, the
start of which necessarily was imposed to coincide with
a “receive & stream record” operation, here, denoted
by the stopwatch. In other words, this imposed the
algorithmic constraint that all flowchart paths staring at
a stopwatch and ending in a stopwatch, itself or other,
were constrained to a 260 ns execution timing. A second
key timing constraint was imposed by the time required
to propagate signals between the different FPGA cards,
which corresponded to a minimum branching-instruction
duration of 76 ns.
The corresponding demodulation-module flowchart is
identical to that shown of panel (b); hence, it is not
shown. This routine functions in following manner: If
a |Bi outcome is detected, the controller jumps to the
“declare B” block of the monitor & catch ∆ton routine;
otherwise, when only not-|Bi outcomes are observed, and
the counter reaches the critical value Noff , corresponding
to ∆tcatch = ∆ton +∆toff , the controller exits the routine.
Data availability The data that support the findings of this
study are available from the corresponding authors on reasonable
request.
9
EXTENDED DATA FIGURES
a
b
master
demodulator
receive & stream record
start
128 ns
∆toff>0
NOT (Irec B
OR Qrec>QB)
260 ns cycle duration
decl. B
∆toff=0
initialize
T
Irec>IB
OR
Qrec>QB
F
demodulate
132 ns
F
T
declare B
0 → cnt
declare not-B
cnt → cnt+1
cnt = Non?
T*
F*
{Irec, Qrec}
T*
Tint = 260 ns
36 ns
wait
catch
F*
exit
exit
feedback pulse
c
enter
tomography
advance tomo.
128 ns
master
catch
reverse
T*/F*
96 ns
monitor &
catch ∆toff
reverse
enter
{Irec, Qrec}
128 ns
>I _
prepare B
monitor &
catch ∆ton
enter
Irec>IB
OR
Qrec>QB
F
96 ns
T
jump to monitor
& catch ∆ton
“delcare B”
T*
36 ns
F*
declare not-B
cnt → cnt+1
cnt = Noff?
wait
exit
Extended Data Figure 1. Control flow of the experiment. a, Flowchart illustrating the control flow of the catch and
reverse experiments, whose results are shown in Figs. 3 and 4. See Methods for the description of each block. b, Flowchart of the
master and demodulator modules chiefly involved in the “monitor and catch ∆ton ” routine. The modules execute concurrently
and share data synchronously, as discussed in Methods. c, Flowchart of the processing involved in the master module of the
‘monitor and catch ∆toff ” routine; see Methods.
Input: Qrec ≥ QB or Irec > IB Qrec < QB and Irec < IB̄ Qrec < QB and IB̄ ≤ Irec ≤ IB
Output:
|Bi
not-|Bi
previous
Extended Data Table I. Input-output table summarizing the behavior of the IQ filter implemented on the FPGA controller.
10
Symbol
Γ−1
κ−1
Tint
Γ−1
BG
∆tmid
Γ−1
GD
T1D
D
T2R
D
T2E
Γ−1
DG
Value
≈ 8.8 ns
Description
Effective measurement time of |Bi, approximately given by 1/κn̄, where n̄ = 5 ± 0.2 in the
main experiment (see Supplement Sec. II)
44.0 ± 0.06 ns Readout cavity lifetime
260.0 ns
Integration time of the measurement record, set in the controller at the beginning of the
experiment
0.99 ± 0.06 µs Average time the atom rests in |Gi before an excitation to |Bi, see Fig. 2b
3.95 µs
No-click duration for reaching ZGD = 0 in the flight of the quantum jump from |Gi to |Di, in
the full presence of ΩDG , see Fig. 3b
30.8 ± 0.4 µs Average time the atom stays in |Di before returning to |Gi and being detected, see Fig. 2b
116 ± 5 µs
Energy relaxation time of |Di
120 ± 5 µs
Ramsey coherence time of |Di
162 ± 6 µs
Echo coherence time of |Di
220 ± 5 µs
Average time between two consecutive |Gi to |Di jumps
Normalized counts
Extended Data Table II. Summary of timescales. List of the characteristic timescales involved in the catch and reverse
experiment. The Hamiltonian parameters of the system are summarized in Sec. I of the Supplementary Information.
τB (µs)
Extended Data Figure 2. Waiting time to switch from a |Bi to not-|Bi state assignment result. Semi-log plot
of the histogram (shaded green) of the duration of times corresponding to |Bi-measurement results, τB , for 3.2 s of continuous
data of the type shown in Fig. 2a. Solid line is an exponential fit, which yields a 4.2 ± 0.03 µs time constant.
a
Re ρc(∆tmid)
b
Im ρc(∆tmid)
Extended Data Figure 3. Mid-flight tomogram. The plots show the real (a) and imaginary (b) parts of the conditional
density matrix, ρc , at the mid flight of the quantum jump (∆tcatch = ∆tmid ), in the presence of the Rabi drive from |Gi to |Di
(∆toff = 0). The population of the |Bi state is 0.023, and the magnitude of all imaginary components is less than 0.007.
11
Population
ΩDG off
PG
PD
control
expt.
Reverse pulse angle φI
Extended Data Figure 4. Reversing the quantum jump mid-flight in the absence of ΩDG . Success probabilities
PG (purple) and PD (orange) to reverse to |Gi and complete to |Di the quantum jump mid-flight at ∆tcatch = ∆t′mid , defined
in Fig. 3b, in the absence of the Rabi drive ΩDG , where ∆ton = 2 µs and θI = π/2. Black dots: success probability for |Gi
(closed dots) and |Di (open dots) for the control experiment where the intervention is applied at random times, see Fig. 4b.
12
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
N. Bohr, “On the Constitution of Atoms and Molecules,” Philosophical Magazine 26, 476 (1913).
W. Nagourney, J. Sandberg, and H. Dehmelt, “Shelved optical electron amplifier: Observation of quantum jumps,” Physical
Review Letters 56, 2797 (1986).
T. Sauter, W. Neuhauser, R. Blatt, and P. E. Toschek, “Observation of Quantum Jumps,” Physical Review Letters 57,
1696 (1986).
J. C. Bergquist, R. G. Hulet, W. M. Itano, and D. J. Wineland, “Observation of Quantum Jumps in a Single Atom,”
Physical Review Letters 57, 1699 (1986).
H. J. Carmichael, An Open Systems Approach to Quantum Optics (Springer, Berlin, Heidelberg, 1993).
C. W. Gardiner, A. S. Parkins, and P. Zoller, “Wave-function quantum stochastic differential equations and quantum-jump
simulation methods,” Physical Review A 46, 4363 (1992).
J. Dalibard, Y. Castin, and K. Mølmer, “Wave-function approach to dissipative processes in quantum optics,” Physical
Review Letters 68, 580 (1992).
M. B. Plenio and P. L. Knight, “The quantum-jump approach to dissipative dynamics in quantum optics,” Reviews of
Modern Physics 70, 101 (1998), arXiv:9702007 [quant-ph].
A. N. Korotkov, “Continuous quantum measurement of a double dot,” Physical Review B 60, 5737 (1999).
A. Einstein, “Strahlungs-emission und -absorption nach der Quantentheorie (Emission and absorption of radiation in quantum
theory),” Verhandlungen der Deutschen Physikalischen Zeitschrift 18, 318 (1916).
A. Einstein, “Quantentheorie der Strahlung (On the quantum theory of radiation),” Physikalische Zeitschrift 18, 121 (1917).
E. Schrödinger, “Are There Quantum Jumps?” British Journal for the Philosophy of Science 3, 109 and 233 (1952).
T. Basche, S. Kummer, and C. Brauchle, “Direct spectroscopic observation of quantum jumps of a single molecule,” Nature
373, 132 (1995).
S. Peil and G. Gabrielse, “Observing the Quantum Limit of an Electron Cyclotron: QND Measurements of Quantum Jumps
between Fock States,” Physical Review Letters 83, 1287 (1999).
S. S. Gleyzes, S. Kuhr, C. Guerlin, J. Bernu, S. Deléglise, U. Busk, M. Brune, J.-M. Raimond, S. Haroche, S. Deleglise,
U. Busk Hoff, M. Brune, J.-M. Raimond, and S. Haroche, “Observing the quantum jumps of light : birth and death of a
photon in a cavity,” Nature 446, 297 (2007).
C. Guerlin, J. Bernu, S. Deléglise, C. Sayrin, S. Gleyzes, S. Kuhr, M. Brune, J.-M. Raimond, and S. Haroche, “Progressive
field-state collapse and quantum non-demolition photon counting.” Nature 448, 889 (2007), arXiv:0707.3880.
F. Jelezko, I. Popa, A. Gruber, C. Tietz, J. Wrachtrup, A. Nizovtsev, and S. Kilin, “Single spin states in a defect center
resolved by optical spectroscopy,” Applied Physics Letters 81, 2160 (2002).
P. Neumann, J. Beck, M. Steiner, F. Rempp, H. Fedder, P. R. Hemmer, J. Wrachtrup, and F. Jelezko, “Single-Shot Readout
of a Single Nuclear Spin,” Science 329, 542 (2010).
L. Robledo, L. Childress, H. Bernien, B. Hensen, P. F. A. Alkemade, and R. Hanson, “High-fidelity projective read-out of
a solid-state spin quantum register,” Nature 477, 574 (2011), arXiv:1301.0392v1.
R. Vijay, D. H. Slichter, and I. Siddiqi, “Observation of Quantum Jumps in a Superconducting Artificial Atom,” Physical
Review Letters 106, 110502 (2011), arXiv:1009.2969.
M. Hatridge, S. Shankar, M. Mirrahimi, F. Schackert, K. Geerlings, T. Brecht, K. M. Sliwa, B. Abdo, L. Frunzio, S. M.
Girvin, R. J. Schoelkopf, and M. H. Devoret, “Quantum Back-Action of an Individual Variable-Strength Measurement,”
Science 339, 178 (2013).
