Chiral cat code: Enhanced error correction induced by higher-order nonlinearities

Adrià Labay-Mora,1, ∗ Alberto Mercurio,2, 3 Vincenzo Savona,2, 3 Gian Luca Giorgi,1 and Fabrizio Minganti2, 4, †
1
Institute for Cross-Disciplinary Physics and Complex Systems (IFISC) UIB-CSIC,
Campus Universitat Illes Balears, 07122 Palma de Mallorca, Spain
2
Institute of Physics, Ecole Polytechnique Fédérale de Lausanne (EPFL), CH-1015 Lausanne, Switzerland
3
Center for Quantum Science and Engineering, Ecole Polytechnique
Fédérale de Lausanne (EPFL), CH-1015 Lausanne, Switzerland
4
Alice & Bob, 53 Boulevard du Général Martial Valin, 75015, Paris, France
(Dated: March 17, 2025)
We introduce a Schrödinger chiral cat qubit, a novel bosonic quantum code generalizing Kerr
cat qubits that exploits higher-order nonlinearities. Compared to a standard Kerr cat, the chiral
cat qubit allows additional correction of bit-flip errors within the Hilbert space of a single bosonic
arXiv:2503.11624v1 [quant-ph] 14 Mar 2025

oscillator. Indeed, this code displays optical bistability, i.e., the simultaneous presence of multiple
long-lived states. Two of them define the code space and two define an error space. Thanks to the
chiral structure of the phase space of this system, the error space can be engineered to “capture” bit
flip events in the code space (a bit-flip trap), without affecting the quantum information stored in the
system. Therefore, it is possible to perform detection and correction of errors. We demonstrate how
this topological effect can be particularly efficient in the presence of large dephasing. We provide
concrete examples of the performance of the code and show the possibility of applying quantum
operations rapidly and efficiently. Beyond the interest in this single technological application, our
work demonstrates how the topology of phase space can enhance the performance of bosonic codes.

I. INTRODUCTION Bosonic quantum codes (BQCs) offer a promising sim-

plification of this complex challenge [15–22]. Rather than
Fault-tolerant quantum computing relies on the devel- storing information in the large Hilbert space of multiple
opment of efficient error-correction protocols, creating physical qubits, BQCs achieve redundancy in the larger
logical qubits through redundant encoding of informa- Hilbert spaces of a single bosonic oscillator. A key exam-
tion [1–3], protecting the system from the detrimental ple are Schrödinger cat qubits that use coherent states
effect of the environment [4–7]. A mainstream approach of opposite phase as logical codewords, with photonic
achieves redundancy using multiple physical qubits (two- Schödinger cats representing the logical x states. When
level systems). Upon successful completion of an error successfully operated, cat qubits are noise biased as they
correction protocol, the decay rate of the quantum infor- efficiently suppress bit-flip errors while still being sus-
mation stored in the logical qubit is lower than that of ceptible to phase-flip ones [16, 23, 24]. The size of the
the original components. Superconducting circuits are cat (i.e., the number of photons) is the key parameter
candidates for the development of error-corrected quan- determining the efficiency of the encoding. Larger cats
tum hardware [8–13]. The two main mechanisms inducing are expected to exponentially reduce the rate of bit-flip
errors are dephasing and photon loss noise, which combine errors at the price of a linear increase in the rate of phase
to generate bit- and phase-flips. Photon loss is due to flips. A possible strategy then relies on the preparation
the coupling of the system to spurious electromagnetic of sufficiently large and controlled cat states, almost im-
degrees of freedom, inducing a loss of energy of the res- mune to bit-flip errors and requiring only the correction
onator. Instead, dephasing emerges when coupling the of phase flips [25–27]. This task can be achieved using
device to nonlinear elements [14]. Given the rapid decay concatenated cats in a repetition code, significantly reduc-
of quantum information even in state-of-the-art devices, ing the hardware footprint with respect to a surface code
error correction requires the fabrication and control of by requiring fewer physical qubits per logical qubit, as re-
many physical qubits. The simultaneous correction of cently experimentally demonstrated [13]. More advanced
bit- and phase-flip errors was recently achieved by real- error correction schemes have also been recently proposed
izing a surface code, a highly controllable 2D geometry [28]. However, coupling multiple nonlinear elements can
of qubits encoding the logical state in tens of data qubits lead to an increase in dephasing noise, as discussed for
and ancillas [12]. Ultimately, realizing logical operation instance in Ref. [13].
between error-corrected qubits would require connecting In superconducting architectures operating cat states
and controlling many components, thus posing a major rely on an effective two-photon driving mechanism, to-
challenge towards the scaling-up of these devices. gether with two-photon dissipation (dissipative cat [16, 29–
32]), Kerr nonlinearity (Kerr cat [33–38]), or a combina-
tion of the two (hybrid cat [39–41]) that stabilize the
logical manifold spanned by coherent states. This task
∗ alabay@[Link] requires nonlinear and parametric processes. The for-
† [Link]@[Link] mer can be tailored by combining Josephson junctions,
 2

resulting in an anharmonic structure of the energy level

of the resonator; within a first approximation, the po- Code Space
tential of these nonlinear elements can be described as a
photon-photon interaction term, known as Kerr nonlin-
earity. Parametric processes result from applying external
drives to nonlinear elements. For example, two-photon
pump and two-photon dissipation mechanisms can be
generated by engineering a two-photon exchange between
the memory, hosting the cat state, and a very dissipative
mode (the buffer) [29, 42, 43].
What differentiates Kerr, dissipative, and hybrid cats
is their operations (gates) and response to errors, with
dephasing and photon loss degrading the quantum in- Error Space
formation stored in the logical qubit. Pioneering works
on cat states, both Kerr and dissipative, demonstrated
their possibilities and value for quantum information tasks
[29, 36]. However, Kerr cats are very susceptible to de-
phasing, as discussed below. More recently, several theo-
retical proposals have explored how additional features,
such as squeezing [44–47], Hamiltonian terms [40, 41],
or engineered dissipation [39, 48], can provide additional
protection from these dissipative events, with experimen-
tal results supporting these claims [49]. However, these
models are often idealizations of actual devices, where
the higher-order effects of the nonlinear elements are ne-
glected. FIG. 1. Bloch sphere representation of the code and error
In this article, we show that higher-order nonlinear spaces of a chiral cat, spanned by the low- (|±α⟩) and high-
processes can also be harnessed to improve the perfor- amplitude (|±αH ⟩) lobes, respectively. Upon an appropriate
choice of the system parameters, dephasing and photon loss
mance of the logical qubit, providing a pathway to protect
induce a passage from the low- to the high-amplitude lobes
Kerr cats from errors. Two facts that both depend on (a bit-flip trap that captures the state without degrading
the simultaneous presence of lower- and higher-order non- quantum information), making it possible to perform error
linearities permit this advantage: (i) The competition detection and correction.
between nonlinear terms can generate two metastable
manifolds, characterized by a different phase and photon
number, and in each of them quantum information can be Here, ∆ is the pump-to-cavity detuning, K2 is the
stored; (ii) Within an appropriate choice of parameters, Kerr term describing boson-boson interaction, and K3 is
the phase space of the bosonic system can be endowed the first higher-order correction describing three-body
with a chiral structure. Thus, the emergent structure processes. Higher-order nonlinearities, of the form
is one where two quasiorthogonal code and error spaces Kn (↠)n an , can also be included. This Hamiltonian typi-
coexist (see Fig. 1), with the error space capturing bit-flip cally emerges when developing the sinusoidal potential of
errors. As we show below, the presence of these features nonlinear superconducting elements based on Josephson
permits an error correction of bit-flip errors, significantly junctions, such as SQUIDs and SNAILs [51]. Normally,
reducing the error rates of Kerr cats, even in the presence K2 dominates over other nonlinearities, which are in many
of large dephasing. Our work shows that states with cases neglected.
Kerr and higher-order nonlinearities can still be used as Moreover, the system is subject to the action of the en-
logical qubits, and even outperform idealized Kerr cats. vironment, whose action can be described by the Lindblad
All the full quantum simulations were carried out using master equation:
the [Link] package in Julia [50].
Lρ = −i[Ĥ, ρ] + κ2 D[â2 ]ρ + κ1 D[â]ρ + κϕ D[↠â]ρ. (2)

II. MODEL AND WORKING PRINCIPLE Here, κ2 is the rate of two-photon dissipation, κ1 is the
single photon loss rate, and κϕ is the dephasing rate.
While κ1 and κϕ tend to be determined by the design and
The system we consider is a nonlinear bosonic resonator,
fabrication of the device, κ2 can be obtained by reservoir
driven by a two photon pump of amplitude ϵs , which in
engineering. Typically, a second mode characterized by
the frame of the pump reads (setting ℏ = 1)
large photon loss rate, called the buffer, is used to tune
K2 † 2 2 K3 † 3 3 two-photon dissipation [29]. Here and in the following,
Ĥ = −∆↠â + (â ) â + (â ) â + ϵs [(↠)2 + â2 ]. we will consider a Kerr-dominated confinement, assuming
2 3
(1) |K2 | ≫ κ2 > 0. Furthermore, we will consider configura-
 3

Dissipative/Kerr Critical Chiral

6
(a) 10 (b) (c)
5 30
Mean field
4
Quantum
20
⟨n ̂⟩

3
5
2
10
1

0 0 0
0 50 100 150 200 0 50 100 150 200 0 50 100 150 200
𝜖 s (MHz) 𝜖 s (MHz) 𝜖 s (MHz)

