Non-Markovian Quantum Dynamics in Strongly Coupled Multimode Cavities

Conditioned on Continuous Measurement

Valentin Link,1 Kai Müller,1 Rosaria G. Lena,2 Kimmo Luoma,3
François Damanet,4 Walter T. Strunz,1 and Andrew J. Daley2
1
Institut für Theoretische Physik, Technische Universität Dresden, D-01062 Dresden, Germany.
2
Department of Physics and SUPA, University of Strathclyde, G4 0NG Glasgow, United Kingdom.
3
Department of Physics and Astronomy, University of Turku, FI-20014 Turun Yliopisto, Finland.
4
Department of Physics and CESAM, University of Liège, B-4000 Liège, Belgium.
(Dated: February 10, 2022)
An important challenge in non-Markovian open quantum systems is to understand what infor-
mation we gain from continuous measurement of an output field. For example, atoms in multimode
cavity QED systems provide an exciting platform to study many-body phenomena in regimes where
arXiv:2112.09499v2 [quant-ph] 9 Feb 2022

the atoms are strongly coupled amongst themselves and with the cavity, but the strong coupling
makes it complicated to infer the conditioned state of the atoms from the output light. In this
work we address this problem, describing the reduced atomic state via a conditioned hierarchy of
equations of motion, which provides an exact conditioned reduced description under monitoring
(and continuous feedback). We utilise this formalism to study how different monitoring for modes
of a multimode cavity affects our information gain for an atomic state, and to improve spin squeez-
ing via measurement and feedback in a strong coupling regime. This work opens opportunities to
understand continuous monitoring of non-Markovian open quantum systems, both on a practical
and fundamental level.

I. INTRODUCTION

In quantum mechanics, a continuous measurement in-

trinsically influences the dynamics of a system such that
its state depends on the particular measurement record
[1–9]. Theoretical techniques to describe this conditioned
state of the system have been well established [10–12],
and play a central role in quantum control theory. In
particular, the measurement record can be used as a
feedback signal to drive the system in a desired state
[13, 14], counter decoherence [15] or increase entangle-
ment [16, 17]. In recent years, interest has shifted from FIG. 1. Schematic setup, showing the gain of information
few particle systems to the control of many-body quan- about a many-body system of atoms in a multimode cavity,
tum systems. For example, there are opportunities to through detection of light leaking out of the cavity. While the
explore particularly interesting many-body physics [18– combined cavity-atom system (red box) can be Markovian, it
32] with many atoms in optical cavities in regimes where is a highly complex system with many modes. In the case of
the atoms interact strongly amongst themselves and also strong coupling between the atoms and the cavity, the reduced
with the cavity mode(s), as depicted in Fig. 1. In prin- system of atoms (blue box) exhibits non-Markovian dynamics.
Our formalism allows us to determine the information gained
ciple, we would like to understand what information we
about this reduced system from the measurements, determin-
can infer about the state of the atoms from light leak- ing the measurement-conditioned dynamics, and allowing for
ing out of the cavity. However, especially in the case continuous feedback.
of multimode cavities [18–20, 23, 25–27, 33], the prob-
lem becomes rapidly intractable as the system size and
number of cavity modes grows.
In this work we present a new method to describe Markovian. The efficiency of the hierarchical approach
the information we obtain about the state of the atoms, allows us to go beyond previous reduced descriptions for
which also contains a framework for understanding and the state of the atoms in cases where the cavity modes
utilising conditioned dynamics with continuous feedback. can be adiabatically eliminated, addressing directly the
Our exact theory for the reduced state of the atoms strong coupling limit in a form that is also numerically
(dashed blue box in Fig. 1) takes the form of a condi- tractable.
tioned hierarchy of equations of motion, with the hierar- Although for concreteness we discuss the example of
chy capturing the information collectively present in the cavity QED, our theory can be applied to arbitrary quan-
cavity mode(s). This theory thus accounts for a nontriv- tum systems coupled to a non-Markovian bath which can
ial interaction of the atoms with the cavity modes and be split into two parts with a larger Markovian open sys-
vice versa, which makes the dynamics of the atoms non- tem in front of the detector. This includes networks of
 2