S. Deléglise, I. Dotsenko, C. Sayrin, J. Bernu, M. Brune, J.-M. Raimond, and S. Haroche, “Reconstruction of non-classical
cavity field states with snapshots of their decoherence,” Nature 455, 510 (2008), arXiv:0809.1064.
C. Sayrin, I. Dotsenko, X. Zhou, B. Peaudecerf, T. Rybarczyk, S. Gleyzes, P. Rouchon, M. Mirrahimi, H. Amini, M. Brune,
J.-M. Raimond, and S. Haroche, “Real-time quantum feedback prepares and stabilizes photon number states,” Nature 477,
73 (2011), arXiv:1107.4027.
L. Sun, A. Petrenko, Z. Leghtas, B. Vlastakis, G. Kirchmair, K. M. Sliwa, A. Narla, M. Hatridge, S. Shankar, J. Blumoff,
L. Frunzio, M. Mirrahimi, M. H. Devoret, and R. J. Schoelkopf, “Tracking Photon Jumps with Repeated Quantum NonDemolition Parity Measurements,” Nature 511, 444 (2013), arXiv:1311.2534.
N. Ofek, A. Petrenko, R. Heeres, P. Reinhold, Z. Leghtas, B. Vlastakis, Y. Liu, L. Frunzio, S. M. Girvin, L. Jiang, M. Mirrahimi, M. H. Devoret, and R. J. Schoelkopf, “Demonstrating Quantum Error Correction that Extends the Lifetime of
Quantum Information,” Nature 536, 441 (2016), arXiv:1602.04768.
M. Porrati and S. Putterman, “Wave-function collapse due to null measurements: The origin of intermittent atomic fluorescence,” Physical Review A 36, 929 (1987).
R. Ruskov, A. Mizel, and A. N. Korotkov, “Crossover of phase qubit dynamics in the presence of a negative-result weak
measurement,” Physical Review B 75, 220501 (2007).
J. Volz, R. Gehr, G. Dubois, J. Esteve, and J. Reichel, “Measurement of the internal state of a single atom without energy
exchange,” Nature 475, 210 (2011).
D. Ristè, M. Dukalski, C. A. Watson, G. De Lange, M. J. Tiggelman, Y. M. Blanter, K. W. Lehnert, R. N. Schouten, and
L. Dicarlo, “Deterministic entanglement of superconducting qubits by parity measurement and feedback,” Nature 502, 350
(2013), arXiv:1306.4002.
K. W. Murch, S. J. Weber, K. M. Beck, E. Ginossar, and I. Siddiqi, “Reduction of the radiative decay of atomic coherence
in squeezed vacuum,” Nature 499, 62 (2013), arXiv:1301.6276.
13
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
S. J. Weber, A. Chantasri, J. Dressel, A. N. Jordan, K. W. Murch, and I. Siddiqi, “Mapping the optimal route between two
quantum states,” Nature 511, 570 (2014), arXiv:1403.4992.
A similar phase reference for a non-unitary, yet deterministic, evolution induced by measurement was previously found in a
different context in: N. Katz, M. Ansmann, R. C. Bialczak, E. Lucero, R. McDermott, M. Neeley, M. Steffen, E. M. Weig,
A. N. Cleland, J. M. Martinis, and A. N. Korotkov, Science (New York, N.Y.) 312, 1498 (2006).
Reversal of quantum jumps have been theoretically considered in a different context, see H. Mabuchi and P. Zoller, Phys.
Rev. Lett. 76, 3108 (1996).
R. J. Cook, “What are Quantum Jumps?” Physica Scripta 1998, 49 (1988).
N. Bergeal, F. Schackert, M. Metcalfe, R. Vijay, V. E. Manucharyan, L. Frunzio, D. E. Prober, R. J. Schoelkopf, S. M.
Girvin, and M. H. Devoret, “Phase-preserving amplification near the quantum limit with a Josephson ring modulator,”
Nature 465, 64 (2010), arXiv:0912.3407.
Z. K. Minev, K. Serniak, I. M. Pop, Z. Leghtas, K. Sliwa, M. Hatridge, L. Frunzio, R. J. Schoelkopf, and M. H. Devoret,
“Planar Multilayer Circuit Quantum Electrodynamics,” Physical Review Applied 5, 044021 (2016), arXiv:1509.01619.
F. Lecocq, I. M. Pop, Z. Peng, I. Matei, T. Crozes, T. Fournier, C. Naud, W. Guichard, and O. Buisson, “Junction
fabrication by shadow evaporation without a suspended bridge,” Nanotechnology 22, 315302 (2011), arXiv:1101.4576.
C. T. Rigetti, Quantum Gates for Superconducting Qubits, Ph.D. thesis, Yale University (2009).
Z. Minev, Z. Leghtas, S. Mudhada, I. Pop, L. Christakis, R. Schoelkopf, and M. Devoret, “Energy Participation Approach
to the Design of Quantum Josephson Circuits,” (in preparation) (2018).
J. M. Chow, L. DiCarlo, J. M. Gambetta, F. Motzoi, L. Frunzio, S. M. Girvin, and R. J. Schoelkopf, “Optimized driving
of superconducting artificial atoms for improved single-qubit gates,” Physical Review A 82, 040305 (2010).
M. D. Reed, Entanglement and Quantum Error Correction with Superconducting Qubits, Ph.D. thesis, Yale University (2013).
J. Bylander, S. Gustavsson, F. Yan, F. Yoshihara, K. Harrabi, G. Fitch, D. G. Cory, Y. Nakamura, J.-S. Tsai, and W. D.
Oliver, “Noise spectroscopy through dynamical decoupling with a superconducting flux qubit,” Nature Physics 7, 565 (2011).
C. Eichler, C. Lang, J. M. Fink, J. Govenius, S. Filipp, and A. Wallraff, “Observation of Entanglement between Itinerant
Microwave Photons and a Superconducting Qubit,” Physical Review Letters 109, 240501 (2012).
Y. Liu, Quantum Feedback Control of Multiple Superconducting Qubits, Ph.D. thesis, Yale University (2016).
D. Ristè, C. C. Bultink, K. W. Lehnert, and L. DiCarlo, “Feedback Control of a Solid-State Qubit Using High-Fidelity
Projective Measurement,” Physical Review Letters 109, 240502 (2012).
Supplementary Information for:
“To catch and reverse a quantum jump mid-flight”
Z.K. Minev,1, ∗ S.O. Mundhada,1 S. Shankar,1 P. Reinhold,1 R. Gutiérrez-Jáuregui,2
R.J. Schoelkopf,1 M. Mirrahimi,3, 4 H.J. Carmichael,2 and M.H. Devoret1
1
arXiv:1803.00545v3 [quant-ph] 12 Feb 2019
2
Department of Applied Physics, Yale University, New Haven, Connecticut 06511, USA
The Dodd-Walls Centre for Photonic and Quantum Technologies, Department of Physics,
University of Auckland, Private Bag 92019, Auckland, New Zealand
3
Yale Quantum Institute, Yale University, New Haven, Connecticut 06520, USA
4
QUANTIC team, INRIA de Paris, 2 Rue Simone Iff, 75012 Paris, France
(Dated: February 14, 2019)
CONTENTS
I. Experimental characterization of the system
II. Quantum trajectory theory
A. Fluorescence monitored by photon counts
1. Coherent Bright drive
2. Incoherent Bright drive
3. Dark drive off
4. Completed and aborted evolutions of the jump transition
B. Bright state monitored by dispersive cavity readout
1. Stochastic Schrödinger equation
2. Independently measured imperfections
3. Leakage from the GBD-manifold
III. Comparison between experiment and theory
A. Simulated data sets
B. Error budget
1. Imperfections
2. Budget for lost coherence
C. Signal-to-noise ratio (SNR) and de-excitation measurement efficiency
1. Indirect monitoring method with superconducting circuits
References
2
4
4
4
5
6
7
7
7
8
8
9
9
10
10
10
12
13
14
2
I.
EXPERIMENTAL CHARACTERIZATION OF THE SYSTEM
Hamiltonian of the device. The two-transmon, single-readout-cavity system is well described, in the low-excitation
manifold, by the approximate dispersive Hamiltonian1,2 :
1
1
Ĥ/~ = ωB b̂† b̂ − αB b̂†2 b̂2 + ωD dˆ† dˆ − αD dˆ†2 dˆ2 − χDB b̂† b̂dˆ† dˆ
2
2
†
†ˆ †
ˆ
+ ωC + χB b̂ b̂ + χD d d ĉ ĉ ,
(1)
where ωC , ωB and ωD are the cavity, bright, and dark qubit transition frequencies, ĉ, b̂ and dˆ are the associated ladder
operators, and α and χ are the modal anharmonicities and dispersive shifts, respectively.3 The states |Bi and |Di
correspond to one excitation in the bright (b̂† |0i) and dark (dˆ† |0i) modes, respectively. The measured parameters of
the device and the mode coherences are summarized in Table S1.
DG coherence & tomography control. In Fig. S1, we show the results of a control experiment where we verified the
D
Ramsey coherence (T2R
) and energy relaxation (T1D ) times of the DG transition with our tomography method. Solid
D
= 119.2 µs
lines are fitted theoretical curves for the free evolution of the prepared initial state √12 (|Di − |Gi). The T2R
value gained from the simultaneous fit of XDG (t) and YDG (t) matches the lifetime independently obtained from a
standard T2R measurement. Similarly, the value of T1D = 115.4 µs extracted from an exponential fit of ZDG (t) matches
the value obtained from a standard T1 measurement. We note that our tomography procedure is well calibrated and
skew-free, as evident in the zero steady-state values of XDG and YDG . The steady state ZDG corresponds to the
thermal population of the dark state nD
th . It has recently been shown that residual thermal populations in cQED
systems can be significantly reduced by properly thermalizing the input-output lines.4,5
Measurement-induced energy relaxation T1 (n̄). Figure S2 shows a characterization of the parasitic measurementinduced energy relaxation of |Bi and |Di. As is the case in standard cQED systems6–8 , the |Bi level shows strong
T1 degradation9–12 as a function of the readout drive strength n̄. However, the lifetime of the dark state (|Di) is
protected, and remains largely unaffected even at large drive strengths (n̄ ≈ 50).