FIG. 2. Schematics of various types of cat states. The panels show the mean photon number as a function of the two-photon
drive amplitude, while the insets show a sketch of the Wigner function, with the arrows indicating the error processes. Solid lines
represent the steady-state photon number, while the dashed line represents the prediction of the semiclassical approximation.
(a) For a pure dissipative [29]/Kerr [36] cat, increasing the two-photon drive linearly increases the photon number. For a given
drive, the system spans a manifold characterized by two states of opposite phase ρ±α ≃ |±α⟩⟨±α|. Bit flip errors than take the
form of a passage between these two states. (b) For a critical cat [41] (a hybrid cat is simultaneously stabilized by Kerr and
two-photon loss and operated in the presence of detuning), the system displays optical bistability, i.e., the simultaneous presence
of multiple solution according to the semiclassical approximation. The optimal regime of operation is one where a cat state will
eventually decay into the vacuum, and quantum information is encoded in the metastable manifold spanned by squeezed states
of opposite phases. The dominant source of errors is a leakage, corresponding to a jump to the vacuum (ρ±α → |0⟩⟨0|). (c) The
chiral cat is a hybrid cat with additional higher-order nonlinearity. The manifold ρ±α ≃ |±α⟩⟨±α| coexists with a larger-photon
number one at ρ±αH ≃ |±αH ⟩⟨±αH |, both capable of hosting cat-like states. The high manifold acts as a bit-flip trap where the
jump ρα ↔ ρ−α is suppressed in favour of ρ±α → ρ±αH . Finally, as the photon number between the two manifolds is different,
the error can be detected and corrected. Parameters (MHz): κ1 /2π = 0.01, K2 /2π = −10, κϕ = 0 and incrementally we set (a)
∆ = K3 = 0, (b) ∆/2π = −50 and κ2 /|K2 | = 0.01, and (c) K3 /2π = 0.3.

tions with large κϕ , imagining concatenated architectures ∆ ̸= 0 and κϕ ̸= 0 the system will still host two steady
of highly-nonlinear and driven elements, where dephasing states, but no steady coherences. In this idealized picture,
is expected to be further increased [13] with respect to we can define the “standard” representation of cat states
that of a single component. on the Bloch sphere where we set the logical |±z ⟩ = |±α⟩,
while |±x ⟩ = |Cα± ⟩ (see the definition of the code space
in Fig. 1).
A. Kerr, dissipative, and critical cats Even if the cat properties as a memory do not change
between dissipative and Kerr cats, different gates have
In this part, we will review the operating principles of to be tailored for each configuration. Furthermore, the
ideal Kerr and dissipative cats by neglecting the effect of presence of additional Hamiltonian or dissipative terms
high-order nonlinearities (K3 = 0). can induce errors during the performance of logical oper-
ations. For example, the Z gate requires the action of a
one-photon drive of the form ϵz (â+↠). This term induces
1. Cats in the absence of photon loss and dephasing leakage outside of the cat manifold, whose magnitude de-
pends on whether cats are generated by a Hamiltonian
In this configuration, both dissipative, Kerr, and hybrid or a dissipative confinement. In particular, one can show
cats display similar properties, and the combined action that Kerr nonlinearity makes the cat more resilient to ϵz ,
of ϵs , K2 , and κ2 makes it possible to stabilize cat states allowing for stronger drives and thus faster gates than a
if ∆ = κ1 = κϕ = 0. The even and odd cat states are dissipative cat [39].
defined by |Cα± ⟩ ∝ |α⟩ ± |−α⟩, with |α⟩ a coherent state
such that â |α⟩ = α |α⟩. Any combination of cat states is
also stationary and the system admits two steady states 2. The effect of photon loss and dephasing
of the form ρss = |Cα± ⟩⟨Cα±p | and two steady coherences
ρsc = |Cα± ⟩⟨Cα∓ | with α = −ϵs /(K2 − iκ2 ) [52]. Thus, This picture drastically changes in the presence of
the entire manifold spanned by cat (and coherent) states photon loss and dephasing that degrade the quantum
is stable, generating what is called a decoherence-free information stored in the cat. A single steady state
subspace [21, 53]. Finally, in this configuration, but with ρss ≃ (|Cα+ ⟩⟨Cα+ | + |Cα− ⟩⟨Cα− |)/2 characterizes the system
 4

of bit-flip errors as in the dissipative case. Nevertheless,

𝜅𝜙 = 10 − 4𝜅 1 the Kerr cat shows protection from bit-flip errors with a
10 −4 (−1)
“staircase”-like behavior of λ0 , observed and discussed
− Re(𝜆(0 − 1))/𝜅 1

in, e.g., [56].

10 −6 The error suppression capability of cat states worsens
in the presence of dominating dephasing (see Fig. 4). For
10 −8 Kerr dissipative cats in Fig. 4(a), the suppression of bit-flip
Dissipative errors is still exponential in the photon number, even
if not as efficient as in the photon-loss dominated case.
10 −10 Nonetheless, dissipative cats remain a promising platform
2 6 10 14 18
Mean photon number
for biased-noise qubit repetition codes, and candidates
for low-hardware footprint error correction strategies. For
FIG. 3. The Liouvillian gap describing the bit-flip error rate as
Kerr cats [Fig. 4(b)], instead, the state loses its capability
a function of the mean photon number (|α|2 ) for the dissipative to suppress bit-flip errors, and thus it does not show all
(orange) and Kerr (blue) cats in a regime with negligible the beneficial properties that would make it a biased-noise
dephasing. Parameters as in Fig. 2 qubit. This can be a major challenge towards the con-
. catenation of multiple Kerr cats, as dephasing is expected
to emerge in nonlinear coupled systems.
[54, 55]. In the Bloch sphere, any point on the surface
will decay towards the center, and dissipation will even- 3. Phase space structure
tually lead to a completely mixed logical state. We can
understand this mixture as the simultaneous occurrence A useful tool to understand errors and their effects is to
of bit-flip errors |α⟩ → |−α⟩ [sketched in Fig. 2(a)] and use a combination of quantum trajectories (see details in
phase-flip errors |Cα± ⟩ → |Cα∓ ⟩. Here, |α|2 still grows the Appendix C) and semiclassical analysis. Quantum tra-
linearly with the amplitude of the drive. Cats are then jectories, unlike master equation approaches that directly
predicated on the idea that, increasing their size, i.e., their evolve the system’s density matrix, describe individual
photon number, bit-flip errors should be exponentially wave functions |ψ(t)⟩ through sequences of deterministic
suppressed (coherent states are long-lived). In contrast, and probabilistic events. For example, consider a state
the rate of phase-flip errors increases linearly with the initialized in |α⟩. As discussed in the Appendix C, around
cat size. The advantageous scaling of bit- over phase-flip this initial state, dissipation and Hamiltonian terms act as
errors makes cat states biased-noise qubits. noise displacing an initial coherent state, with dephasing
In all configurations, and even for the chiral cat in- inducing large fluctuations in the form of temperature-like
troduced in the following, it can be shown that the rate jumps, whose rate scales as κϕ |α|.
of phase-flip errors scales as κ1 ⟨↠â⟩. To determine the To determine the effect of dephasing, we can then use a
bit-flip rate, it is practical to resort to the Liouvillian semiclassical approximation. The semiclassical (coherent
eigenspectrum. Indeed, the Liouvillian is a linear superop- state) approximation neglects quantum fluctuations, as-
erator, that can be diagonalized to retrieve the eigenvalues suming ρ(t) = |α(t)⟩⟨α(t)| with |α(t)⟩ the coherent state.
λj and the eigenoperators ρj , defined by Lρj = λj ρj . The The evolution of the system, where we neglect dephasing,
model is further characterized by a Z2 weak symmetry is captured by the C-number equation
(an invariance of the equation of motion [53]) as â → −â κ 
leaves Eq. (2) unchanged. This, in turn, implies that, for 1 2
α̇(t) = − − κ2 |α(t)| α(t)
each eigenoperator, Π̂ρj Π̂ = ±ρj , with Π̂ = exp iπ↠â .

2 (4)
2
We can thus associate the quantum number k = ±1 to − i2ϵs α∗ (t) − iK2 |α(t)| α(t)
each eigenvalue and eigenoperator, to get
We then study the vector field indicating α̇(t), and the
(k)
Lρj
(k) (k)
= λ j ρj , k = ±1. (3) direction of the evolution of a point in phase space. Similar
analysis has been used in, e.g., Refs. [57–60] to study chaos
In particular, the Liouvillian eigenvalue λ0 describes
(−1) and topological effects in driven two-photon systems. We
the bit-flip error rate [41], where we have ordered the eigen- now have to compare the orbits of this system with the
(k) (k) thermal-like noise induced by dephasing.
values by their real part [| Re(λ0 )| = 0 < | Re(λ1 )| <
(k)
In a dissipative cat, shown in Fig. 5(a), the vector field
| Re(λ2 )|]. will always point back to one of the coherent states. For
(−1)
First, we study the bit-flip error rates plotting λ0 in the cat to display a bit-flip error, large fluctuation are
Fig. 3 with moderate values of dephasing. For dissipative required, making the jump improbable. The situation
cats, the bit-flip error rate becomes exponentially smaller is drastically different in a Kerr cat, where the system
with the number of photons. For Kerr cats, despite an ini- forms periodic orbits around the minima of the potential.
tial suppression of the bit-flip rate, increasing the photon This is shown in Fig. 5(b). Small fluctuations can thus
number does not lead to the same exponential suppression accumulate, eventually leading to bit-flip errors.
 5

(a) Dissipative (b) Kerr (c) Hybrid (Δ = 0) (d) Critical

𝜅𝜙 = 10 𝜅 1
10 0
− Re(𝜆(0 − 1))/𝜅 1

𝜅𝜙 = 𝜅 1
𝜅𝜙 = 0.1 𝜅 1

10 −5

10 −10
2 6 10 14 2 6 10 14 2 6 10 14 2 6 10 14
Mean photon number Mean photon number Mean photon number Mean photon number

FIG. 4. As a function of the mean-photon number, the Liouvillian gap describing the bit-flip error rate in (a) dissipative, (b)
Kerr, (c) hybrid and (d) critical cat. We show three different regimes: single-photon dominated at κϕ = 0.1κ1 (solid lines),
equal error rate κϕ = κ1 (dashed lines) and dephasing dominated κϕ = 10κ1 (dotted lines). Parameters as in Fig. 2. Detuning is
zero for dissipative, Kerr and hybrid cats, but an optimal detuning is chosen for the critical cat.