standard single-mode cavity QED, coupled resonator ar- where HA and L are respectively the Hamiltonian and
rays and circuit QED [34–36], systems of electrons cou- an arbitrary coupling operator for the atoms, a is the
pled to damped phonons (i.e., a dissipative Hubbard- cavity mode annihilation operator, and g is the atom-
Holstein model [37]) relevant for the solid-state, systems cavity mode coupling strength.
of cold atoms immersed in a BEC [38] or coupled coher- Multimode generalizations of this model constitute
ently to an untrapped level [39, 40], or quantum emitters good candidates for example to explore many-body spin
coupled to plasmonic cavities [41] which can be modeled models from glassiness to associative memories [28] or to
in terms of leaky quasinormal modes [42]. solve specific NP hard problems [29], as they are able
Furthermore, the formalism is itself an interesting ex- to generate tunable-range interactions between atoms
tension to the general theory of open quantum systems. placed inside the cavities [25, 52, 53].
While established non-Markovian open system methods By way of an introduction think, however, of the sim-
make it possible to compute the reduced state of the sys- plest possible example, the Jaynes-Cummings model [54]
tem (the atoms in Fig. 1) in the unmonitored case, mak- consisting of a single two-level atom only. This is de-
ing a connection to continuous measurement was possible scribed by the choices
only in some very special cases [43]. In contrast, here the
exact non-Markovian dynamics of the atoms directly in- ω
HA = σz , L = σ− , (2)
herits a measurement interpretation from the full Marko- 2
vian system of atoms and cavity modes. The embedding
of a non-Markovian system in a larger Markovian system where σα (α = x, y, z) are the standard Pauli opera-
is a well known strategy [44–51]. However, our results tors σx = |0ih1| + |1ih0|, σy = i(|0ih1| − |1ih0|) and
provide a systematic exact theory to determine the con- σz = |1ih1| − |0ih0| with |0i and |1i the two relevant
ditioned states for the reduced system for different mea- atomic states, separated by the atomic transition fre-
surement schemes of the outgoing cavity field, connecting quency ω. This model is particularly simple because the
directly to various experimental setups. Hamiltonian conserves the total number of atomic exci-
tations and photons. In order to gain information on the
Below in Section II, we introduce details on the cav- state of the atoms via the cavity field, the outgoing cavity
ity QED systems that we consider, and revise the corre- field can be continuously monitored, for instance with ho-
sponding continuous measurement theory. We also illus- modyne detection. The joint atom-cavity time evolution
trate why in the strong coupling regime, standard adia- is then conditioned on the measured homodyne current
batic elimination of the cavity mode is not sufficient to Jhom . For perfect detection efficiency, the detected signal
accurately predict the dynamics of the atoms. In Sec- can be written in the form [1]
tion III, we present the derivation of our new exact de-
scription of the conditioned reduced atomic state, which √
takes the form of a conditoned hierarchy of mixed-state Jhom dt = 2κha + a† idt + dW. (3)
quantum trajectories for atom-only density matrices. In
Section IV, the application of our method to multimode Throughout the paper, the brackets h...i denote the quan-
cavities is discussed, and we study how the information tum expectation value with respect to the current con-
we gain about the atomic system and the resulting cor- ditioned state. The signal thus contains information on
relations depend on the way in which the cavity modes a particular quadrature of the mode. The second term
are monitored. In Section V, we extend the theory to that contributes to the homodyne current is white noise
include quantum feedback based on the continuous mea- which arises due to the randomness of quantum measure-
surement, which has direct applications in quantum con- ment outcomes, written here in terms of increments dW
trol. We show how squeezing of a collective spin in cavity of the Wiener process with zero mean E [dW ] = 0, obey-
QED can be improved via feedback, beyond the results ing Ito’s rule dW 2 = dt. The measurement strength κ
known in an adiabatic regime [13]. Finally, in Section is determined by the rate of light leaking out from the
VI, we present our conclusions and discuss further future cavity mirrors. Through continuous measurement, infor-
perspectives arising from this work. mation on the joint system of cavity field and atoms is
continuously acquired. Thus, the state of atoms and field
ρAC is influenced by, or conditioned on, the measurement
outcome and obeys the stochastic evolution equation [1]
II. CONDITIONED ATOM DYNAMICS
dρAC = − i[HAC , ρAC ]dt + κ 2aρAC a† − {a† a, ρAC } dt


√
+ 2κ (a − hai)ρAC + ρAC (a† − ha† i) dW .