XGD
YGD
ZGD
N
Time from start to tomography (µs)
Supplementary Figure S1.
Control experiment: time-resolved tomogram of the free evolution of a DG
superposition. The atom is prepared in √12 (|Di − |Gi) and tomography is performed after a varied delay. Dots: reconstructed
conditional GD tomogram (XDG , YDG , and ZDG ) and population in DG manifold, N (see Methods). Solid lines: theoretical
fits.
3
Readout cavity
BG transition
DG transition
Mode frequencies and non-linear parameters
ωC /2π = 8979.640 MHz
ωBG /2π = 5570.349 MHz
ωDG /2π = 4845.255 MHz
χB /2π = −5.08 ± 0.2 MHz
χD /2π = −0.33 ± 0.08 MHz
αB /2π = 195 ± 2 MHz
αD /2π = 152 ± 2 MHz
χDB /2π = 61 ± 2 MHz
Coherence related parameters
T1B = 28 ± 2 µs
T1D = 116 ± 5 µs
B
= 18 ± 1 µs
T2R
D
= 120 ± 5 µs
T2R
Tint = 260.0 ns
B
T2E
D
= 162 ± 6 µs
T2E
nC
th ≤ 0.0017 ± 0.0002
nB
th ≤ 0.01 ± 0.005
κ/2π= 3.62 ± 0.05 MHz
η = 0.33 ± 0.03
= 25 ± 2 µs
nD
th ≤ 0.05 ± 0.01
Drive amplitude and detuning parameters
n̄ = 5.0 ± 0.2
ΩB0 /2π = 1.2 ± 0.01 MHz
ΩDG /2π = 20 ± 2 kHz
ΩB1 /2π = 600 ± 10 kHz
∆B1 /2π = −30.0 MHz
∆R = χ B
∆DG /2π = −275.0 kHz
Supplementary Table S1. Compilation of the experimental parameters.
Bright B
T1-1 (µs-1)
Dark D
guide
for eye
Readout drive strength n
Supplementary Figure S2. Measurement-induced energy relaxation T1 (n̄). Energy relaxation rate (T1−1 ) of |Bi
(blue dots) and |Di (red dots) as a function of n̄, measured with the following protocol: after the atom is prepared in either
|Bi or |Di, the readout tone (R) is turned on for duration tread with amplitude n̄ (corresponding to the number of steady-state
photons in the readout cavity when excited on resonance), whereafter the population of the initial state is measured. As in
all other experiments, the readout drive is applied at the |Bi cavity frequency (ωC − χB ). The relaxation rates are extracted
from exponential fits of the population decay as a function of tread , from 1.3 × 107 experimental realizations. The solids lines
are guides to the eye: blue line indicates the rapid degradation of T1B as a function of the readout strength, while the red line
indicates the nearly constants T1D of the protected dark level.
4
II.
A.
QUANTUM TRAJECTORY THEORY
Fluorescence monitored by photon counts
1.
Coherent Bright drive
The experiments with trapped ions13–15 monitor intermittent fluorescence from the bright state |Bi to track jumps
between |Gi and |Di.16 In the simplest three-level scheme,15 and using coherent radiation to excite both transitions,
the master equation for the reduced density operator ρ of the three-level system, written in the interaction picture, is
dρ
= (i~)−1 [Ĥdrive , ρ] + γB D [|GihB|] ρ + γD D [|GihD|] ρ,
dt
(2)
ˆ · } denotes the Lindblad superoperator, γB and γD are radiative decay rates, and
ˆ · = ξˆ · ξ † − 1 {ξˆ† ξ,
where D[ξ]
2
Ĥdrive = i~
ΩBG
ΩDG
|BihG| − |GihB| + i~
|DihG| − |GihD| ,
2
2
(3)
with ΩBG and ΩDG the Rabi drives. The quantum trajectory description17–19 unravels ρ into an ensemble of pure
states whose ket vectors evolve along stochastic paths conditioned on the clicks of imaginary photon detectors that
monitor fluorescence from |Bi (and much less frequently from |Di). In recognition of each click the ket vector is reset
to |Gi, while otherwise it follows a deterministic evolution as a coherent superposition,
|ψ(∆tcatch )i = CG (∆tcatch )|Gi + CB (∆tcatch )|Bi + CD (∆tcatch )|Di,
(4)
where CG (0) = 1, CB (0) = CD (0) = 0, with ∆tcatch = 0 marking the time of the last click reset, and
i~
γB
γD
d|ψi
= Ĥdrive − i~ |BihB| − i~ |DihD| |ψi.
d∆tcatch
2
2
(5)
The norm of the ket |ψ(∆tcatch )i is not preserved, but rather gives the probability that the evolution will continue,
with no interruption by further clicks, up to time ∆tcatch ; clearly it must decay with this probability to zero. If we
then define
WDG (∆tcatch ) ≡
CD (∆tcatch )
,
CG (∆tcatch )
(6)
the normalized ket vector in the GD-subspace has Bloch vector components
ZGD (∆tcatch ) =
−1
WDG (∆tcatch ) − WDG
(∆tcatch )
,
−1
WDG (∆tcatch ) + WDG (∆tcatch )
(7)
XGD (∆tcatch ) =
2
,
−1
WDG (∆tcatch ) + WDG
(∆tcatch )
(8)
(9)
YGD (∆tcatch ) = 0,
where, using Eqs. (4) and (5),
CG
CG
0 −ΩBG −ΩDG
d
1
CB = ΩBG −γB
0 CB .
d∆tcatch C
2 Ω
CD
0
−γD
D
DG
(10)
In general this 3 × 3 system does not have a closed solution in simple form, although there is a particularly simple
solution under conditions that produce intermittent fluorescence, i.e., rare jumps from |Gi to |Di (“shelving” in the
dark state13 ) interspersed as intervals of fluorescence “off” in a background of fluorescence “on”. The conditions follow
−1
naturally if |Di is a metastable state,13–15,20 whose lifetime γD
is extremely long on the scale of the mean time
between photon detector clicks for a weak ΩBG Rabi drive,
Γ−1
BG
=
Ω2BG
γB
−1
.
(11)
5
Thus, for (ΩDG , γD ) ≪ Ω2BG /γB ≪ γB , Eq. (10) yields the equation of motion
dWDG
Ω2
ΩDG
= BG WDG +
,
d∆tcatch
2γB
2
whose solution for the click reset initial condition, WDG (0) = 0, is
2
ΩDG
ΩBG
WDG (∆tcatch ) = 2
∆tcatch − 1 ,
exp
ΩBG /γB
2γB
(12)
(13)
from which a long enough interval with no clicks gives WDG (∆tcatch ) ≫ 1 and leads to the conclusion that the
ket vector is |Di. The time scale for the transition, ∆tmid , is defined by ZGD (∆tmid ) = 0, which corresponds to
WDG (∆tmid ) = 1.21 Simply inverting Eq. (13) produces the formula
∆tmid =
Ω2BG
2γB
−1
ln
Ω2BG /γB
+1 ,
ΩDG
(14)
but strong monitoring, Ω2BG /γB ≫ ΩDG , allows the −1, in Eq. (13), and +1, in Eq. (14), to be dropped. Equations
(7)–(9), (13), and (14) then provide simple formulas for the continuous, deterministic, and coherent evolution of the
quantum jump when completed:
2
ΩBG
ZGD (∆tcatch ) = tanh
(∆tcatch − ∆tmid ) ,
(15)
2γB
2
ΩBG
(∆tcatch − ∆tmid ) ,
(16)
XGD (∆tcatch ) = sech
2γB
YGD (∆tcatch ) = 0.
(17)
These formulas execute a perfect jump, ZGD (∞) = 1, XGD (∞) = YGD (∞) = 0. The ideal arises from the assumed
strong monitoring, ΩDG ≪ Ω2BG /γB . Departures from it can be transparently analyzed by adopting an incoherent
Bright drive, see Sec. II A 2. An elegant analysis of the no-click evolution for arbitrary amplitude of the Dark Rabi
drive can be found in Refs. 22 and 23. For an interesting connection to of the three-level intermittent dynamics to
dynamical phase transitions, see Refs. 24 and 25.
Application of the photon counting model to the experiment. The photon-counting theory presented in this section
provides the background to the experiment along with a link to the original ion experiments. It captures a core set
of the ideas, even though the monitoring of |Bi implemented in the experiment is diffusive — the opposite limit of
the point-process description presented here, see Sec. II B. Nevertheless, the photon-counting theory even provides a
quantitative first approximation of the experimental results. For definitiveness, consider the flight of the quantum
jump shown in Fig. 3b. The measured mid-flight time, ∆tmid = 3.95 µs, is predicted, in a first approximation, by
Eq. (14). Using the (independently measured) values of the experimental parameters, summarized in Table S1 (setting
ΩBG equal to ΩB0 = 2π×1.2 MHz, the BG drive when the atom is not in |Bi) and extracting the effective measurement
rate of |Bi, γB = 2π × 9.0 MHz (which follows from Eq. (11) where ΓBG = 2π × 1.01 MHz, the average click rate on the
BG transition), Eq. (14) predicts ∆tmid ≈ 4.3 µs — in fair agreement with the observed value ∆tmid = 3.95 µs. The
photodetection theory presented in in Sec. II A 2 further improves the agreement. These calculations serve to generally
illustrate the theory and ideas of the experiment; the quantitative comparison between theory and experiment is only
presented in Sec. III.