4. Hybrid and critical cats these solutions become metastable. The steady state ac-
cording to full quantum solution will display an abrupt
Hybrid cats use both Kerr and dissipative confinement. change between the two states around a critical point,
Considering a hybrid configuration where |K2 | ≫ κ2 > κϕ , connected to the presence of a dissipative phase transition
but with ∆ = 0, makes the cat partially resilient to [43, 61–63]. Around the critical point, the system will
dephasing, suppressing bit flip errors by increasing the switch between the vacuum and the populated phase at
cat size [39]. This is shown in Fig. 4(c). However, this random times. Instead, on either side of the transition,
configuration is still severely limited in its bit-flip error instead, the system will remain for long time in one of
correction capability when compared to a fully dissipative the metastable manifolds to then irreversibly decay to
cat. Several works then noticed that detuning can be the steady state [64, 65]. There is a profound connection
a resource to enhance the performance of a hybrid cat between second-order phase transitions and quantum in-
[40, 41, 56]. Indeed, an appropriate choice of detuning formation encoding [66–68], that we further extend here
suppresses bit-flip errors, as shown in Fig. 4(d). This to first-order ones.
feature allows for accessing specific spectral degeneracies
in cat states [39, 56] and enables better error correction A cat operated in this latter regime, called a critical
thanks to the introduction of squeezing [44]. cat [41], can thus still preserve quantum information on
Detuning can trigger bistability. Indeed, it is possible long timescales, even if the system eventually reaches the
to find regions of the parameter space where the minimal vacuum. Operating this critical cat in the metastable
bit-flip rate occurs when the system is bistable, with regime requires some specific procedures [41]. When
a vacuum-like steady state [see Fig. 2(b)]. In driven- comparing the performance of the critical cat with an
dissipative nonlinear systems, bistability occurs when hybrid cat at ∆ = 0, we see an improvement not only in
the system admits multiple long-lived (almost-stationary) the presence of one-photon dissipation, but also in the
states, each characterized by a different photon number. presence of dephasing as shown in Fig. 4. As the hybrid
Its emergence is linked to a balance between the various cat was already outperforming the Kerr cat, we conclude
parameters of the system: on the one hand, detuning that the critical cat selects the optimal parameters to
and nonlinearity determine the energy of a photon, while encode quantum information in a Kerr dominated cat.
drive and dissipation compete to populate or empty it. Finally, we observe that the system is again capable of
To understand this phenomenon, one can again resort to exponentially suppressing errors, making it viable even
a semiclassical approximation that in this case reads in the presence of large dephasing.
κ 
1 2
α̇(t) = − − κ2 |α(t)| α(t)
2   (5) When investigating the phase-space structure of the
2
− i2ϵs α∗ (t) − iα(t) −∆ + K2 |α(t)| critical cat, as done in Fig. 5(c) we observe how it adds a
layer of complexity compared to dissipative and Kerr cats,
This nonlinear equation admits, thanks to the addition with detuning deforming the shape of the orbits, thus
of detuning, multiple solutions for the steady-state equa- requiring larger fluctuations for the system to show bit
tion α̇ = 0 for an appropriate choice of parameters. A flips. In the limit of large detuning where the critical cat
stability analysis then finds that both the vacuum α = 0 is operated, however, the vacuum becomes the attractor
and a high-photon manifold |α| ̸= 0 can coexist [see of the dynamics, implying that, eventually, a system will
Fig. 2(b)]. Once quantum fluctuations are re-introduced, decay to it.
 6

Dissipative Kerr/Hybrid Critical Chiral

Fast
(a) (b) (c) (d)
5

|d𝛼/dt|
Im(𝛼)

−5

Slow
−5 0 5 −5 0 5 −5 0 5 −5 0 5
Re(𝛼) Re(𝛼) Re(𝛼) Re(𝛼)

FIG. 5. Phase-space representation of the manifold for (a) dissipative, (b) Kerr or hybrid, (c) critical, and (d) chiral cat states,
and corresponding vector field obtained by the semiclassical analysis. While for a standard dissipative, Kerr or hybrid cat errors
can bring the manifold out of its bottom of the potential, in the chiral case the system is attracted to the high-photon manifold.
Parameters (MHz) as in Fig. 2 with κϕ = 10−4 κ1 . Detuning and two-photon driving are chosen as to keep the code space with
eight photons: (a) ϵs /2π = 40, (b) ϵs /2π = 40, (c) ∆/2π = −64 and ϵs /2π = 7.2, and (d) ∆/2π = −8 and ϵs /2π = 26.4.

B. The chiral cat state |−α⟩. We conclude that the manifold spanned by
|±αH ⟩ acts as a trap for bit flip events, and one can
Let us now include K3 = ̸ 0. In the idealized cat config- engineer a right-bottom/left-top decay structure in phase
uration, where κ1 = κϕ = ∆ = 0 also this term prevents space, where the transitions are mainly |+α⟩ → |+αH ⟩
the steady state from defining a decoherence-free subspace. and |−α⟩ → |−αH ⟩, thus inducing a chiral structure.
However, in the presence of photon loss and dephasing This emergent chirality allows error correction to be
the inclusion of K3 can improve the cat’s performance, in applied. Indeed, calling |±α⟩ the code space and |±αH ⟩
analogy to what discussed for detuning in the critical cat. the error space (as in Fig. 1), one can determine whether
As the competition between ∆ and K2 generates bista- the system is in the code or error space without acquiring
bility, so does that between K2 and K3 . This can be any information on the phase of the field, as these states
straightforwardly argued from the equation of motion of have different photon number. This makes it possible to
the mean field, which now reads [63] detect errors and correct them, sending back |±αH ⟩ →
κ  |±α⟩.
1 2 Coarse-graining the dynamics of the quantum system
α̇(t) = − − κ2 |α| α − i2ϵs α∗
2  (6) so that dephasing and photon loss errors accumulate to
2 4 determine an overall bit- or phase-flip errors, we can
− iα −∆ + K2 |α| + K3 |α| .
visualize the process leading to an error in the logical
We show the solution to this equation in Fig. 2(c). Here, if space, as done in Fig. 6. For a logical error to occur,
the relative sign of K2 and K3 is different, two metastable one needs both |±α⟩ → |±αH ⟩ and |±αH ⟩ → |∓αH ⟩.
manifolds emerge: one spanned by low-photon states We can say that this error correction protocol allows
ρ±α = |±α⟩⟨±α|, and one by high-photon number ones the error correction rate of the large cat |±αH ⟩, while
ρ±αH = |±αH ⟩⟨±αH |. Compared to the bistability be- maintaining the phase flip error rate of the small encoding
tween ∆ and K2 , where the two metastable states are the |±α⟩. Further error correction capability comes from the
cat manifold and the vacuum, here both manifolds can squeezed nature of the states in |±αH ⟩. This code thus
encode quantum information, and define cat states and permits an efficient correction of bit-flip errors in Kerr
qubits. cat even in the presence of large dephasing. The rest of
Suppose now that we take a state initialized in |+α⟩. the paper is dedicated to rigorously demonstrating this
In a generic bistable configuration this state can jump statement.
to |−α⟩, generating bit flip errors, but also explore the
states |±αH ⟩. However, by choosing ∆ < 0, K2 < 0,
and K3 > 0, it is possible to suppress some of these III. PERFORMANCE OF THE CHIRAL CAT
transitions. The mechanism underlying this behavior
can be understood again using the semiclassical analysis To determine the optimal regime of operation for the
shown in Fig. 5(d). Any perturbation that takes the cat chiral cat qubit, we first assume that the system admits
qubit out of the |+α⟩ manifold will either bring it back coherent states as attractor of the dynamics. Namely,
to itself, or will mainly lead to a decay into |+αH ⟩. The the system is either in |±α⟩ or |±αH ⟩. We then imag-
opposite passage from |+α⟩ → |−αH ⟩ is far less probable, ine a protocol where, if the system is detected in |±αH ⟩,
as well as |+α⟩ → |−α⟩. Similar relations hold for the we project it back onto the corresponding |±α⟩; oth-
 7

erwise no operation is applied. Overall, the measure- 1.0

(a) |𝛼⟩ 𝜏|𝜆chiral| (b) |𝛼H⟩
ment and recovery procedure takes the form of the pro-
0.1

Projection
jector Π̂αH = |αH ⟩⟨αH | + |−αH ⟩⟨−αH | and recovery
R̂ = |α⟩⟨αH | + |−α⟩⟨−αH |, that combined give 10.0
0.5
1000.0
R̂ Π̂αH + (1̂ − Π̂αH ) = ∞
(7)
|α⟩ (⟨α| + ⟨αH |) + |−α⟩ (⟨−α| + ⟨−αH |) . 0.0
1.0
While this intuition is valid for the coherent-state ap- (c) | − 𝛼⟩ (d) | − 𝛼H⟩
proximation and if we assume at all times to be either in