While the physics that we address in this work is far
more general, we will use the language of cavity QED. (4)
First, we choose as a concrete example a simple and well
known model: a collection of atoms A, coupled to a single Note that here we may in fact have pure state solutions
mode of an optical cavity C via the Hamiltonian ρAC = |ψAC ihψAC | (quantum trajectories, see later). To
obtain the evolution of the average state, that is the av-
HAC = HA + ∆a† a + g(aL† + a† L) (1) erage with respect to all measurement realizations, we
 3

simply have to omit the terms involving the stochastic in-

crement dW . We are left with the standard master equa-
tion of cavity decay in Gorini-Kossakowski-Sudarshan-
Lindblad (GKSL), or Lindblad, form [55, 56]. The
main goal of this work is to develop a theory describ-
ing the continuously monitored state of the atoms only,
ρA = trC ρAC , by tracing out the cavity field in (4), with-
out further approximation. It is clear that while the mon-
itored ρAC may well be pure, ρA will almost always be
mixed. In the next section we will derive our exact hier-
archical scheme for non-Markovian quantum trajectories
of the monitored, mixed ρA . FIG. 2. Example of homodyning atomic trajectories for the
Before doing so, we discuss two known approaches, Jaynes-Cummings model with g = 2ω, ∆ = ω and κ = 3ω for
where the desired ρA is obtained through approxima- different approximation schemes. Left: Purity of the atomic
tions: the simplest way is via adiabatic elimination – state tr[ρ2A ] as a function of time obtained in the bad cavity
formally a second order perturbation theory in the cou- limit [green line, Eq. (7)], the Redfield limit [blue line, Eq. (5)]
pling strength g, leading to an effective theory for the and the exact equation [red line, Eq. (4)]. As seen on the
reduced state of the atoms described by the conditioned plot, in the bad cavity approximation the state is pure, which
Redfield master equation is not in agreement with the exact dynamics. Thus, in this
approximation, the state of the atom remains on the Bloch
  sphere (right figure).
dρA ≈ − i[HA , ρA ] + L̄ρA , L† + L, ρA L̄† dt
  

i√ 

+ 2κ ρA (L̄† − hL̄† i) − (L̄ − hL̄i)ρA dW.
g timescale κ  ω, ∆. Then we can approximate the Red-
(5) field operator as L̄ ≈ g 2 L/κ which leads to an effective
stochastic master equation of the same form as (4),
Here, the operator L̄ is time-dependent and given as
Z t
g2
dsg 2 e−(i∆+κ)(t−s) e−iHA s LeiHA s . 2LρA L† − {L† L, ρA } dt


L̄(t) = (6) dρA ≈ − i[HA , ρA ]dt +
0 √ κ
ig 2
(L − hLi)ρA + ρA (L† − hL† i) dW .


This equation is explicitly derived later in this work, and + √
also in Ref. [49]. When the average over all measure- κ
ment results is taken, i.e. the second line is neglected, (7)
equation (5) reduces to the deterministic Redfield mas-
ter equation of atoms coupled to a non-Markovian reser- In this approximation all memory effects of the cavity
voir with Lorentzian spectral density, reflecting the leaky field vanish and the atoms obey GKS-Lindblad dynamics.
cavity mode. Also, within the approximation, the mea- To give an example we consider a single realization
sured homodyne current√relates directly to the state of of the Jaynes-Cummings model conditioned on a given
the atoms Jhom dt ≈ −i g2κ hL̄ − L̄† idt + dW , signaling homodyne detection signal, shown in Fig. 2 as a trajec-
that the cavity field is assumed to effectively follow the tory on or inside the Bloch sphere. The dynamics in
state of the atoms. The Redfield equation cannot capture the Jaynes-Cummings model is simply a relaxation of
strong memory effects which may occur in the interaction the atomic excitation to the ground state, which is a
of the atoms with the cavity field. Recent work shows stationary state. The different plots in the figure com-
that for instance in the U (1)-symmetric Dicke model, a pare an exact solution of the Jaynes Cummings model
higher order perturbative evolution equation is required via master equation (4) with the Redfield approximation
to correctly capture the state of the atoms and the phase (5) and the bad cavity limit (7). Clearly, the bad cavity
transition [57], while in the standard dissipative Dicke limit is not applicable in this parameter regime. Note
model Redfield theory does yield accurate results [58]. that, in contrast to the bad cavity limit, the true condi-
Hence, even in relatively simple models, it is not obvious tioned atomic state does not remain pure, as indicated
how to eliminate the cavity modes appropriately. We by a purity trρ2A (t) of less than one. This is because
will see later that Eq. (5) will drop out naturally as only the atom-field interaction leads to finite entanglement
the first order approximation in our general hierarchical between atom and the cavity mode. The Redfield ap-
approach. proximation is close to the exact solution, but does not
On top of the weak-coupling limit leading to the Red- match perfectly even though, here, the cavity field can
field theory, one can make further simplifications based have at most a single photon occupation. This exam-
on assumptions on the timescales of the cavity and atom ple highlights that a more systematic method is needed
processes. A popular approximation is the ’bad cavity’ in order to compute the conditioned state of the atoms
limit, where the cavity decay is assumed to be the fastest within an atom-only description.
 4