2.
Incoherent Bright drive
If the coherent Rabi drive ΩBG is replaced by an incoherent drive ΓBG , the master equation in the interaction
picture becomes
dρ
= (i~)−1 [Ĥdrive , ρ] + ΓBG D[|BihG|]ρ + (γB + ΓBG )D[|GihB|]ρ + γD D[|GihD|]ρ,
dt
(18)
where
Ĥdrive = i~
ΩDG
|DihG| − |GihD| .
2
(19)
6
The previous weak drive assumption, Ω2BG /γB ≪ γB , is now carried over with the assumption ΓBG ≪ γB , which
says that the time between clicks in fluorescence is essentially the same as the time separating photon absorptions
from the incoherent drive, as absorption is rapidly followed by fluorescence (γB + ΓBG ≫ ΓBG ). This brings a useful
simplification, since, following each reset to |Gi, the unnormalized ket evolves in the GD-subspace,
d|ψi
ΓBG
γD
i~
= Hdrive − i~
|GihG| − i~ |DihD| |ψi,
(20)
d∆tcatch
2
2
thus replacing Eqs. (10) and (12) by the simpler 2 × 2 system
d
1 −ΓBG −ΩDG CG
CG
=
,
CD
d∆tcatch CD
2 ΩDG −γD
(21)
and, if γD ≪ ΓBG , the equation of motion
dWDG
ΓBG
ΩDG
2
=
WDG +
(1 + WDG
),
d∆tcatch
2
2
(22)
exp (V − V −1 )ΩDG ∆tcatch /2 − 1
,
WDG (∆tcatch ) =
V − V −1 exp [(V − V −1 )ΩDG ∆tcatch /2]
(23)
with solution, for WDG (0) = 0,
where
1 ΓBG
V =
+
2 ΩDG
s
2
1 ΓBG
− 1.
4 ΩDG
(24)
In Ref. 22, a general form of the Bloch vector equations for arbitrary amplitude of the Rabi drive was found. Inversion
of the condition WDG (∆tmid ) = 1 gives the characteristic time scale
−1
V +1
.
(25)
∆tmid = 2 (V − V −1 )ΩDG
ln
V −1 + 1
Equations (23)–(25) replace Eqs. (13) and (14); although, under strong monitoring (ΓBG ≫ ΩDG ), they revert to these
results with the substitution Ω2BG /2γB → ΓBG /2, recovering Eqs. (7)–(9) with the same substitution. More generally,
WDG (∆tcatch ) goes to infinity at finite ∆tcatch , changes sign, and returns from infinity to settle on the steady value
WDG (∞) = −V . The singular behavior marks a trajectory passing through the north pole of Bloch sphere. It yields
the long-time solution
s
2
ΩDG
ΩDG
ZGD (∞) = 1 − 4
,
XGD (∞) = −2
,
YGD (∞) = 0,
(26)
ΓBG
ΓBG
in contrast to the perfect jump of Eqs. (15)–(17).
3.
Dark drive off
Turing the Dark drive off shortly after a click reset demonstrates the connection between the flight of a quantum
jump and a projective measurement. From the point of view of the trajectory equations, the only change is the setting
of ΩDG to zero at time ∆ton on the right-hand side of Eqs. (12) and (22). Subsequently, WDG (∆tcatch ) continues its
exponential growth at rate Ω2BG /2γB [Eq. (12)] or ΓBG /2 [Eq. (22)]. Equations (7)–(9) for the GD Bloch components
still hold, but now with
∆tmid =
Ω2BG ΓBG
,
2γB
2
−1
−2
ln WDG
(∆ton ) ,
which can provide an estimate of ∆t′mid , specifying the time at which ZGD = 0.
(27)
7
The evolution during ∆toff , in the absence of ΩDG , in effect realizes a projective measurement of whether the state
of the atom is |Gi or |Di, where the normalized state at ∆ton is
|ψ(∆ton )i
CG (∆ton )|Gi + CD (∆ton )|Di
p
p
=
,
N (∆ton )
N (∆ton )
(28)
2
2
with N (∆ton ) = CG
(∆ton ) + CD
(∆ton ) the probability for the jump to reach ∆tcatch = ∆ton after a click reset to |Gi
at ∆tcatch = 0. The probability for the jump to continue to ∆tcatch > ∆ton (given ∆ton is reached) is then
2
2
2
ΩBG
CD
(∆ton ) CG
(∆ton )
N (∆tcatch )
=
+
exp −
, ΓBG ∆tcatch .
(29)
N (∆ton )
N (∆ton )
N (∆ton )
γB
4.
Completed and aborted evolutions of the jump transition
In this simple model, the probability for the trajectory to complete — for the measurement to yield the result
|Di — is obtained in the limit ∆tcatch → ∞, and, as expected, is equal to the probability to occupy the state |Di
2
at time ∆ton ; i.e., the completion probability is PD (∆ton ) = CD
(∆ton )/Norm(∆ton ). It is helpful to illustrate this
idea with an example. Consider the catch experiment of Fig. 3b in the absence of the Dark Rabi drive, ΩDG . From
ZGD , we can estimate that out of all the trajectories that pass though the ∆ton mark approximately PD (∆ton ) =
(1 + ZGD (∆ton ))/2 ≈ 8% fully complete without an interruption. On the other hand, for those that pass the ∆t′mid
mark, approximately 50% complete. It follows from Eq. (29), that the probability of the evolution to complete
increases the further along the trajectory is. Although some of the jump evolutions abort at random, importantly,
every single jump evolution that completes, and is thus recorded as a jump, follows not a random but an identical
path in Hilbert space, i.e., a deterministic one. This path (of any jump) is determined by Eq. (23), or, in the simpler
model, by the Eqs. (15)-(17) for the components of the GD Bloch vector.
B.
Bright state monitored by dispersive cavity readout
1.
Stochastic Schrödinger equation
Monitoring the quantum jump through fluorescence photon counts provides a clean and simple way of analyzing
the deterministic character of the evolution. It is prohibitively challenging for an experiment, though, as the time
origin ∆tcatch = 0 is set by the click reset to |Gi, and in an ensemble of measurements, all resets must be aligned on
the very last click before an interval of deterministic evolution (Eqs. (7)–(9)) in order for ∆tmid to be aligned over
the ensemble; low detection efficiency—on the order of 10−3 or less13–15 —in the first ion experiments does not permit
this. Monitoring through a dispersive cavity readout provides a robust way of aligning ∆tcatch = 0, hence ∆tmid , over
an ensemble of measurements. Leaving aside imperfections (see below), the master equation in the interaction picture
is
dρ
= (i~)−1 [Ĥdrive , ρ] + (i~)−1 [ĤR , ρ] + κD[ĉ]ρ,
(30)
dt
with
ΩBG (t)
Ω∗BG (t)
ΩDG
Ĥdrive = i~
|DihG| − |GihD| ,
(31)
|BihG| −
|GihB| + i~
2
2
2
and
ĤR = −~∆R ĉ† ĉ + i~
κ√ †
n̄(ĉ − ĉ) + ~ χB |BihB| + χD |DihD| ĉ† ĉ,
2
(32)
where the bi-chromatic drive ΩBG (t) = ΩB0 + ΩB1 exp(−i∆B1 t) replaces the Rabi drive ΩBG of Eq. (3), n̄ is the
mean photon number in the readout cavity when driven on resonance, and ∆R is the detuning of the probe from
the unshifted cavity resonance; the bi-chromatic drive facilitates transitions in both directions between |Gi and |Bi,
given that the bright level shifts when the cavity fills with photons. The quantum trajectory unraveling monitors the
reflected probe with efficiency η and accounts for residual photon loss through random jumps; thus, the stochastic
Schrödinger equation combines a continuous evolution (heterodyne readout channel),
1
κ †
√ √
∗
d|ψi =
Ĥdrive + ĤR − i~ ĉ ĉ dt + η κdζ ĉ |ψi,
(33)
i~
2
8
where
√ √ hψ|â|ψi
dt + dZ,
(34)
η κ
hψ|ψi
dZ is a complex Wiener increment, obeying E [dZ] = 0, E dZ 2 = 0, and Var [dZ] = E [dZ ∗ dZ] = dt, with random
jumps (photon loss),
dζ =
|ψi → ĉ|ψi
at rate
(1 − η)κ
hψ|ĉ† ĉ|ψi
.
hψ|ψi
(35)
The monitored output dζ is scaled—to units of (readout cavity photon number)1/2 —and filtered to generate simulated
quadratures Irec and Qrec of the measurement record:
κ −1/2
κfilter
Re(dζ) ,
(36)
Irec dt − η
dIrec = −
2
2
κ −1/2
κfilter
Im(dζ) ,
(37)
Qrec dt − η
dQrec = −
2
2
where κfilter is the bandwidth of the experimental readout amplifier chain.
2.