Projection
|±α⟩⟨±α| or |±αH ⟩⟨±αH |, the mixed nature of the encod-
ing and the fact that the system will also explore other 0.5
states require being more mathematically careful. To gen-
eralize this idea to the actual regime of operation, we call
ρ±α the coherent states associated with the low-photon 0.0
number manifold, and ρ±αH that of the high-photon num- 10 0 10 1 10 2 10 3 10 4 10 0 10 1 10 2 10 3 10 4
ber states. Periodically, we detect whether the system Time (μs) Time (μs)
transitioned to the high-photon manifold. To do that, we
select the most probable states of ρ±αH and construct FIG. 7. Example evolution of a state |α⟩ with the recovery
the projector onto them, which we call Π̂αH [69]. If the operation applied at viarious recovery times τ . Without cor-
system is not in the ρ±αH manifold, we do not perform rection (dotted black with τ = ∞) the population on the left
lobe is driven towards the high-manifold and ends in an equal
classical mixture of |±αH ⟩. Similarly, if correction is too slow,
Code Space Error Space the recovery fails at preserving the code space. Instead, for
τ |λchiral | <
∼ 100, the code space is preserved and the decay
towards the steady state |α⟩⟨α| + |−α⟩⟨−α| is slowed down.
Parameters: κϕ = 2κ1 and |α|2 = 8 in Fig. 8. For this choice
of parameters, λchiral ≃ 10−2 .

any error correction. Otherwise, we act with an idealized

recovery operation R̂ sending each of the components of
the state ρ±αH to the most probable one in ρ±α . Namely,
the overall recovery procedure reads

ρR = Rρ = R̂ Π̂αH ρ Π̂†αH R̂† +(1̂− Π̂αH )ρ(1̂− Π̂†αH ). (8)

Time

To find the optimal working point for this algorithm,

we then search for the point where the system admits
Recovery the chiral structure with four modes. Then we check
that the passage |α⟩ → |αH ⟩ is slow, but still faster than
|α⟩ → |−α⟩. This translates into investigating three Li-
ouvillian eigenvalues: (i) λleak that describes the leakage
rate of the whole code space, spanned by the metastable
manifold ρ±α , into the error space spanned by ρ±αH ; (ii)
λchiral that describes how |±α⟩ → |±αH ⟩ occurs; (iii) λH
that describes the passage |±αH ⟩ → |∓αH ⟩. While this
restriction may provide a suboptimal solution to the prob-
lem, all these elements can be automatically constructed
using the eigendecomposition of the Liouvillian, and the
FIG. 6. Within a coarse-grained quantum trajectory picture, in automated search discussed in the Appendix B.
order for the system to display a logical error, two consecutive We then assume to monitor the system at a certain rate
processes must occur. First, the state has to pass from the code τ and we perform the recovery operation R at this fre-
space to the error space. Such an event, although improbable, quency, as shown in Fig. 7. Therefore, the time evolution
occurs on a timescale that is not the shortest one in the of the system is of the form
problem. Then, a second jump in the error manifold must
take place. This second jump is significantly more improbable. ρ(t + τ ) = ReLτ ρ(t). (9)
Error correction can then be performed by detecting if the
qubit jumped to the error manifold, and in that case by The optimal performance of the code is then found by
reverting to the code space. investigating the bit-flip error rate sending τ → 0. In
 8

(a) 𝜅𝜙 = 0.1𝜅 1 (b) 𝜅𝜙 = 𝜅 1 (c) 𝜅𝜙 = 10𝜅 1

10 −5.0
|𝜆error/H ̂gap|

10 −7.5

Critical
10 −10.0 Chiral

5 10 15 5 10 15 5 10 15
Mean photon number Mean photon number Mean photon number

FIG. 8. Results of the parameter optimization for both critical (blue) and chiral (green) cats for increasing values of dephasing:
(a) κϕ /κ1 = 0.1, (b) κϕ /κ1 = 1, and (c) κϕ /κ1 = 10. As a function of the mean photon number of the code space, we show the
ratio between the error rate (the effective error rate obtained by applying detection and recovery protocol to the chiral cat) with
respect to the Hamiltonian gap. For the chiral cat we find an exponential scaling of the form ∼ exp(−γn) where γ is 0.71, 0.54
and 0.48 for each value of κϕ respectively. This is attained at an almost constant Kerr nonlinearity of 45 · 2πMHz and K3 of the
order 2πMHz. Parameters as in Fig. 4, optimized over ∆, K2 and ϵs and fixed |κ2 /K2 | = 0.01.

this limit, we can define an effective error rate λerror 10 −0.5

(a)
describing the rate at which the bit flip |α⟩ → |−α⟩
occurs. In practice, we verified that taking λchiral τ ≪ 1 10 −1.0
does not significantly change this rate. 1−𝓕
We show the performance of this chiral cat in suppress- 10 −1.5
ing bit flip errors and compare it to a critical (i.e., an
optimal hybrid detuned) cat in Fig. 8. To provide a fair 10 −2.0
comparison, we compare the bit flip rate to the Hamil-
tonian gap between the ground and first excited state
10 −2.5
E1 − E0 , with Ej the Hamiltonian energies. This quantity
captures the speed at which Hamiltonian operations can 10 −2.0 10 −1.5 10 −1.0 10 −0.5 10 0.0
be performed. While in photon-loss dominated config- 𝜖z/𝜖 s
urations the chiral cat matches the performance of the 1 (b)
hybrid one (demonstrating that higher-order nonlinearies
are not necessarily detrimental), for dephasing dominated
systems, the chiral cat outperforms the critical cat.
Parity

0
Dissipative
Kerr
Gates Critical
Chiral
−1
Since the chiral cat behaves as a Kerr cat in the code 0.00 0.01 0.02 0.03
space, gates can be directly adapted from those of stan- Time (μs)
dard Kerr cats [70].
A gate that may be adversely affected by the presence FIG. 9. Performance of z gate in a dissipative (orange), Kerr
of K3 is the z gate, which has form (dark blue), critical (light blue), and chiral (green) config-
uration. (a) As a function of the ratio ϵz /ϵs , the error in
Ĥz = ϵz (â + ↠). (10) performing the z gate. (b) As a function of time, parity oscil-
lations at the optimal ϵz point. Parameters as in Fig. 5.
Indeed, the presence of the drive may induce transitions
between the low-and high manifold. This is not the case,
as we show in Fig. 9. We demonstrate the gate can be IV. IMPLEMENTATION AND
performed with an efficiency similar to that of the Kerr CONCATENATION
cat limit [Fig. 9(a)]. Compared to a dissipative cat, we see
that the operation can be performed much more rapidly, Having demonstrated the capability of the code in cor-
and with a significantly smaller error rate. We conclude recting for bit-flip events in an idealized scenario, here we
that the chiral cat maintains the capability of Kerr cat discuss how to actually perform their correction, while
to be rapidly operated [Fig. 9(b)]. simultaneously correcting for phase-flip ones. The basic
 9

strategy we propose, shown in Fig. 10 is to perform a A. Simultaneous detection of bit and phase flip
repetition code using nonlinear ancillas, as in the experi- errors
mental realization of a repetition code done in Ref. [13].
The role of the ancilla can be dual: they can be used We consider here a simplified architecture, where we
to perform the parity checks needed to detect phase flip assume the ancilla to perform a parity and photon-number
errors and to detect the |α⟩ → |αH ⟩ chiral errors. For readout without any errors. We then simulate a repetition
example, the latter could be checked using a dispersive code with 7 logical qubits. Upon detection of a bit flip,
readout protocol. We also stress that in configurations we apply the idealized recovery protocol described above.
where two-photon dissipation is achieved through an ad- Similarly, if a phase flip is detected, we correct for the
ditional buffer mode, the emission spectroscopy of the parity error by injecting a photon.
buffer also carries information on the state of the memory. In Fig. 10, we show a quantum trajectory for a system
initialized in an even cat, where several phase flip events
occur, as well as the jump from |α⟩ → |αH ⟩. As a jump in
parity [Fig. 10(a)] does not change the photon number, a
jump in photon number does not affect parity [Fig. 10(b)].
The Wigner functions, shown at various time along the
dynamics, clearly demonstrate the capability of the code
to store and preserve quantum information over long
times.
It should be noticed that the chiral error rate |α⟩ →
|αH ⟩ also affects the performance of the repetition code.
Indeed, the phase flip error rate is significantly larger
in the αH manifold and a chiral jump will effectively
reduce the distance of the error correction code, removing
one of the qubits from the repetition array. In practice,
calling pchiral the probability of a qubit to jump in the
time interval τ of one detection and recovery cycle, we
can model the overall phase-flip error rate as pphase−flip ≃
κ1 |α|2 τ (1 + pchiral |αH |2 /|α|2 ).