III. ATOM-ONLY CONDITIONED non-Markovian dynamics of those atoms, leading to non-

HIERARCHICAL EQUATIONS OF MOTION Markovian atomic quantum trajectories which are mixed
states, and have a clear interpretation in terms of con-
To overcome the clear limitation of an adiabtic treat- tinuous measurement by construction.
ment of the environmental modes, we introduce a fully In order to derive an equation for the state of the
non-Markovian theory for the atoms which is designed to atoms, the cavity field in equation (8) must be traced out.
capture the information we obtain from continuous mon- For this we first project the equation onto a Bargmann
itoring of the environment. If the measurement outcome coherent state of the cavity |yi = exp(ya† )|0i, as in non-
is averaged out, our theory reduces to the well known hi- Markovian quantum state diffusion [70–72] or in Ref. [49],
erarchical equations of motion (HEOM) method, which which yields
has been used successfully for a numerical treatment of
dhy|ψAC i =
non-Markovian open quantum dynamics [59–63]. Our 
theory provides a general non-Markovian analogue to − iHA − ig(L† ∂y∗ + Ly ∗ ) − (κ + i∆)y ∗ ∂y∗
equations for the conditioned state in continuous mea-  (9)
surement theory for Markovian systems [1]. In this anal- + κha + a† i∂y∗ dthy|ψAC i
ogy, the HEOM alone is the counterpart to a master √ 
equation that averages over the measurement results. In + 2κ∂y∗ dW + dN hy|ψAC i
the limit of weak coupling, as well as for special integrable
models, our approach reproduces the results of [49], and To handle the derivative terms we define n-th order aux-
towards Markovian measurements with additional non- iliary states as |ψ (n) (y ∗ , t)i = (gi∂y∗ )n hy|ψAC (t)i which
Markovian baths connects to the results of Refs. [64, 65]. themselves obey the coupled evolution equations
Our method does not rely on any assumptions on the
atom-cavity coupling or the cavity quality, and can in- d|ψ (n) i =
clude measured and unmeasured environmental modes
 