Independently measured imperfections
The stochastic Schrödinger equation is supplemented by spontaneous and thermal jumps on both the |Gi to |Bi
and |Gi to |Di transitions, and by pure dephasing of the GB and GD coherences. With these processes included, the
term
h
i
hγ
i
D
γ B nB
γB B
D
φ
φ
th + γD nth
− i~
(nth + 1) + γB
|BihB| +
(nD
|DihD|
+
|GihG|
+
1)
+
γ
D
2
2 th
2
is added to the non-Hermitian Hamiltonian Ĥdrive + ĤR − i~(κ/2)ĉ† ĉ on the right-hand side of Eq. (33), and there
are additional random jumps
hψ|BihB|ψi
hψ|DihD|ψi
+ γD (nD
,
th + 1)
hψ|ψi
hψ|ψi
|ψi → |Gi
at rate
γB (nB
th + 1)
|ψi → |Bi
at rate
γ B nB
th
hψ|GihG|ψi
φ hψ|BihB|ψi
+ 2γB
,
hψ|ψi
hψ|ψi
(39)
|ψi → |Di
at rate
γ D nD
th
hψ|GihG|ψi
φ hψ|DihD|ψi
+ 2γD
.
hψ|ψi
hψ|ψi
(40)
(38)
φ
G,D
B,D
1
The parameters γB,D , nB,D
th , and γB,D are mapped to the independently measured parameters TB,D , nth , and T2R
listed in Table S2 (see below).
3.
Leakage from the GBD-manifold
The three-state manifold, |Gi, |Bi, and |Di, is not strictly closed. Rare transitions to higher excited states of the
two-transmon system may occur. This possibility is included with the addition of the further term
γFG
γFD
γGF + γDF
− i~
|GihG| +
|DihD| +
|FihF|
2
2
2
to the non-Hermitian Hamiltonian, and the associated additional random jumps,
|ψi → |Fi
at rate
γFG
hψ|DihD|ψi
hψ|GihG|ψi
+ γFD
,
hψ|ψi
hψ|ψi
(41)
|ψi → |Gi
at rate
γGF
hψ|FihF|ψi
,
hψ|ψi
(42)
|ψi → |Di
at rate
γDF
hψ|FihF|ψi
,
hψ|ψi
(43)
9
where |Fi is a single catch-all higher excited state.
III.
COMPARISON BETWEEN EXPERIMENT AND THEORY
A.
Simulated data sets
Independently measured parameters. The parameters used in the simulations are listed in Table S2. In most cases
they are set to the value at the center of the range quoted in Table S1, but with three exceptions: (i) T1B and T1D are
set to lower values in response to the photon number dependence of the readout displayed in Fig. S2; (ii) ΩDG /2π is
set higher, but still falls inside the experimental error bars, and (iii) nC
th = 0. Of the three exceptions, only ΩDG /2π
has a noticeable effect on the comparison between simulated and experimental data sets.
Leakage from the GBD-manifold. Additional random jumps to state |Fi are governed by four parameters that are
not independently measured; they serve as fitting parameters, required to bring the simulation into agreement with
the asymptotic behavior of Z(∆tcatch ), which, without leakage to |Fi, settles to a value higher than is measured in
the experiment. The evolution of the X(∆tcatch ) is largely unaffected by the assignment of these parameters, where
any change that does occur can be offset by adjusting ΩDG /2π while staying within the experimental error bars.
Ensemble average. Simulated data sets are computed as an ensemble average by sampling an ongoing Monte Carlo
simulation, numerically implementing the model outlined in Eqs. (33)–(43). Quadratures Irec and Qrec are computed
from Eqs. (36) and (37), digitized with integration time Tint = 260 ns, and then, as in the experiment, a hysteric filter
is used to locate “click” events (∆tcatch = 0) corresponding to an inferred change of state from |Bi to not-|Bi. During
the subsequent sampling interval (∆tcatch ≥ 0), the four quantities
j
rec
rec
rec
ZjGD , XjGD , YGD
, PjBB (∆tcatch ) = Zrec
GD , XGD , YGD , PBB (tj + ∆tcatch ),
(44)
with tj is the click time and
Zrec
GD (t) =
hD|ψ(t)ihψ(t)|Di − hG|ψ(t)ihψ(t)|Gi
,
hψ(t)|ψ(t)i
rec
Xrec
GD (t) + iYGD (t) = 2
Prec
BB (t) =
hD|ψ(t)ihψ(t)|Gi
,
hψ(t)|ψ(t)i
hB|ψ(t)ihψ(t)|Bi
,
hψ(t)|ψ(t)i
(45)
(46)
(47)
are computed, and running sums of each are updated. The sample terminates when the measurement record indicates
a change of state from not-|Bi back to |Bi. Finally, for comparison with the experiment, Bloch vector components
are recovered from the average over sample intervals via the formula
ZGD , XGD , YGD
PN (∆tcatch )
j
ZjGD , XjGD , YGD
(∆tcatch )
(∆tcatch ) =
,
PN (∆tcatch ) j
N (∆tcatch ) − j
PBB (∆tcatch )
j
(48)
where N (∆tcatch ) is the number of sample intervals that extend up to, or beyond, the time ∆tcatch . The simulation
and sampling procedure is illustrated in Fig. S3, and a comparison between the experiment and the simulation is
provided in Fig. S4.
The simulated and measured Bloch vector components are fit with expressions motivated by Eqs. (15)-(17) and (26),
modified to account for the effect of non-idealities in the experiment,
ZGD (∆tcatch ) = a + b tanh(∆tcatch /τ + c),
XGD (∆tcatch ) = a′ + b′ sech(∆tcatch /τ ′ + c′ ) ,
YGD (∆tcatch ) = 0 .
(49)
(50)
(51)
The fit parameters (a, a′ , b, b′ , c, c′ , τ, τ ′ ) for the simulated and experimental data shown in Fig. S4 are compared in
Table S3. As imposed by Eq. (26), in the absence of ΩDG (turned off at time ∆ton = 2 µs) a′ , the offset of XGD , is
strictly enforced to be zero. The extracted simulation and experiment parameters are found to agree at the percent
level.
10
Readout cavity
BG transition
DG transition
Non-linear parameters
χB /2π = −5.08 MHz
χD /2π = −0.33 MHz
Coherence related parameters
κ/2π = 3.62 MHz
η = 0.33
T1B = 15 µs
T1D = 105 µs
T2B = 18 µs
T2D = 120 µs
nB
th = 0.01
nD
th = 0.05
Tint = 260.0 ns
nC
th = 0
Drive amplitude and detuning parameters
n̄ = 5.0
ΩB0 /2π = 1.2 MHz
ΩDG /2π = 21.6 kHz
ΩB1 /2π = 600 kHz
∆R = χ B
∆B1 /2π = −30.0 MHz
∆DG /2π = −274.5 kHz
Supplementary Table S2. Compilation of the simulation parameters.
B.
1.
Error budget
Imperfections
Various imperfections are expected to reduce the maximum coherence recovered in the measurement of XGD (∆tcatch ).
They include:
(i) Readout errors when inferring |Bi to not-|Bi transitions and the reverse. Such errors affect the assignment of
∆tcatch , which can be either too short or too long to correlate correctly with the true state of the system.
(ii) Leaks from the GBD-manifold to higher excited states. These errors mimic a |Bi to not-|Bi transition, as in the
first sample interval of Fig. S3, yet the anticipated coherent evolution within the GBD-manifold does not occur.
(iii) Thermal jumps from |Gi to |Di. Such incoherent transitions contribute in a similar way to ZGD (∆tcatch ), while
making no contribution to the measured coherence.
(iv) Direct dephasing of the DG-coherence.
(v) Partial distinguishability of |Gi and |Di. The readout cavity is not entirely empty of photons when the state is
not-|Bi, in which case the cross-Kerr interaction χD |DihD|ĉ† ĉ shifts the ΩDG Rabi drive from resonance; hence,
backaction noise is transferred from the photon number to XGD (∆tcatch ).
2.
Budget for lost coherence
The maximum coherence reported in the experiment is 0.71 ± 0.005. In the simulation it is a little lower at 0.69.
By removing the imperfections from the simulation, one by one, we can assign a fraction of the total coherence loss
to each. Readout errors are eliminated by identifying transitions between |Bi and not-|Bi in the ket |ψi rather than
from the simulated measurement record; all other imperfections are turned off by setting some parameter to zero.
The largest coherence loss comes from readout errors, whose elimination raises the XGD (∆tcatch ) maximum by 0.09.
The next largest comes from leakage to higher excited states, which raises the maximum by a further 0.06. Setting χD
to zero adds a further 0.04, and thermal transitions and pure dephasing together add 0.02. Figure S5 illustrates the
change in the distribution of XjGD (∆tcatch ) samples underlying the recovery of coherence. The removal of the finger
pointing to the left in panel (a) is mainly brought about by the elimination of readout errors, while the reduced line
of zero coherence marks the elimination of leakage to higher excited states. Aside from these two largest changes,
there is also a sharpening of the distribution, at a given ∆tcatch , when moving from panel (a) to panel (b). Having
addressed the five listed imperfections, a further 10% loss remains unaccounted for, i.e., the distribution of panel (b)
11
j
ZGD
a
1
1
∆tcatch
j
YGD
0
Irec (photon)1/2
b
tj
j
XGD
-1
j
ZGD
j
XGD
0
j
√5
YGD
0
0
100
200
300
Time-record position (µs)
400
-1
500
204
208
212
Time-record position (µs)
216
Supplementary Figure S3. Sampling of the Monte-Carlo simulation. a, Simulated measurement quadrature Irec and
correlated trajectory computed from Eqs. (45) and (46). Three sample intervals are shown. The earliest corresponds to leakage
from the GBD-manifold, where a jump from |Gi to |Fi is followed by a jump from |Fi to |Di. The second and third sample
intervals correspond to direct transitions from |Gi to |Di, which are continuously monitored and the object of the experiment.
b, Expanded view of the shaded region of the second sample interval in panel (a). The evolution is continuous but not smooth,
due to backaction noise from the continuously monitored readout. This feature is in sharp contrast to the perfect “no-click”
readout upon which the simple theory of Sec. II A is based.
a
1.0
ZGD
simulation
XGD
0.5
experiment
0.0
YGD
-0.5
-1.0
0
2
4
6
8
10
12
10
12
Catch time ∆tcatch(µs)
b
1.0
ZGD
XGD
0.5
0.0
YGD
-0.5
-1.0
0
2
4
6
8
Catch time ∆tcatch(µs)
Supplementary Figure S4. Comparison between simulation and experiment. a, Simulated data set obtained with
Rabi drive ΩDG turned on for the entire ∆tcatch ; parameters taken from Table S2 and leakage from the GBD-manifold included
with (γFG , γFD )/2π = 0.38 kHz and (γGF , γDF )/2π = 11.24 kHz. b, Simulated data set obtained with Rabi drive ΩDG turned off
at time ∆ton = 2 µs; parameters taken from Table S2 and leakage from the GBD-manifold included with γFG /2π = 0.217 kHz,
γFD /2π = 4.34 kHz, γGF /2π = 11.08 kHz, and γDF /2π = 15.88 kHz. When leakage from the GBD-manifold is omitted, the ZGD
curve rises more sharply and settles to a value that is 10% (20%) higher in panel (a) (panel (b)).