V. CONCLUSIONS

We introduced a chiral cat qubit, a logical qubit that

works in the presence of higher-order nonlinearities. We
show how these higher-order nonlinear processes can be
harnessed to improve the performance of the logical qubit,
providing a pathway to protect Kerr cats from dephasing.
The system we consider displays optical bistability, i.e.,
the simultaneous presence of multiple solutions according
to the semiclassical approximation. We show that these
states can form a logical code and error space, and we
demonstrate how the chiral structure acquired by the
FIG. 10. (a) Repetition code with cat data and ancillas phase space can be used for error correction. Our work
qubit. The latter are used both to perform parity measurement shows that the chiral cat has the desired error correction
(green boxes) and photon number discrimination measurement feature, while also maintaining the capability of Kerr cats
(orange boxes). Upon the detection of an error either in to be rapidly operated. At the same time, the chiral cat
parity or in photon number, the appropriate error correction is characterized by exponential error correction capabili-
protocol is applied. Along a single quantum trajectory and
ties in the size of the cat, even in the presence of large
as a function of time evolution of (b) parity and (c) photon
number. A jump in parity corresponds to a phase flip error
dephasing.
in one qubit, which can be corrected by the parity check Beyond the specific example investigated here, our work
performed by the repetition code. Parity jumps do not affect demonstrates how the rich landscape of parametric, criti-
the photon number. A jump in the photon number, instead, cal, and topological phenomena in open bosonic systems
corresponds to the |±α⟩ → |±αH ⟩ transition. As it does not can endow bosonic codes with error correction capabili-
yet correspond to a bit flip, this error can be detected and ties. We plan to investigate how chiral properties can be
corrected, as shown in the trajectory. Notice that this jump applied to different driving schemes, or investigate colored
does not directly entail a jump in the parity. Parameters as dissipation and autonomous stabilization possibilities for
in Fig. 7. chiral codes.
 10

ACKNOWLEDGMENTS (as in Fig. 4), but the chiral strucutre and its associated
eigenoperators have to be properly accounted between
We thank Filippo Ferrari, Luca Gravina, and Roberta the four slowest eigenmodes (see Appendix A). To exten-
Zambrini for insightful discussions. FM acknowledges sively search the parameter (see Eq. (2)), we employed
insightful discussion with team members at Alice & evolutionary minimization algorithms to determine the
Bob, in particular Giulio Campanaro, Joachim Co- parameter set that yields the optimal bit-flip time using
hen, André Melo, and Sébastien Jezouin. We acknowl- the correction method described in Section III. Such a
edge the Spanish State Research Agency, through the choice was mainly motivated by the non-differentiable
Marı́a de Maeztu project CEX2021-001164-M funded nature of the minimization function. Specifically, we
by MCIU/AEI/10.13039/501100011033, through the choose utilized the Evolutionary Centers Algorithm
COQUSY project PID2022-140506NB-C21 and -C22 from [Link] [71]. This algorithm leverages
funded by MCIU/AEI/10.13039/501100011033, and the genetic operations such as selection, crossover, and mu-
QuantERA QNet project PCI2024-153410 funded by tation to guide an initial population of Npop individuals
MCIU/AEI/10.13039/501100011033 and cofounded by toward the global minimum over Niter iterations [72]. The
the European Union; MINECO through the QUANTUM following sections detail the optimization process to ob-
SPAIN project, and EU through the RTRP - NextGener- tain the bit-flip time between |+α⟩ ↔ |−α⟩ as a function
2
ationEU within the framework of the Digital Spain 2025 of |α| .
Agenda. V.S. acknowledges support by the Swiss National
Science Foundation through Projects No. 200020 185015,
200020 215172, 200021-227992, and 20QU-1 215928.

Appendix A: Interpretation of the Liouvillian

eigenoperators

As we discussed in the main text, three Liouvillian eigen-

values and eigenoperators come into play when discussing 1. Search space
the chiral cat error: (i) λleak and the corresponding ρleak ;
(ii) λchiral and ρchiral ; and (iii) λH and ρH . To understand
why this is the case, and their role in determining the
qubit performance, in Fig. 11 we plot the various state The problem is characterised by seven parameters. We
that lead to the construction of the logical state ρ+α . assume three of them to be fixed: the one-photon dissi-
First, in Fig. 11(a) we plot the steady state of the system. pation (κ1 ), two-photon dissipation (κ2 ), and dephasing
This is of the form ρss ≃ (ρ+αH + ρ−αH )/2. To obtain the (κϕ ) rates. While various relative ratios of κ1 and κϕ are
state ρ+αH we need to add the eigenstate ρH , as shown explored, κ2 is chosen to be 1% of the Kerr interaction K2 .
in Fig. 11(b). Similarly, to pass from the error manifold The others, detuning (∆), two-photon drive (ϵs ), Kerr
to the centre of the Bloch sphere in the code manifold, (K2 ) and the first high-order correction (K3 ), are instead
we have ρα + ρ−α = ρss + ρleak , as shown in Fig. 11(c). free parameters. Note that the driving strength ϵs is not
Finally, the state ρα is obtained by summing all these a free parameter: as we fix the mean photon number of
2
processes, so that, as shown in Fig. 11(d), the code space (|α| ) we can use the mean field equation
(6) to it [63].
ρα = ρss + ρleak + ρH + ρchiral . (A1)

The time evolution of this state is of the form All parameters are bounded by experimentally realiz-
able values. While the Kerr nonlinearity can be tuned to
eLt ρα = ρss + eλleak t ρleak + eλH t ρH + eλchiral t ρchiral . (A2) tens of MHz [33, 36], for the higher-order correction we
have no clear literature comparison. As we do not want
Remarkably, here there is no direct bit-flip error be- to fix a particular superconducting element to operate the
tween ρα and ρ−α , but the passage to the high-photon chiral cat, we use the transmon model to bound K3 , and
manifold is always required. Finally, we remark that possible higher-order order corrections. The transmon
|λH | < |λleak | < |λchiral |. Hamiltonian is given by Ĥ = 4EC n̂2 − EJ cos ϕ̂, where
 1/4
ϕ̂ = 2E C
EJ (↠+ â) is the phase operator, EC the
Appendix B: Parameter optimization charge energy and EJ the Josephson energy [73, 74]. Ex-
panding the cosine term, and discarding with different
To realize a chiral cat, it is crucial to identify the op- number of creation andPannihilation operators, leads to
timal parameters, as chirality manifests within a narrow the Hamiltonian Ĥ = m Km â†m âm where the expres-
parameter regime. Moreover, the Liouvillian gap is no sions can be found by normal ordering the annihilation
longer the quantity that determines the bit flip error rate and creation operators, setting K1 = ∆. For instance, we
 11

Code Space Error Space

(a) |𝛼⟩ |𝛼H⟩

|C𝛼− ⟩ |C𝛼+ ⟩ |C𝛼− ⟩

H
|C𝛼+ ⟩
H

| − 𝛼⟩ | − 𝛼H⟩

(b) |𝛼⟩ |𝛼H⟩

+ = |C𝛼− ⟩ |C𝛼+ ⟩ |C𝛼− ⟩

H
|C𝛼+ ⟩
H

| − 𝛼⟩ | − 𝛼H⟩

(c) |𝛼⟩ |𝛼H⟩

+ = |C𝛼− ⟩ |C𝛼+ ⟩ |C𝛼− ⟩

H
|C𝛼+ ⟩
H

| − 𝛼⟩ | − 𝛼H⟩

(d) |𝛼⟩ |𝛼H⟩

+ + + = |C𝛼− ⟩ |C𝛼+ ⟩ |C𝛼− ⟩

H
|C𝛼+ ⟩
H

| − 𝛼⟩ | − 𝛼H⟩

FIG. 11. Construction of the logical states using the Liouvillian eigenvalues. On the left we plot the Wigner function of the
Liouvillian eigenoperators. In the center the resulting state. On the right, their representation in the error and code space. (a)
The steady state ρss . (b) The state associated with ραH → ρ−αH is ρH . Added to ρss gives ραH . (c) The leakage matrix ρleak
describes how the center of the Bloch sphere in the code space passes to the centre of the Bloch sphere in the error space. (d)
Adding the chiral decay into the mixture, we finally obtain ρα .

have [75] EC and EJ rather then K2 and K3 . Finally, the bounds

"  3/2 for the optimization are
EC 90 2EC
K2 ≈ − + EJ
2 6! EJ Parameter Min Max
 2  5/2 # ∆ (MHz/2π) −1000 0
1260 2EC 18900 2EC
− + (B1) EC (MHz/2π) 1 100
8! EJ 10! EJ EJ /EC 5 150
"  3/2
20 2EC
K3 ≈EJ
6! EJ
2. Minimization function
 2  5/2 #
560 2EC 12600 2EC
− + (B2)
8! EJ 10! EJ The target quantity to be minimized is the left-right
transition rate once the correction operation introduced
We note that the cosine potential already ensures an in Section III is used. However, chirality only occurs in
alternating sign between each term, needed to trigger a very small parameter regime of the entire search space.
multistability and chirality. Higher-order corrections are Therefore, for each parameter set, we need to identify
also significantly smaller than the K2 and K3 . Nonethe- if there is a chiral mode (see Appendix A), extract the
less, we include them in our optimization protocol to density matrices of both ρ±α and ρ±αH , construct the
demonstrate the robustness of chirality to the presence of recovery operation [Eq. (7)], and finally, determine the
higher-order terms. Notice that including these does not new transition rate. For this, we will use the fact that the
increase the search space, that still require determining eigenoperators describing the code and error spaces are
 12