− iHA − igLy ∗ − n(κ + i∆) dt|ψ (n) i
alike.
To sketch the derivation, we will focus on the simplest  i i 
− L† + κha + a† i − (κ + i∆)y ∗ dt|ψ (n+1) i (10)
case of homodyne detection of the cavity output field g g
from a single cavity mode only. The derivation of the var- i√
(n)
ious generalizations, which are presented later on, follows + dN |ψ i − 2κdW |ψ (n+1) i
g
similar lines. For homodyne detection the joint atom and
cavity state obeys the following stochastic Schrödinger + g 2 nLdt|ψ (n−1) i
equation [1, 66, 67], which is the pure state version of
Eq. (4) reminiscent of the hierarchy of pure states (HOPS) in
non-Markvoian quantum state diffusion [79, 80]. Here,
  however, we determine conditioned atomic states un-
d|ψAC i = − iHAC − κa† a + κha + a† ia |ψAC idt der continuous (homodyne) measurement of the cavity
√ (8)
modes. As with HOPS, the reduced state of the atoms is
+ 2κa|ψAC idW + |ψAC idN .
recovered by taking a Gaussian average over the y ∗ vari-
√ able upon acknowledging the completeness of coherent
Here, dN = − κ4 (ha + a† i)2 dt − 12 2κha + a† idW is a fac- states with respect to a Gaussian measure
tor ensuring normalization. The pure conditioned states
|ψAC i are standard Markovian quantum trajectories, and ρA (t) = trC |ψAC (t)ihψAC (t)|
have a clear physical interpretation in terms of continu- Z 2
d y −|y|2 (0) ∗
ous measurement. = e |ψ (y , t)ihψ (0) (y ∗ , t)| (11)
π
Non-Markovian generalizations of quantum trajecto- h i
ries are well known in the literature, for instance the ≡ Ey |ψ (0) (y ∗ , t)ihψ (0) (y ∗ , t)| .
non-Markovian version of quantum jumps [68, 69] and,
more closely related to the results presented here, non- Note that ρA is the state of the atoms conditioned on
Markovian quantum state diffusion [70–72]. They have the homodyne measurement record of the output field.
been used to compute the unmonitored dynamics of the Similar to [81], we can replace the hierarchy of condi-
atoms by propagating stochastic pure states. However, tioned pure states (10) by matrix hierarchichal equations
following [73–78], while a pure state solution of a gen- of motion for the y-averaged, conditioned auxiliary ma-
eral non-Markovian SSE can be interpreted as a condi- trices
tioned state at any particular time [73, 74], joining up h i
(n,m)
these solutions to form a continuously monitored quan- ρA (t) = Ey |ψ (n) (y ∗ , t)ihψ (m) (y ∗ , t)| . (12)
tum trajectory is, up to special exceptions [43], gener-
ally impossible [76]. In our case, the joint atom-cavity To find a closed evolution equation for these
quantum trajectory is Markovian, and, if pure, follows objects one has to employ partial integra-
(8). We now aim to find an atom-only description of the tion under the Gaussian y-mean, which allows
 5

Ey igy|ψ (n) (y ∗ , t)ihψ (m) (y ∗ , t)|

 
to evaluate =
(n+1,m)
Ey ig∂y∗ |ψ (n) (y ∗ , t)ihψ (m) (y ∗ , t)| = ρA
 
(t). The
resulting cHEOM reads
(n,m)
dρ =
 A
(n,m) (n,m)
− i[HA , ρA ] − [(n − m)i∆ + (m + n)κ] ρA
 
(n−1,m) (n,m−1) †
+ g 2 nLρA + mρA L
h
(n+1,m)
i h
(n,m+1)
i (13)
+ ρA , L† + L, ρA dt
 √
(n,m)
+ − 2κhXiρA
i √  (n,m+1) FIG. 3. Average trace distance E ||ρA − ρexact
 