12
(a) In presence of ΩDG
Parameter
Experiment
-0.07
-0.21
0.94
0.93
-2.32
-2.04
1.64
1.74
a
a′
b
b′
c
c′
τ
τ′
±
±
±
±
±
±
±
±
0.005
0.005
0.005
0.005
0.03
0.03
0.01
0.01
(b) In absence of ΩDG
Simulation
-0.07
-0.22
0.95
0.91
-2.27
-2.05
1.65
1.76
±
±
±
±
±
±
±
±
0.005
0.005
0.005
0.005
0.03
0.03
0.01
0.01
Error
Parameter
0.5%
2%
1%
2%
2%
0.5%
0.5%
1%
a
a′
b
b′
c
c′
τ
τ′
Experiment
-0.11
0
0.92
0.61
-1.96
-1.97
2.17
1.98
±
±
±
±
±
±
±
±
Simulation
0.005 -0.10 ± 0.005
0
0 ± 0
0.008 0.91 ± 0.008
0.005 0.60 ± 0.005
0.05 -2.10 ± 0.05
0.05 -2.05 ± 0.05
0.05
2.03 ± 0.05
0.05
1.92 ± 0.05
Error
8%
0%
1%
2%
7%
4%
6%
3%
Supplementary Table S3. Comparison between parameters extracted from the simulation and those from the
experiment. a, Parameters obtained from fits of the simulated and measured data for the catch protocol in the presence of
the Rabi drive ΩDG throughout the entire duration of the quantum jump, data shown in Fig. S4a. b, Parameters obtained
from fits of the simulated and measured data for the catch protocol in the absence of the ΩDG during the flight of the quantum
jump for ∆ton = 2 µs, data shown in Fig. S4b.
is not a line passing through XjGD (∆tmid ) = 1. The final 10% is explained by the heterodyne detection backaction
noise, a function of the drive and measurement parameters, displayed in panel (b) of Fig. S3.
a
b
1.0
0.10
0.050
0.040
0.030
0.020
0.015
0.010
0.0075
0.0050
0.0025
0.5
j
XGD
0.0
∆tmid
-0.5
∆tmid
-1.0
0
2
4
6
8
Catch time ∆tcatch(µs)
10
12
0
2
4
6
8
10
12
Catch time ∆tcatch(µs)
Supplementary Figure S5.
Coherence loss through sample to sample fluctuations. a, Contour plot of the
distribution of XjGD (∆tcatch ) samples corresponding to the simulated data set displayed in panel (a) of Fig. S4. b, Same as
panel (a) but with transitions between |Bi and not-|Bi identified in the ket |ψi rather than from the simulated measurement
D
D
D
record, and with changed parameters: (γFG , γFD , γGF , γDF )/2π = 0, nB
th = nth = 0, T2 = 2T1 , and χD /2π = 0.
C.
Signal-to-noise ratio (SNR) and de-excitation measurement efficiency
As discussed in the Methods section, the catch protocol hinges on the efficient detection of de-excitations from
|Bi to |Gi. In atomic physics, de-excitations are typically monitored by a direct detection method, employing a
photodetector. Alternatively, de-excitations can be monitored by an indirect method, as done in our experiment.
In this subsection, we discuss the efficiency of both methods. For the indirect method, using simple analytics, we
estimate the total efficiency of time-continuous, uninterrupted monitoring of de-excitations from |Bi to |Gi to be
ηeff,clk = 0.90 ± 0.01 for the parameters of our experiment, with integration time Tint = 0.26 µs. The simple analysis
of this section complements the numerical one of the previous section, Sec. III B 2.
Direct monitoring method in atomic physics. The direct method monitors for a |Bi de-excitation by collecting and
absorbing the photon radiated in the de-excitation. The total measurement efficiency of this method is limited by i)
collection efficiency — the fraction of emitted photons collected by the detector in its own input spatial modes (for
instance, as determined by the solid angle) — typically falls in the range 0.1 - 50%,26 ii) the efficiency of detecting
the absorption of a single photon, which falls in the range 1 - 90%,27 and iii) non-idealities of the photodetector
apparatus, including its dead time, dark counts, jitter, etc.27 The combination of these inefficiencies presents an
almost insurmountable challenge in experimental atomic physics for realizing continuous, time-resolved detection of
13
nearly every single photon emitted by the three-level atom, required to faithfully catch the jump.
Direct monitoring method with superconducting circuits. While technologically very different, the direct monitoring
method with superconducting circuits is conceptually similar to atomic method but can readily achieve high collection
efficiencies.28–49 However, the energy of the emitted microwave photon is exceedingly small — 23 µeV, about a part
per 100,000 of the energy of a single optical photon — which essentially forbids the direct detection of the photon with
near-unit efficiency. This is because the propagating photon is unavoidably subjected to significant loss, added spurious
noise, amplifier non-idealities, etc. In our experiment, these imperfections reduce the full measurement/amplification
chain efficiency from its ideal value32,38,49 of 1 to a modest η = 0.33 ± 0.03, corresponding to the direct detection of
approximately only one out of every three single photons — insufficient for the catch protocol.
1.
Indirect monitoring method with superconducting circuits
Alternatively, the indirect monitoring method couples the atom to an ancillary degree of freedom, which is itself
monitored in place of the atom. In our experiment, the atom is strongly, dispersively coupled to the ancillary readout
cavity. The cavity scatters a probe tone, whose phase shift constitutes the readout signal, as discussed in the Methods
section. Since the probe tone can carry itself many photons, this scheme increases the signal-to-noise ratio (SNR)
and, hence, the total efficiency (ηeff,clk ) of detecting a |Bi de-excitation. Note that the efficiency ηeff,clk should not
be confused with the efficiency of a photodetector or the efficiency η of the measurement/amplification chain, since
ηeff,clk includes the effect of all readout imperfections and non-idealities, state discrimination and assignment errors,
etc. see below. In the remainder of this section, we estimate the SNR and efficiency ηeff,clk of the experiment.
SNR of the indirect (dispersive) method. The output of the measurement and amplification chain monitoring the
readout cavity is proportional to the complex heterodyne measurement record ζ (t), which obeys the Itô stochastic
differential equation, see Eq. (34),50
dζ (t) =
√
ηκ
hψ (t) |â|ψ (t)i
dt + dZ (t) ,
hψ (t) |ψ (t)i
(52)
where â is the cavity amplitude operator in the Schrödinger picture, η is the total measurement efficiency of the
amplification chain — again, not to be confused with the de-excitation measurement efficiency, ηeff,clk — and dZ is
the complex Wiener process increment, defined below Eq. (52). A somewhat counterintuitive property of Eq. (52) is
that the heterodyne record increment dζ (t) is stochastic and noisy even when η = 1, the case of ideal measurement in
which no signal is lost — the stochastic term, dZ, represents pure quantum vacuum fluctuations, which are inherent
in the case of heterodyne detection.17,51,52 Due to the unavoidable presence of these fluctuations, only an infinitesimal
amount of information about the system can be extracted from dζ at an instant of time. Finite amount of information
is extracted by integrating dζ for a finite duration Tint ,
Z Tint
dζ (t) ,
(53)
s ≡ Irec + iQrec ≡
0
where Irec and Qrec are the in- and out-of-phase quadrature components of one segment of the record. What does s
correspond to? Its value depends on dζ, which depends on the state of the cavity, |ψi, which itself depends on the
occupation of |Bi — and therefore s contains the occupation of |Bi. A de-excitation of |Bi to |Gi can thus be detected
by monitoring s, whose value is different for the two states, since the cavity is generally in the coherent state |αB i
or |αG i when the atom is in |Bi or |Gi, respectively. For the moment, assuming the atom and cavity do not change
states during the course of the measurement duration Tint , the stochastic integral in Eq. (53) explicitly evaluates to
1
1
√
√
sB,G =
ηκRe [αB,G ] Tint + √ WI (Tint ) + i − ηκIm [αB,G ] Tint + √ WQ (Tint ) ,
(54)
2
2
where WI,Q denote independent Wiener processes, obeying the conventional rules, E [W (t)] = 0 and Var [W (t)] = t2 .