metastable, resulting in a separation in the Liouvillian right eigenoperators. We then calculate the coefficient ma-
spectrum that distinguishes the four slowest eigenmodes trix [Ĉ]j,β = tr σj† ρβ , where β ∈ {+α, −α, αH , −αH }, and
from the rest. We also check that the vacuum in not P3
the corresponding projectors P̂β = j=0 (C −1 )β,j σj . The
the steady state state, ensuring that all dynamics are classical transition matrix between the metastable states
contained within the manifold spanned by {ρ±α , ρ±αH }. is given by V̂ = ĈΛĈ −1 , where Λ = diag(λ0 , . . . , λ4 ).
Below, we describe how this process is performed. This matrix describes the evolution of the jumps within
a. Detection of chiral eigenoperator The chiral mode the metastable manifold. To account for the correction,
is an antisymmetric operator which, if exist, lives in the we consider the total evolution as R ◦ L. This can be
second symmetry sector [k = −1 in Eq. (3)]. Moreover, evaluated as rγ→β = tr P̂β R(ργ ), which determines the
since the chiral mode should be long-lived, we can restrict probability that the lobe γ is found in β after the correc-
the search to only the two slowest eigenmodes. We know tion. Thus, the transition matrix with correction is given
that this mode is decomposed as ρchiral = c1 (ρα − ρ−α ) − by
c2 (ραH − ρ−αH ) where c1 and c2 are two proportionality Xh i
constants resulting from numerical diagonalization. A [V̂ ]α→β = Ĉ −1 ΛĈ rγ→β . (B4)
direct method would be to eigendecompose the operator γ,α
γ
and filter the eigenvectors depending on the position in
phase space. However, this leads to incorrect results, since Diagonalizing the symmetric matrix V̂ allows us to iden-
the high-photon manifold may not lie on the imaginary tify the eigenvector corresponding to the transitions be-
axis in the Wigner function. Hence, we used a different tween +α and −α. The associated eigenvalue λerror de-
approach which does not depend on the actual position termines the total bit-flip time of the lobes, such that
of the lobes. We calculate the Wigner function Wchiral of τerror = −1/λerror .
ρchiral which results in a matrix that is mostly zero except e. Fitness evaluation All the previous steps allow us
in around ρ±α and ρ±αH . Then, we calculate the contour to calculate the desired transition rate conditioned on the
lines for a given value ℓ of the normalized matrix |Wchiral |. existence of the chiral eigenoperator. To prevent the algo-
The chiral mode is identified if only four contour lines rithm from failing in the other cases, we use several checks
are found. We then check the relative sign the Wigner to stop the process as soon as possible and save compu-
function within those contour [c.f. Fig. 11(d)]. During tational resources. For instance, before diagonalising the
the optimization we fixed ℓ = 0.05. Liouvillian, we check whether the mean field solution pre-
b. Extraction of density matrices Given the four slow- dicts the existence of the four metastable states. Follow
est eigenmodes, we use the procedure outlined in Ap- up checks deal with the existence of metastability or the
pendix A to construct the four density matrices. detection of the chiral eigenoperator. At each halting step,
c. Construction of the recovery operation In gen- we also adapt the fitness function to guide the algorithm
eral, the low manifold is well approximated by the pure towards a regime where chirality might be present. A
coherent states |±α⟩ but this is not the case for the summary of this conditions and the corresponding fitness
high manifold. In order to construct the projection function can be seen in Table I.
and recovery operation we eigendecompose ρ±αH to ob- Finally, for those parameters where the algorithm is
tain {(λj±αH , vj±αH )}j where, due to the symmetry, able to evaluate τerror , we evaluate the following quantity
λαj
H
= λ−α
j
H
and the eigenvectors are equal up to a f (x) = −Ĥgap (τerror − τleak ) (B5)
rotation. Only a few eigenvalues are relevant in the de-
composition so we set a threshold value λth = 10−4 . where the first term corresponds to the Hamiltonian gap
Hence, the projection and recovery operators are Ĥgap = E1 − E0 > 0 and the last term is the difference
X between the bit-flip time (α ↔ −α) with correction and
Π̂αH = vj+αH vj+αH + vj−αH vj−αH (B3a) without correction. The latter can also be evaluated from
±αH the semiclassical transition matrix V̂ as explained above.
λj >λth
X Defining the minimizing function in this way implies that
R̂ = α vj+αH + −α vj−αH (B3b) the larger the Hamiltonian gap, the smaller will it be,
±α
λj H >λth
as desired. Similarly, this function allows to find those
parameters where the chiral correction is more favourable.
This guarantees that any component of a state in the high
manifold is taken to the low manifold.
3. Results
d. Calculation of transition rate To efficiently calcu-
late the transition rate, we employ the theory of classical
metastability [65], which allows us to describe the quan- We will perform the minimization forκ1 /2π = 10 kHz,
2
tum jumps as classical transitions. This approach enables κϕ /κ1 ∈ {0.1, 1, 10} and |α| = 2, . . . , 15. We also com-
us to map the full Hilbert space into a four-dimensional pare to the results with best known encoding so far, the
manifold spanned by the four lobes. We begin by deter- critical cat. For this, we repeat the simulations without
mining the left eigenoperators of the Liouvillian, defined EJ /EC and replace Eq. (B5) with the fitness function
(−1)
as L† σj = λ∗j σj , and sorting them in the same order as the that is just − Re(λ0 )/Ĥgap (as in Fig. 4).
 13

TABLE I. Minimization function used if the algorithm is not able to calculate the recovery time.
Halt reason Return value
Number of mean field solutions n is smaller than 5 1 × 1018 (1 + 5 − n)
Mean field mean photon number is different from target 1 × 1016 (1 + ||α|2 − ⟨n̂⟩target |)
(−1)
Chiral eigenoperator not found 1 × 1014 (1 + Re(λ1 ))
13
No metastability 1 × 10 (1 + |Re(λ4 )/ Re(λ5 )|)
Failed to extract all the metastable states 1 × 1012 (1 + |Re(λchiral )|)
Classical α → −α transition is not found 1 × 109 (1 + [V̂ ]α→−α )

The results are presented in Fig. 8 and the optimal with the environment, which in our case are the photon
√ √
parameters in Fig. 12. As can be seen in panel (a), the loss L̂1 = κ1 â, two-photon dissipation L̂2 = κ2 â2 ,
chiral correction allows to improve the bit flip error time √
and dephasing L̂3 = κϕ ↠â. Evolution under Ĥeff is
up to six orders of magnitude with respect to the bit-flip interrupted by quantum jumps, acting as L̂k |ψ(t)⟩, oc-
time in uncorrected nonlinear cats. In the other panels curring in an infinitesimal time dt by jump probabilities
(b-d), we show the optimal driving strength, detuning
pk = ⟨ψ(t)|L̂†k L̂k |ψ(t)⟩ dt. Thus, each stochastic trajec-
and Kerr, respectively. We clearly note that large mean
tory corresponds to a single realization of measurement
photon number requires larger drive and detuning, as
outcomes of the environment of an ideal measurement
expected. But, in all cases, the optimal values are smaller
instrument.
for the chiral compared to the critical cat. We also notice
In an undriven system, quantum jumps corresponding
how the Kerr nonlinearity remains almost constant for
to both one- and two-photon loss progressively collapse
all values of |α|2 , indicating the possibility of operating a
the superposition states towards the vacuum. Dephasing,
single device in the chiral regime by simply detuning and
instead, reduces coherence between components without
increasing photon drive.
energy loss. This is not the case in driven cat states.
As for the Kerr, we note that the critical cat uses
2 To understand the effect of each term, we resort to the
all available nonliearity for any |α| , i.e., it hits the up-
shifted-Fock basis introduced in [80] decomposing the
per bound set for this parameter. Increasing the up-
annihilation operator into a qubit and gauge sectors, and
per bounds of all parameters to K2max /2π = 1000 MHz,
defined by:
∆max /2π = 5000 MHz and ϵmax s /2π = 5000 MHz, still
leads to a saturation of K2 at very large mean photon â = Ẑ ⊗ (â′ + α) . (C2)
number. For intermediate values, instead, it converges to
a very large K2 or the order of hundreds of MHz and the Ẑ is a 2 × 2 Pauli Z operator, acting on the qubit sector.
other parameters, ∆ and ϵs , in the order of GHz. This â′ is a bosonic annihilation operator of the gauge sector.
results are unrealistic for state-of-the-art devices. Im- α is the amplitude of the coherent state. The coherent
portantly, using the chiral cat encoding, we can increase state is then |±α⟩ = |±z , 0⟩, so that the cat qubit is the
the protection against bit flips with feasible experimental ground state manifold of â′ . For simplicity, we will assume
parameters. α to be real. The non-Hermitian Hamiltonian then reads
κ κ2 κϕ
Ĥeff = Ĥ − i ↠â − i ↠2 â2 − i (↠â)2
Appendix C: Quantum trajectories in the displaced 2 2 2
Fock basis κ + κϕ † κ2 + κϕ † 2 2
= Ĥ − i â â − i â â
2 2
≃ −4W α2 Iˆ ⊗ â′† â′ − 2W αIˆ ⊗ â′†2 â′ + â′† â′2


The quantum trajectories method, also known as the
Monte Carlo wave function method, is a stochastic de- − cαIˆ ⊗ (â′† + â′ ),
scription of open quantum systems [76–79]. This for- (C3)
malism uses a stochastic Schrödinger equation, and the
resulting wave function |ψ(t)⟩ is known as a quantum where Iˆ is the identity and
trajectory. Each trajectory evolves continuously by
d |ψ(t)⟩ /dt = −iĤeff |ψ(t)⟩, where the non-Hermitian ef- κ2 + κϕ κ1 + κϕ
W = K2 + i , c=∆+i . (C4)
fective Hamiltonian is defined by 2 2
iX † At the same time, the jump operators read
Ĥeff = Ĥ − L̂k L̂k . (C1)
2 √ √
k L̂1 = κ1 Z ⊗ (â′ + α)
κ1 â =
√ √
Here, Ĥ is the system Hamiltonian defined in Eq. (1), L̂2 = κ2 â2 ≃ κ2 αIˆ ⊗ â′ (C5)
√ √
while L̂k are jump operators representing the interaction L̂3 = κϕ ↠â ≃ κϕ αIˆ ⊗ (â′ + â′† ).
 14

(a) Notice that here we neglected K3 , as we make the hypoth-

10 6 esis confined around the low-excitation code manifold,
𝜏error − 𝜏leak (𝜇s)

𝜅𝜙 = 0.1𝜅 1 since we are interested in describing how loss events dete-

10 4 riorate the cat state performance.
𝜅𝜙 = 1𝜅 1
𝜅𝜙 = 10𝜅 1
10 2

(b)
These formulas are very instructive. First, in the non-
Hermitian Hamiltonian, the only term “heating” up the
750
|Δ| (MHz/2𝜋)

system is c = ∆ + i(κ1 + κϕ )/2, exciting the system out

of its ground states |±z , 0⟩. As for the quantum jump
500
operators L̂1 and L̂2 associated with κ1 and κ2 , they do
not induce any bit flip, but rather cool down the system
250 and “clean” the excitation effects induced by the spurious
terms ∆, κ1 and κϕ . We conclude that bit-flip errors
50 are due to the lack of quantum jumps induced by one-
(c)
and two-photon loss, and not because of them. Instead,
Kerr (MHz/2𝜋)

dephasing can be interpreted as an effective thermal-

45 like effect, whose strength increases with the number of
photons in the system. Indeed,

40

(d)
1.50
K3/3 (MHz/2𝜋)

1.25

√ √
1.00 κϕ ↠â |±α⟩ ≃ κϕ αIˆ ⊗ â′† |±z , 0⟩ . (C6)

5 10 15
Mean photon number

FIG. 12. (a) Difference between the bit flip error time with
τerror and without recovery τleak as a function of the mean
photon number. (b-d) Optimal parameters necessary to min-
imise the ratio between error rate and Hamiltonian gap of
Fig. 8: (b) detuning and (c) Kerr (−K2 /2) and (d) first high
order correction (K3 /3). This picture highlights the detrimental effect of dephasing.