||1 of ho-

(n+1,m)
+ 2κ ρA − ρA dW, A
g modyning trajectories ρA (t) [Eq. (13)] to the exact result
ρexact
A (t) for the driven Jaynes-Cummings model with g = 2ω,
a main result of our work. Here, arising from the mea- ε = 0.5ω, ∆ = ω and κ = √ 2ω using different approximation
† schemes, where ||O||1 = tr O† O denotes the trace norm of
 homodyne current, a term hXi = ha + a i =
sured
(1,0) (0,1) an operator O. The average is taken with respect to 1000
tr ρA − ρA /(ig) appears, that can be determined
samples of the homodyne current and the initial state is |1i.
from the auxiliary matrices of first order. The zeroth or-
(0,0)
der auxiliary state ρA is the exact physical state of the
atoms conditioned on the measurement record, as can be homodyne current (3) from the first level auxiliary states.
seen from Eq. (11). As the above equation is expressed Note that in the following we consider an initial condi-
in Ito formalism, it is written in terms of the stochas- tion with no photons in the cavity, so that all higher
tic increment dW , related to the physical measurement order hierarchy states are initially zero. Their norm then
current via Eq. (3). For completeness, this main result increases over time, as the modes become occupied in the
is presented in the Stratonovich formulation of stochastic dynamics.
calculus in appendix A, where the explicit dependence on Nicely, a second order perturbation theory can be de-
the actual measurement current Jhom becomes obvious. rived with ease from the full hierarchy by formally inte-
To recover the unobserved average reduced atomic grating the equations for the first level auxiliary states
state, one additionally has to take the average over the and neglecting all contributions of higher order. In this
homodyne current Jhom , which amounts to taking the approximation, the first auxiliary states can be expressed
average with respect to the increments dW in (13): one as
simply has to omit all terms proportional to dW . As (1,0)
ρA
(0,0)
(t) = L̄(t)ρA (t) ,
could be expected, this yields the standard HEOM for a (15)
(0,1) (0,0)
quantum system in a bath with Lorentzian spectral den- ρA (t) = ρA (t)L̄† (t) ,
sity [81]. Here, remarkably, we obtain HEOM from an
entirely different, Markovian continuous measurement- where L̄(t) is the Redfield operator (6). Inserting this
based approach. in the zeroth order equation of the hierarchy (13) gives
Clearly, the derivation can be straightforwardly gener- the closed stochastic master equation for the conditioned
alized to other measurement schemes such as direct pho- state of the atoms (5). As expected, taking the aver-
todetection or heterodyne detection. This is shown in age with respect to the increments dW results in the
appendix B, where also the corresponding hierarchical standard deterministic Redfield equation for a bath with
equations are provided. Lorentzian spectral density. Equation (5) can thus be
While the full hierarchy (13) is in principle exact, the seen as a generalization of the Redfield theory to contin-
main practical advantage of the cHEOM arises from the uous measurement.
fact that it can be truncated at finite order, and the con- To showcase the developed cHEOM method, we go be-
sistency of that truncation can be checked: depending on yond the Jaynes-Cummings model (2) and include a driv-
the excitation of the modes, only a few auxiliary states ing term in the Hamiltonian 2ε σx . For a nonzero driving
need to be taken into account. In fact, the trace of aux- the number of excitations is no longer preserved and the
iliary states gives the moments of cavity operators via hierarchy does not truncate at the second order. In Fig. 3
we show the convergence of the solutions with respect to
(n,m) the hierarchy depth, that is we simply truncate the hi-
trρA (t)
han (a† )m i(t) = (14) erarchy by setting ρA
(n,m)
= 0 for n + m larger than the
(ig)n (−ig)m
maximal depth kmax . The figure clearly shows how a
so that a neglect of high order auxiliary states amounts truncated cHEOM gives a systematic expansion beyond
to neglecting corresponding higher order moments. As Redfield theory which converges to the exact result as
used earlier, Eq. (14) makes it possible to compute the the hierarchy depth is increased.
 6

In fact, for this simple example with only a single cav- may not be continuously monitored by any of the mea-
ity mode, solving the full Markovian stochastic master surement schemes discussed above, offering numerical ad-
equation (4) is possible, and our hierarchical atom-only vantages, as we shall see. The general Hamiltonian we
formulation is not required to numerically determine the like to consider reads
atomic time evolution. In the following section we intro-
duce a generalization of the hierarchy to multiple bath M
modes. In this case the cHEOM can open new possibil-
gk (L†k ak + Lk a†k ) + ∆k a†k ak .
X
H = HA + (16)
ities to tackle the challenging description of cavity QED
k=1
systems in multimode cavities.

Here, M cavity modes couple to the atoms with differ-

IV. MULTIPLE CAVITY MODES ent strengths gk and possibly different coupling opera-
tors Lk . For simplicity, we assume homodyne detec-
Exploiting the opportunities of multimode cavities or tion on all modes. In the multimode case the auxil-
ensembles of coupled single-mode cavities experimentally iary density operators in the cHEOM acquire an index
has offered new fascinating possibilities for cavity QED for each mode so that it is useful to define a vector no-
physics simulations of many-body phenomena. The ex- tation where n = (nk ) and m = (mk ) are vectors of
ponential size of the combined atom and cavity modes indices and w = (κk + i∆k ) is a vector storing the cav-
Hilbert space poses problems for a straightforward nu- ity mode detunings and decay rates. Then the cHEOM
merical treatment. Our cHEOM formalism can be gen- for the atom state conditioned on all homodyne currents
eralized easily to multiple cavity modes which may or Jhom,k (t) reads