Equation (54) shows that the distribution of the stochastic variable s is a Gaussian blob in the IQ plane centered at
√
2
≡ Var [sB,G ] = 21 Tint . We can thus define
s̄B,G ≡ E [sB,G ] = ηγTint αB,G with width determined by the variance σB,G
the SNR of the experiment by comparing the distance between the two pointer distributions to their width,
SNR ≡
s̄B − s̄G
σB + σG
2
,
(55)
where the B (resp., G) subscript denotes signals conditioned on the atom being in |Bi (resp., |Gi). In terms of |αB i
and |αG i,
SNR =
1
2
ηκTint |αB − αG | ,
2
(56)
14
which can be expressed in terms of the parameters of the experiment, summarized in Table S1,
2
1
κ
SNR = ηκTint cos arctan
n̄ ,
2
2χBG
(57)
Holding other parameters fixed, according to Eq. (57), the SNR can be increased arbitrarily by increasing n̄, which can
be readily done by increasing the amplitude of the cavity probe tone. A higher SNR for s corresponds to a higher SNR
for measuring an atom de-excitation, since s is a proxy of the |Bi population. Thus, the indirect cavity monitoring can
overcome the typical degradation in SNR imposed by the inefficiencies and non-idealities of the measurement chain,
η. In practice, the SNR increase with n̄ is bounded from above, since with sufficiently high n̄ spurious non-linear
effects become significant8–12,53–58 . The cavity and non-linear coupling to the atom serve in effect as a rudimentary
embedded pre-amplifier at the site of the atom, which transduces with amplification the de-excitation signal before
its SNR is degraded during propagation and further processing.
Discrimination efficiency of the indirect method. While the SNR provides a basic characterization of the measurement, it is useful to convert it to a number between 0 and 1, which is called the discrimination efficiency, ηdisc . It
quantifies the degree to which the two Gaussian distributions of s are distinguishable,59
#
" r
1
SNR
,
(58)
ηdisc = erfc −
2
2
where erfc denotes the complementary error function. Equation (58) shows that increasing the SNR by separating the
sB and sG distributions far beyond their spread, σB/G , provides only marginal gain as ηdisc saturates to 1. Next, we
calculate the SNR and ηdisc for the parameters of the experiment and discuss corrections due to readout non-idealities.
A first comparison to the experiment. A first estimate of the SNR and ηdisc of the experiment are provided by
Eqs. (57) and (58). Using the parameters of the experiment, summarized in Table S1, from these two equations, we
find SNR = 4.3 ± 0.6 and ηdisc = 0.98 ± 0.007. Using data from the experiment, in particular, a second long IQ record
trace, represented by a short segment in Fig. 2a, we find the SNR of the jumps experiment, by fitting the histogram of
the trace with a bi-Gaussian distribution, to be SNR = 3.8 ± 0.4, corresponding to ηdisc = 0.96 ± 0.01. The measured
values are slightly lower than the analytics predict due to readout imperfections not included in the calculation so
far, such as state transitions during Tint , cavity transient dynamics, additional pointer-state distributions, etc.
Effective click detection efficiency. The dominant next-order error is due to atom state transitions during the
measurement window, Tint , which contributes an assignment error of approximately 1 − ηasg = 1 − exp (Tint /τB ) =
0.06 ± 0.001 to the detection of a |Bi de-excitation. Combining ηdisc with ηasg , we obtain the total efficiency for
detecting |Bi de-excitations ηeff,clk = ηdisc ηasg = 0.90 ± 0.01, consistent with the total readout efficiency of 0.91 that
is independently estimated using the trajectory numerics, see Sec. III B 2.
∗
1
2
3
4
5
6
7
8
9
10
zlatko.minev@aya.yale.edu | zlatko-minev.com
S. E. Nigg, H. Paik, B. Vlastakis, G. Kirchmair, S. Shankar, L. Frunzio, M. H. Devoret, R. J. Schoelkopf, and S. M. Girvin,
“Black-Box Superconducting Circuit Quantization,” Physical Review Letters 108, 240502 (2012).
Z. K. Minev et al., Energy-participation approach to the design of quantum Josephson circuits, in prep.
For related work on three-level superconducting systems, see Refs. 60–65.
J.-H. Yeh, J. LeFebvre, S. Premaratne, F. C. Wellstood, and B. S. Palmer, “Microwave attenuators for use with quantum
devices below 100 mK,” Journal of Applied Physics 121, 224501 (2017).
Z. Wang, S. Shankar, Z. Minev, P. Campagne-Ibarcq, A. Narla, and M. H. Devoret, “Cavity Attenuators for Superconducting
Qubits,” Physical Review Applied 11, 014031 (2019), arXiv:/arxiv.org/pdf/1807.04849 [https:].
D. H. Slichter, R. Vijay, S. J. Weber, S. Boutin, M. Boissonneault, J. M. Gambetta, A. Blais, and I. Siddiqi, “MeasurementInduced Qubit State Mixing in Circuit QED from Up-Converted Dephasing Noise,” Physical Review Letters 109, 153601
(2012).
D. H. Slichter, C. Müller, R. Vijay, S. J. Weber, A. Blais, and I. Siddiqi, “Quantum Zeno effect in the strong measurement
regime of circuit quantum electrodynamics,” New Journal of Physics 18, 53031 (2016).
D. Sank, Z. Chen, M. Khezri, J. Kelly, R. Barends, B. Campbell, Y. Chen, B. Chiaro, A. Dunsworth, A. Fowler, E. Jeffrey,
E. Lucero, A. Megrant, J. Mutus, M. Neeley, C. Neill, P. J. J. O’Malley, C. Quintana, P. Roushan, A. Vainsencher,
T. White, J. Wenner, A. N. Korotkov, and J. M. Martinis, “Measurement-Induced State Transitions in a Superconducting
Qubit: Beyond the Rotating Wave Approximation,” Physical Review Letters 117, 190503 (2016).
M. Boissonneault, J. M. Gambetta, and A. Blais, “Nonlinear dispersive regime of cavity QED: The dressed dephasing
model,” Physical Review A 77, 060305 (2008).
M. Boissonneault, J. M. Gambetta, and A. Blais, “Dispersive regime of circuit QED: Photon-dependent qubit dephasing
and relaxation rates,” Physical Review A 79, 013819 (2009).
15
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
L. Verney, R. Lescanne, M. H. Devoret, Z. Leghtas, and M. Mirrahimi, “Structural Instability of Driven Josephson Circuits
Prevented by an Inductive Shunt,” Physical Review Applied 11, 024003 (2019), arXiv:1805.07542.
R. Lescanne, L. Verney, Q. Ficheux, M. H. Devoret, B. Huard, M. Mirrahimi, and Z. Leghtas, “Escape of a Driven Quantum
Josephson Circuit into Unconfined States,” Physical Review Applied 11, 014030 (2019), arXiv:1805.05198.
W. Nagourney, J. Sandberg, and H. Dehmelt, “Shelved optical electron amplifier: Observation of quantum jumps,” Physical
Review Letters 56, 2797 (1986).
T. Sauter, W. Neuhauser, R. Blatt, and P. E. Toschek, “Observation of Quantum Jumps,” Physical Review Letters 57,
1696 (1986).
J. C. Bergquist, R. G. Hulet, W. M. Itano, and D. J. Wineland, “Observation of Quantum Jumps in a Single Atom,”
Physical Review Letters 57, 1699 (1986).
R. J. Cook and H. J. Kimble, “Possibility of Direct Observation of Quantum Jumps,” Physical Review Letters 54, 1023
(1985).
H. J. Carmichael, An Open Systems Approach to Quantum Optics (Springer, Berlin, Heidelberg, 1993).
J. Dalibard, Y. Castin, and K. Mølmer, “Wave-function approach to dissipative processes in quantum optics,” Physical
Review Letters 68, 580 (1992).
R. Dum, P. Zoller, and H. Ritsch, “Monte Carlo simulation of the atomic master equation for spontaneous emission,”
Physical Review A 45, 4879 (1992).
K. Macieszczak, M. Guţă, I. Lesanovsky, and J. P. Garrahan, “Towards a Theory of Metastability in Open Quantum
Dynamics,” Physical Review Letters 116, 240404 (2016).
M. Porrati and S. Putterman, “Wave-function collapse due to null measurements: The origin of intermittent atomic fluorescence,” Physical Review A 36, 929 (1987).
R. Ruskov, A. Mizel, and A. N. Korotkov, “Crossover of phase qubit dynamics in the presence of a negative-result weak
measurement,” Physical Review B 75, 220501 (2007).
R. Ruskov, V. V. Dobrovitski, and B. N. Harmon, “Manipulation of double-dot spin qubit by continuous noisy measurement,”
(2009), arXiv:0906.0425.
I. Lesanovsky, M. van Horssen, M. Guţă, and J. P. Garrahan, “Characterization of Dynamical Phase Transitions in Quantum
Jump Trajectories Beyond the Properties of the Stationary State,” Physical Review Letters 110, 150401 (2013).
J. P. Garrahan and M. Guţă, “Catching and reversing quantum jumps and thermodynamics of quantum trajectories,”
Physical Review A 98, 052137 (2018), arXiv:1808.00726.
J. Volz, R. Gehr, G. Dubois, J. Esteve, and J. Reichel, “Measurement of the internal state of a single atom without energy
exchange,” Nature 475, 210 (2011).
M. D. Eisaman, J. Fan, A. Migdall, and S. V. Polyakov, “Invited Review Article: Single-photon sources and detectors,”
Review of Scientific Instruments 82, 071101 (2011).
N. Katz, M. Neeley, M. Ansmann, R. C. Bialczak, M. Hofheinz, E. Lucero, A. O’Connell, H. Wang, A. N. Cleland, J. M.
Martinis, and A. N. Korotkov, “Reversal of the Weak Measurement of a Quantum State in a Superconducting Phase Qubit,”
Physical Review Letters 101, 200401 (2008).
R. Vijay, D. H. Slichter, and I. Siddiqi, “Observation of Quantum Jumps in a Superconducting Artificial Atom,” Physical
Review Letters 106, 110502 (2011), arXiv:1009.2969.
D. Ristè, C. C. Bultink, K. W. Lehnert, and L. DiCarlo, “Feedback Control of a Solid-State Qubit Using High-Fidelity
Projective Measurement,” Physical Review Letters 109, 240502 (2012).