[1] M. A. Nielsen and I. L. Chuang, Quantum Computation beyond, Quantum 2, 79 (2018).

and Quantum Information: 10th Anniversary Edition, [7] H.-P. Breuer and F. Petruccione, The Theory of Open
tenth ed. (Cambridge University Press, USA, 2011). Quantum Systems (Oxford University Press, Oxford,
[2] E. T. Campbell, B. M. Terhal, and C. Vuillot, Roads 2007).
towards fault-tolerant universal quantum computation, [8] M. D. Reed, L. DiCarlo, S. E. Nigg, L. Sun, L. Frunzio,
Nature 549, 172 (2017). S. M. Girvin, and R. J. Schoelkopf, Realization of three-
[3] B. M. Terhal, Quantum error correction for quantum qubit quantum error correction with superconducting
memories, Reviews of Modern Physics 87, 307 (2015). circuits, Nature 482, 382–385 (2012).
[4] D. A. Lidar and T. A. Brun, eds., Quantum Error Cor- [9] N. Ofek, A. Petrenko, R. Heeres, P. Reinhold, Z. Leghtas,
rection (Cambridge University Press, Cambridge, 2013). B. Vlastakis, Y. Liu, L. Frunzio, S. M. Girvin, L. Jiang,
[5] S. Haroche and J.-M. Raimond, Exploring the Quantum: M. Mirrahimi, M. H. Devoret, and R. J. Schoelkopf, Ex-
Atoms, Cavities, and Photons (Oxford University Press, tending the lifetime of a quantum bit with error correction
Oxford, 2013). in superconducting circuits, Nature 536, 441 (2016).
[6] J. Preskill, Quantum Computing in the NISQ era and [10] S. Krinner, N. Lacroix, A. Remm, A. Di Paolo, E. Genois,
 15

C. Leroux, C. Hellings, S. Lazar, F. Swiadek, J. Her- rahimi, and M. H. Devoret, Confining the state of light
rmann, G. J. Norris, C. K. Andersen, M. Müller, A. Blais, to a quantum manifold by engineered two-photon loss,
C. Eichler, and A. Wallraff, Realizing repeated quantum Science 347, 853 (2015).
error correction in a distance-three surface code, Nature [30] S. Touzard, A. Grimm, Z. Leghtas, S. O. Mundhada,
605, 669–674 (2022). P. Reinhold, C. Axline, M. Reagor, K. Chou, J. Blumoff,
[11] R. Acharya et al., Suppressing quantum errors by scaling K. M. Sliwa, S. Shankar, L. Frunzio, R. J. Schoelkopf,
a surface code logical qubit, Nature 614, 676–681 (2023). M. Mirrahimi, and M. H. Devoret, Coherent Oscillations
[12] R. Acharya et al., Quantum error correction below the inside a Quantum Manifold Stabilized by Dissipation,
surface code threshold, Nature 638, 920–926 (2024). Physical Review X 8, 021005 (2018).
[13] H. Putterman et al., Hardware-efficient quantum error [31] Q. Xu, J. K. Iverson, F. G. S. L. Brandão, and L. Jiang,
correction via concatenated bosonic qubits, Nature 638, Engineering fast bias-preserving gates on stabilized cat
927–934 (2025). qubits, Physical Review Research 4, 013082 (2022).
[14] K. Noh, C. Chamberland, and F. G. Brandão, Low- [32] H. Goto, Universal quantum computation with a nonlinear
Overhead Fault-Tolerant Quantum Error Correction with oscillator network, Physical Review A 93, 050301 (2016).
the Surface-GKP Code, PRX Quantum 3, 010315 (2022). [33] S. Puri, S. Boutin, and A. Blais, Engineering the quantum
[15] D. Gottesman, A. Kitaev, and J. Preskill, Encoding a states of light in a Kerr-nonlinear resonator by two-photon
qubit in an oscillator, Physical Review A 64, 012310 driving, npj Quantum Information 3, 1 (2017).
(2001). [34] S. Puri, A. Grimm, P. Campagne-Ibarcq, A. Eickbusch,
[16] M. Mirrahimi, Z. Leghtas, V. V. Albert, S. Touzard, R. J. K. Noh, G. Roberts, L. Jiang, M. Mirrahimi, M. H. De-
Schoelkopf, L. Jiang, and M. H. Devoret, Dynamically voret, and S. M. Girvin, Stabilized Cat in a Driven Non-
protected cat-qubits: A new paradigm for universal quan- linear Cavity: A Fault-Tolerant Error Syndrome Detector,
tum computation, New Journal of Physics 16, 045014 Physical Review X 9, 041009 (2019).
(2014). [35] S. Puri, L. St-Jean, J. A. Gross, A. Grimm, N. E. Frattini,
[17] V. V. Albert, Lindbladians with multiple steady states: P. S. Iyer, A. Krishna, S. Touzard, L. Jiang, A. Blais,
Theory and applications (2018), arXiv:1802.00010. S. T. Flammia, and S. M. Girvin, Bias-preserving gates
[18] W. Cai, Y. Ma, W. Wang, C.-L. Zou, and L. Sun, Bosonic with stabilized cat qubits, Science Advances 6, eaay5901
quantum error correction codes in superconducting quan- (2020).
tum circuits, Fundamental Research 1, 50 (2021). [36] A. Grimm, N. E. Frattini, S. Puri, S. O. Mundhada,
[19] A. Joshi, K. Noh, and Y. Y. Gao, Quantum information S. Touzard, M. Mirrahimi, S. M. Girvin, S. Shankar, and
processing with bosonic qubits in circuit QED, Quantum M. H. Devoret, Stabilization and operation of a Kerr-cat
Science and Technology 6, 033001 (2021). qubit, Nature 584, 205 (2020).
[20] B. M. Terhal, J. Conrad, and C. Vuillot, Towards scalable [37] N. E. Frattini, R. G. Cortiñas, J. Venkatraman, X. Xiao,
bosonic quantum error correction, Quantum Science and Q. Su, C. U. Lei, B. J. Chapman, V. R. Joshi, S. M.
Technology 5, 043001 (2020). Girvin, R. J. Schoelkopf, S. Puri, and M. H. Devoret, The
[21] E. Knill, R. Laflamme, and L. Viola, Theory of Quan- squeezed Kerr oscillator: Spectral kissing and phase-flip
tum Error Correction for General Noise, Physical Review robustness (2022), arXiv:2209.03934.
Letters 84, 2525 (2000). [38] J. Venkatraman, R. G. Cortinas, N. E. Frattini,
[22] M. H. Michael, M. Silveri, R. T. Brierley, V. V. Albert, X. Xiao, and M. H. Devoret, Quantum interference
J. Salmilehto, L. Jiang, and S. M. Girvin, New Class of of tunneling paths under a double-well barrier (2022),
Quantum Error-Correcting Codes for a Bosonic Mode, arXiv:2211.04605.
Physical Review X 6, 031006 (2016). [39] R. Gautier, A. Sarlette, and M. Mirrahimi, Combined
[23] L. Gilles, B. M. Garraway, and P. L. Knight, Generation Dissipative and Hamiltonian Confinement of Cat Qubits,
of nonclassical light by dissipative two-photon processes, PRX Quantum 3, 020339 (2022).
Physical Review A 49, 2785 (1994). [40] D. Ruiz, R. Gautier, J. Guillaud, and M. Mirrahimi,
[24] P. T. Cochrane, G. J. Milburn, and W. J. Munro, Macro- Two-photon driven kerr quantum oscillator with multiple
scopically distinct quantum-superposition states as a spectral degeneracies, Phys. Rev. A 107, 042407 (2023).
bosonic code for amplitude damping, Physical Review [41] L. Gravina, F. Minganti, and V. Savona, Critical
A 59, 2631 (1999). schrödinger cat qubit, PRX Quantum 4, 020337 (2023).
[25] J. Guillaud and M. Mirrahimi, Repetition Cat Qubits for [42] R. Lescanne, M. Villiers, T. Peronnin, A. Sarlette, M. Del-
Fault-Tolerant Quantum Computation, Physical Review becq, B. Huard, T. Kontos, M. Mirrahimi, and Z. Leghtas,
X 9, 041053 (2019). Exponential suppression of bit-flips in a qubit encoded in
[26] J. Guillaud and M. Mirrahimi, Error rates and resource an oscillator, Nature Physics 16, 509 (2020).
overheads of repetition cat qubits, Physical Review A [43] C. Berdou, A. Murani, U. Réglade, W. Smith, M. Villiers,
103, 042413 (2021). J. Palomo, M. Rosticher, A. Denis, P. Morfin, M. Del-
[27] J. Guillaud, J. Cohen, and M. Mirrahimi, Quantum com- becq, T. Kontos, N. Pankratova, F. Rautschke, T. Per-
putation with cat qubits, SciPost Phys. Lect. Notes , 72 onnin, L.-A. Sellem, P. Rouchon, A. Sarlette, M. Mir-
(2023). rahimi, P. Campagne-Ibarcq, S. Jezouin, R. Lescanne,
[28] D. Ruiz, J. Guillaud, A. Leverrier, M. Mirrahimi, and and Z. Leghtas, One hundred second bit-flip time in
C. Vuillot, Ldpc-cat codes for low-overhead quantum com- a two-photon dissipative oscillator, PRX Quantum 4,
puting in 2d, Nature Communications 16, 10.1038/s41467- 10.1103/prxquantum.4.020350 (2023).
025-56298-8 (2025). [44] D. S. Schlegel, F. Minganti, and V. Savona, Quantum
[29] Z. Leghtas, S. Touzard, I. M. Pop, A. Kou, B. Vlastakis, error correction using squeezed Schrödinger cat states
A. Petrenko, K. M. Sliwa, A. Narla, S. Shankar, M. J. (2022), arXiv:2201.02570.
Hatridge, M. Reagor, L. Frunzio, R. J. Schoelkopf, M. Mir- [45] T. Hillmann and F. Quijandrı́a, Quantum error correction
 16