 
(n,m) (n,m) (n,m)
dρA (t) = − i[HA , ρA ] − (w · n + w∗ · m) ρA dt
M    h i h i
(n−ek ,m) (n,m−ek ) † (n+ek ,m) † (n,m+ek )
X
2
+ gk nk Lk ρA + m k ρA Lk + ρA , Lk + Lk , ρA dt
(17)
k=1
M 
X √ (n,m) i √ 
(n,m+ek ) (n+ek ,m)

+ − 2κk hXk iρA + 2κk ρA − ρA dWk ,
gk
k=1

one of the central results of our work. Here, we Dicke Hamiltonian

use the unit vectors ek = (δkk0 ) and scalar prod- Nclusters NX
modes
P †
uct a · b = kak bk , and hXk i = hak + ak i = a†k ak + gik Jix (ak + a†k )
X X
H=Ω Jiz + ∆
(e ,0) (0,e )
tr ρA k − ρA k /(igk ) arises from the homodyne i=1 k=1 ik
(18)
current. In addition to taking into account multiple
which can be realized by a double Raman pumping
cavity modes with non-Markovian response, further
scheme [28, 83], and Markovian leakage of cavity pho-
Markovian dissipation channels of the atoms can be
tons with rate κ are assumed to be the only source of
included simply by adding them to the hierarchy at each
dissipation. Even for few cavity modes and moderate
level. On the other hand also further non-Markovian
coupling, this system is numerically very challenging be-
baths could be accounted for by additionally employing
yond an adiabatic regime where the cavity modes are
any other suitable HEOM scheme, like the eHEOM
either largely detuned or strongly damped. Thus it is a
method [82].
perfect setup to test the conditioned hierarchy (17).
As an interesting application, we study how monitoring
of different cavity modes increases our knowledge about
To demonstrate the applicability of our method to a the state of the atoms in a single experimental run. This
nontrivial problem, we consider in the following a three- knowledge is quantified by the von-Neumann entropy of
mode cavity where three localized ’clusters’ of atoms are the state of the atoms S[ρA ] = −trρA ln ρA . In case a
trapped. Each of the identical atoms in a particular clus- mode of the cavity field is continuously monitored, this
ter interacts with the cavity field in the same way, as entropy decreases. The expected information gain from
sketched in Fig. 4(a). Then the interaction is collective the monitoring is given by the difference of the entropy
and the cluster can be described by a large spin J~i of size of the average state S[E [ρA ]] and the mean entropy of
j = Natoms /2. The atoms are assumed to interact with the conditioned states E [S[ρA ]] [10, 84]. The latter is a
the individual cavity modes according to a generalized nontrivial nonlinear average of conditioned states which
 7

FIG. 4. Atomic dynamics in a lossy multimode cavity. (a) Sketch of a possible setup corresponding to Hamiltonian (18) and
parameters (19), where three clusters of atoms are trapped in a plane transverse to the cavity axis and interact with cavity
modes localized around the different clusters. Such spatial-dependent intensity profiles could be obtained in a confocal cavity
QED [23, 25]. We assume that each cavity mode can be monitored independently. The panels (b), (c) and (d) show respectively
the mean entropy, the mutual information and the negativity of the combined state of atomic cluster 1 and 3 depending on
which modes of the cavity are continuously monitored via homodyning. The results are obtained from integrating 3000 samples
of the conditioned HEOM (17) with parameters (19)-(21) and Natoms = 3.