R. Vijay, C. Macklin, D. H. Slichter, S. J. Weber, K. W. Murch, R. Naik, A. N. Korotkov, and I. Siddiqi, “Stabilizing Rabi
oscillations in a superconducting qubit using quantum feedback,” Nature 490, 77 (2012), arXiv:1205.5591.
M. Hatridge, S. Shankar, M. Mirrahimi, F. Schackert, K. Geerlings, T. Brecht, K. M. Sliwa, B. Abdo, L. Frunzio, S. M.
Girvin, R. J. Schoelkopf, and M. H. Devoret, “Quantum Back-Action of an Individual Variable-Strength Measurement,”
Science 339, 178 (2013).
K. W. Murch, S. J. Weber, C. Macklin, and I. Siddiqi, “Observing single quantum trajectories of a superconducting quantum
bit,” Nature 502, 211 (2013), arXiv:1305.7270.
G. De Lange, D. Riste, M. J. Tiggelman, C. Eichler, L. Tornberg, G. Johansson, A. Wallraff, R. N. Schouten, and L. Dicarlo,
“Reversing quantum trajectories with analog feedback,” Physical Review Letters 112, 080501 (2014), arXiv:1311.5472.
N. Roch, M. E. Schwartz, F. Motzoi, C. Macklin, R. Vijay, A. W. Eddins, A. N. Korotkov, K. B. Whaley, M. Sarovar, and
I. Siddiqi, “Observation of measurement-induced entanglement and quantum trajectories of remote superconducting qubits,”
Physical Review Letters 112, 170501 (2014), arXiv:1402.1868.
S. J. Weber, A. Chantasri, J. Dressel, A. N. Jordan, K. W. Murch, and I. Siddiqi, “Mapping the optimal route between two
quantum states,” Nature 511, 570 (2014), arXiv:1403.4992.
P. Campagne-Ibarcq, L. Bretheau, E. Flurin, A. Auffèves, F. Mallet, and B. Huard, “Observing interferences between past
and future quantum states in resonance fluorescence,” Physical Review Letters 112, 180402 (2014), arXiv:arXiv:1311.5605v1.
C. Macklin, K. O’Brien, D. Hover, M. E. Schwartz, V. Bolkhovsky, X. Zhang, W. D. Oliver, and I. Siddiqi, “A nearquantum-limited Josephson traveling-wave parametric amplifier,” Science 350, 307 (2015).
P. Campagne-Ibarcq, P. Six, L. Bretheau, A. Sarlette, M. Mirrahimi, P. Rouchon, and B. Huard, “Observing Quantum
State Diffusion by Heterodyne Detection of Fluorescence,” Physical Review X 6, 011002 (2016), arXiv:1511.01415.
P. Campagne-Ibarcq, S. Jezouin, N. Cottet, P. Six, L. Bretheau, F. Mallet, A. Sarlette, P. Rouchon, and B. Huard,
“Using Spontaneous Emission of a Qubit as a Resource for Feedback Control,” Physical Review Letters 117, 060502 (2016),
arXiv:1602.05479.
S. Hacohen-Gourgy, L. S. Martin, E. Flurin, V. V. Ramasesh, K. B. Whaley, and I. Siddiqi, “Quantum dynamics of
16
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
simultaneously measured non-commuting observables,” Nature 538, 491 (2016).
M. Naghiloo, N. Foroozani, D. Tan, A. Jadbabaie, and K. W. Murch, “Mapping quantum state dynamics in spontaneous
emission,” Nature Communications 7, 11527 (2016).
T. C. White, J. Y. Mutus, J. Dressel, J. Kelly, R. Barends, E. Jeffrey, D. Sank, A. Megrant, B. Campbell, Y. Chen, Z. Chen,
B. Chiaro, A. Dunsworth, I.-C. Hoi, C. Neill, P. J. J. O’Malley, P. Roushan, A. Vainsencher, J. Wenner, A. N. Korotkov, and
J. M. Martinis, “Preserving entanglement during weak measurement demonstrated with a violation of the Bell-Leggett-Garg
inequality,” npj Quantum Information 2, 15022 (2016).
Q. Ficheux, S. Jezouin, Z. Leghtas, and B. Huard, “Dynamics of a qubit while simultaneously monitoring its relaxation and
dephasing,” Nature Communications 9, 1926 (2018), arXiv:1711.01208.
M. Naghiloo, D. Tan, P. M. Harrington, J. J. Alonso, E. Lutz, A. Romito, and K. W. Murch, “Thermodynamics along
individual trajectories of a quantum bit,” (2017), arXiv:1703.05885.
D. Tan, N. Foroozani, M. Naghiloo, A. H. Kiilerich, K. Mølmer, and K. W. Murch, “Homodyne monitoring of postselected
decay,” Physical Review A 96, 022104 (2017), arXiv:1705.04287.
S. Hacohen-Gourgy, L. P. García-Pintos, L. S. Martin, J. Dressel, and I. Siddiqi, “Incoherent Qubit Control Using the
Quantum Zeno Effect,” Physical Review Letters 120, 020505 (2018), arXiv:1706.08577.
J. Heinsoo, C. K. Andersen, A. Remm, S. Krinner, T. Walter, Y. Salathé, S. Gasparinetti, J.-C. Besse, A. Potočnik,
A. Wallraff, and C. Eichler, “Rapid High-fidelity Multiplexed Readout of Superconducting Qubits,” Physical Review Applied
10, 034040 (2018), arXiv:1801.07904.
C. C. Bultink, B. Tarasinski, N. Haandbæk, S. Poletto, N. Haider, D. J. Michalak, A. Bruno, and L. DiCarlo, “General
method for extracting the quantum efficiency of dispersive qubit readout in circuit QED,” Applied Physics Letters 112,
092601 (2018).
Since the bandwidth of the measurement chain, κfilter , is significantly larger than that, κ, of the readout cavity, κfilter ≫ κ,
we can neglect the effect of κfilter for simplicity of discussion, see Eqs. (36) and (37).
M. B. Plenio and P. L. Knight, “The quantum-jump approach to dissipative dynamics in quantum optics,” Reviews of
Modern Physics 70, 101 (1998), arXiv:9702007 [quant-ph].
H. M. Wiseman and G. J. Milburn, Quantum Measurement and Control (Cambridge University Press, 2010).
Z. K. Minev, I. M. Pop, and M. H. Devoret, “Planar superconducting whispering gallery mode resonators,” Applied Physics
Letters 103, 142604 (2013), arXiv:arXiv:1308.1743v1.
M. Khezri, E. Mlinar, J. Dressel, and A. N. Korotkov, “Measuring a transmon qubit in circuit QED: Dressed squeezed
states,” Physical Review A 94, 012347 (2016).
C. C. Bultink, M. A. Rol, T. E. O’Brien, X. Fu, B. C. S. Dikken, C. Dickel, R. F. L. Vermeulen, J. C. de Sterke, A. Bruno,
R. N. Schouten, and L. DiCarlo, “Active Resonator Reset in the Nonlinear Dispersive Regime of Circuit QED,” Physical
Review Applied 6, 034008 (2016).
M. Khezri and A. N. Korotkov, “Hybrid phase-space–Fock-space approach to evolution of a driven nonlinear resonator,”
Physical Review A 96, 043839 (2017).
T. Walter, P. Kurpiers, S. Gasparinetti, P. Magnard, A. Potočnik, Y. Salathé, M. Pechal, M. Mondal, M. Oppliger, C. Eichler,
and A. Wallraff, “Rapid High-Fidelity Single-Shot Dispersive Readout of Superconducting Qubits,” Physical Review Applied
7, 054020 (2017).
K. Serniak, M. Hays, G. de Lange, S. Diamond, S. Shankar, L. D. Burkhart, L. Frunzio, M. Houzet, and M. H. Devoret,
“Hot Nonequilibrium Quasiparticles in Transmon Qubits,” Physical Review Letters 121, 157701 (2018), arXiv:1803.00476.
J. Gambetta, W. A. Braff, A. Wallraff, S. M. Girvin, and R. J. Schoelkopf, “Protocols for optimal readout of qubits using
a continuous quantum nondemolition measurement,” Physical Review A 76, 012325 (2007).
J. M. Gambetta, A. A. Houck, and A. Blais, “Superconducting Qubit with Purcell Protection and Tunable Coupling,”
Physical Review Letters 106, 030502 (2011), arXiv:1009.4470.
S. J. Srinivasan, A. J. Hoffman, J. M. Gambetta, and A. A. Houck, “Tunable Coupling in Circuit Quantum Electrodynamics
Using a Superconducting Charge Qubit with a V-Shaped Energy Level Diagram,” Physical Review Letters 106, 083601
(2011), arXiv:1011.4317.
I. Diniz, E. Dumur, O. Buisson, and A. Auffèves, “Ultrafast quantum nondemolition measurements based on a diamondshaped artificial atom,” Physical Review A 87, 033837 (2013), arXiv:1302.3847.
É. Dumur, B. Küng, A. K. Feofanov, T. Weissl, N. Roch, C. Naud, W. Guichard, and O. Buisson, “V-shaped superconducting
artificial atom based on two inductively coupled transmons,” Physical Review B 92, 020515 (2015), arXiv:1501.04892.
T. Roy, S. Kundu, M. Chand, S. Hazra, N. Nehra, R. Cosmic, A. Ranadive, M. P. Patankar, K. Damle, and R. Vijay,
“Implementation of Pairwise Longitudinal Coupling in a Three-Qubit Superconducting Circuit,” Physical Review Applied
7, 054025 (2017), arXiv:1610.07915.
G. Zhang, Y. Liu, J. J. Raftery, and A. A. Houck, “Suppression of photon shot noise dephasing in a tunable coupling
superconducting qubit,” npj Quantum Information 3, 1 (2017).