with dissipatively stabilized squeezed-cat qubits, Physical 16, 10.1038/s41467-025-56830-w (2025).

Review A 107, 10.1103/physreva.107.032423 (2023). [63] F. Minganti, V. Savona, and A. Biella, Dissipative phase
[46] Q. Xu, G. Zheng, Y.-X. Wang, P. Zoller, A. A. Clerk, and transitions in n-photon driven quantum nonlinear res-
L. Jiang, Autonomous quantum error correction and fault- onators, Quantum 7, 1170 (2023).
tolerant quantum computation with squeezed cat qubits, [64] K. Macieszczak, M. u. u. u. u. Guţă, I. Lesanovsky, and
npj Quantum Information 9, 10.1038/s41534-023-00746-0 J. P. Garrahan, Towards a theory of metastability in open
(2023). quantum dynamics, Phys. Rev. Lett. 116, 240404 (2016).
[47] A. Labay-Mora, R. Zambrini, and G. L. Giorgi, Quantum [65] K. Macieszczak, D. C. Rose, I. Lesanovsky, and J. P. Gar-
memories for squeezed and coherent superpositions in a rahan, Theory of classical metastability in open quantum
driven-dissipative nonlinear oscillator, Phys. Rev. A 109, systems, Physical Review Research 3, 033047 (2021).
032407 (2024). [66] S. Lieu, R. Belyansky, J. T. Young, R. Lundgren, V. V.
[48] R. Gautier, M. Mirrahimi, and A. Sarlette, Designing Albert, and A. V. Gorshkov, Symmetry Breaking and
high-fidelity zeno gates for dissipative cat qubits, PRX Error Correction in Open Quantum Systems, Physical
Quantum 4, 10.1103/prxquantum.4.040316 (2023). Review Letters 125, 240405 (2020).
[49] R. Rousseau et al., Enhancing dissipative cat qubit pro- [67] Y.-J. Liu and S. Lieu, Dissipative phase transitions
tection by squeezing (2025). and passive error correction, Physical Review A 109,
[50] A. Mercurio and Y.-T. Huang, qutip/[Link]: 10.1103/physreva.109.022422 (2024).
v0.24.0 (2024). [68] S. Lieu, E. L. Rosenfeld, K. Noh, and C. T. Hann, View-
[51] A. Blais, A. L. Grimsmo, S. Girvin, and A. Wallraff, ing fluxonium through the lens of the cat qubit (2025),
Circuit quantum electrodynamics, Reviews of Modern arXiv:2501.16425 [quant-ph].
Physics 93, 10.1103/revmodphys.93.025005 (2021). [69] Operationally, we diagonalize ρ±αH =
P (±) (±) (±)
[52] V. V. Albert, B. Bradlyn, M. Fraas, and L. Jiang, Geom- p |ψ ⟩⟨ψ | and define Π̂αH =
j j j j
etry and Response of Lindbladians, Physical Review X 6, P P (k) (k)
041031 (2016). j: pj >10−4 k |ψj ⟩⟨ψj |. We verified that changing

[53] V. V. Albert and L. Jiang, Symmetries and conserved the threshold to smaller values pj does not qualitatively
quantities in Lindblad master equations, Physical Review change the results.
A 89, 022118. [70] S. Puri, L. St-Jean, J. A. Gross, A. Grimm, N. E. Frattini,
[54] F. Minganti, N. Bartolo, J. Lolli, W. Casteels, and P. S. Iyer, A. Krishna, S. Touzard, L. Jiang, A. Blais, S. T.
C. Ciuti, Exact results for Schrödinger cats in driven- Flammia, and S. M. Girvin, Bias-preserving gates with
dissipative systems and their feedback control, Scientific stabilized cat qubits, Science Advances 6, 10.1126/sci-
Reports 6, 26987 (2016). adv.aay5901 (2020).
[55] D. Roberts and A. A. Clerk, Driven-Dissipative Quantum [71] J.-A. M. de Dios and E. Mezura-Montes, Metaheuristics:
Kerr Resonators: New Exact Solutions, Photon Blockade A julia package for single- and multi-objective optimiza-
and Quantum Bistability, Physical Review X 10, 021022 tion, Journal of Open Source Software 7, 4723 (2022).
(2020). [72] D. Goldberg, Genetic Algorithms in Search, Optimization,
[56] N. E. Frattini, R. G. Cortiñas, J. Venkatraman, X. Xiao, and Machine Learning, Addison Wesley series in artificial
Q. Su, C. U. Lei, B. J. Chapman, V. R. Joshi, S. Girvin, intelligence (Addison-Wesley, 1989).
R. J. Schoelkopf, S. Puri, and M. H. Devoret, Observation [73] J. Koch, T. M. Yu, J. Gambetta, A. A. Houck, D. I.
of pairwise level degeneracies and the quantum regime Schuster, J. Majer, A. Blais, M. H. Devoret, S. M. Girvin,
of the arrhenius law in a double-well parametric oscilla- and R. J. Schoelkopf, Charge-insensitive qubit design
tor, Physical Review X 14, 10.1103/physrevx.14.031040 derived from the cooper pair box, Phys. Rev. A 76, 042319
(2024). (2007).
[57] G. Villa, J. del Pino, V. Dumont, G. Rastelli, M. Michalek, [74] A. Blais, A. L. Grimsmo, S. M. Girvin, and A. Wallraff,
A. Eichler, and O. Zilberberg, Topological classifi- Circuit quantum electrodynamics, Rev. Mod. Phys. 93,
cation of driven-dissipative nonlinear systems (2024), 025005 (2021).
arXiv:2406.16591 [[Link]-hall]. [75] The other terms up to order five are:
[58] J. Cohen, A. Petrescu, R. Shillito, and A. Blais, Rem- "  2  5/2 #
iniscence of classical chaos in driven transmons, PRX 70 2EC 3150 2EC
K4 ≈EJ − + (C7)
Quantum 4, 10.1103/prxquantum.4.020312 (2023). 8! EJ 10! EJ
[59] J. Chávez-Carlos, M. A. Prado Reynoso, R. G. Cortiñas, "  5/2 #
252 2EC
I. Garcı́a-Mata, V. S. Batista, F. Pérez-Bernal, D. A. K5 ≈EJ . (C8)
Wisniacki, and L. F. Santos, Driving superconducting 10! EJ
qubits into chaos, Quantum Science and Technology 10, .
015039 (2024). [76] J. Dalibard, Y. Castin, and K. Mølmer, Wave-function
[60] F. Ferrari, L. Gravina, D. Eeltink, P. Scarlino, V. Savona, approach to dissipative processes in quantum optics, Phys-
and F. Minganti, Steady-state quantum chaos in open ical Review Letters 68, 580–583 (1992).
quantum systems (2023), arXiv:2305.15479 [quant-ph]. [77] R. Dum, P. Zoller, and H. Ritsch, Monte carlo simulation
[61] F. Minganti, Out-of-Equilibrium Phase Transitions in of the atomic master equation for spontaneous emission,
Nonlinear Optical Systems, Ph.D. thesis, Université Sor- Physical Review A 45, 4879–4887 (1992).
bonne Paris Cité (2018). [78] K. Mølmer, Y. Castin, and J. Dalibard, Monte carlo wave-
[62] G. Beaulieu, F. Minganti, S. Frasca, V. Savona, S. Fe- function method in quantum optics, Journal of the Optical
licetti, R. Di Candia, and P. Scarlino, Observation of first- Society of America B 10, 524 (1993).
and second-order dissipative phase transitions in a two- [79] H. Carmichael, An Open Systems Approach to Quan-
photon driven kerr resonator, Nature Communications tum Optics: Lectures Presented at the Université Libre
 17

de Bruxelles October 28 to November 4, 1991 (Springer hdanowicz, S. T. Flammia, A. Keller, G. Refael, J. Preskill,
Berlin Heidelberg, 1993). L. Jiang, A. H. Safavi-Naeini, O. Painter, and F. G.
[80] C. Chamberland, K. Noh, P. Arrangoiz-Arriola, E. T. Brandão, Building a Fault-Tolerant Quantum Computer
Campbell, C. T. Hann, J. Iverson, H. Putterman, T. C. Bo- Using Concatenated Cat Codes, PRX Quantum 3, 010329
(2022).