cannot be expressed by the reduced state, i.e. it is a prop- the third cluster [10]. As seen in Fig. 4(c), showing the
erty of the full ensemble and not just its mean. Thus it average mutual information E[I13 ], barely any correla-
can only be evaluated by actually solving the full stochas- tions between cluster 1 and 3 are present in case none
tic master equation for many different noise realizations. of the modes is monitored. In contrast, in case both
For an example calculation we choose three cavity modes mode 1 and 3 are monitored with homodyning, signifi-
and the following parameters cant correlations between the clusters are expected in the
individual experimental realizations. If in addition also
∆ = Ω/2, (19) mode 2 is monitored, these correlations decrease again.
κ = 2Ω, (20) Note however, that quantum correlations cannot increase
  from neglecting the measurement results of mode 2. This
0.4 0.115 0.003 means that the increase in mutual information in the case
g = 0.115 0.4 0.115 Ω, (21) where mode 2 is not monitored, versus the case where all
0.003 0.115 0.4 modes are monitored, is exclusively due to classical cor-
relations.
which are far from an adiabatic regime. The coupling
is chosen such that each cavity mode is localized around The third question we like to address is how quantum
a single cluster in a way that the coupling strength de- correlations between clusters 1 and 3 are affected by the
cays towards neighboring clusters. This induces an ef- monitoring. Specifically, we quantify this by the nega-
fective interaction between the clusters, mediated by a tivity N13 ≡ N [ρ13 ] = (||ρΓ131 ||1 − 1)/2 [85], where ρΓ131
common coupling to the cavity field. In particular, we denotes the partial transpose√ of ρ13 with respect to the
focus on the combined state of the first and last cluster first cluster and ||O||1 = tr O† O the trace norm of an
given by ρ13 = tr2 (ρA ), where tr2 denotes the trace over operator O. In Fig. 4(d), showing the mean negativity
the second cluster. Fig. 4(b) shows the average entropy E[N13 ], we observe how the completely monitored state
E[S13 ] ≡ E[S[ρ13 ]] of the state ρ13 depending on which contains the most negativity and then how it decreases
modes of the cavity field are continuously monitored by when modes 1 and 3 and mode 1 alone are monitored.
homodyning. As should be, the more modes are mon- This is explained by the fact that quantum correlations
itored the more information is gained on the state and cannot increase by classical averaging. When mode 2
the lower the mean entropy. Because the atom clusters alone or none of the modes are monitored, the state con-
become entangled with the cavity modes and the cluster tains no negativity.
2, the entropy is nonzero even if all modes are monitored.
Note that, interestingly, monitoring mode 2, which is lo- The effectiveness of our approach can be assessed when
calized around atom cluster 2, does not give a significant contrasted to the problem size when the cavity mode
information gain on the state of cluster 1 and 3, com- Hilbertspace would be directly truncated. The size of
pared to measuring modes localized around the latter. the space of the density matrices of the atomic system
As a second question of interest, we can study how is r = (2j + 1)2Nclusters . Using our approach and trun-
the monitoring affects our knowledge on correlations be- cating the hierarchy according to a triangular condition
PNmodes
tween the clusters, characterized by the mutual informa- k=1 (mk + nk ) ≤ kmax where mk , nk are the hierar-
tion I13 = S[ρ1 ] + S[ρ3 ] − S[ρ13 ] between the first and chy indices results to total of K auxiliary density matrices
 8

with feedback on mode k:

df b ρ(n,m) =
(2Nmodes + kmax )!
K= . (22) 
(n,m) 1 (n,m)
(2Nmodes )!kmax ! Fk ρA Fk − {Fk2 , ρA }
√ 2
2κk h (n+ek ,m) i h i  (24)
(n,m+ek )
Thus, we need to solve a total of d = rK equations. A + ρA , Fk + Fk , ρA dt
gk
direct truncation of the Fock space of the modes at kmax (n,m)
leads to a full state dimension of D = r(kmax + 1)2Nmodes . − i[Fk , ρA ]dWk
For kmax = 3 we have K = 84 and this already gives
Most relevant is the feedback master equation, i.e. the
two orders of magnitude reduction in the problem size
equation for the averaged state. Again, in the Ito for-
d/D = 84/4096 ≈ 0.02. This reduction becomes even
malism, it is obtained by simply omitting all stochastic
more dramatic when the number of modes is increased
terms.
or the modes are coupled more strongly. Further im-
As an example, we apply the formalism to achieve
provements could be achieved with advanced truncation
unconditioned spin squeezing of an atomic system in a
procedures as in Refs. [80, 86].
single-mode cavity via feedback of the homodyne cur-
rent, following [13, 87]. The results in Ref. [13], however,
have been obtained for the bad cavity limit, where an adi-
abatic elimination of the cavity mode can be performed.
By contrast, our approach allows us to study the case
of a ’good’ cavity and we demonstrate the possibility of
V. FEEDBACK achieving squeezing for longer times in this regime. This
example shows that our formalism opens possibilities to
From an experimental perspective, measuring single investigate quantum state preparation in strong coupling
quantum trajectories is in general not feasible, as a par- regimes without any restrictions on cavity parameters. A
ticular trajectory cannot be prepared twice and any aver- detailed discussion of this example follows.
aging required to reconstruct the state will reduce to the As in [13, 88], we consider a system of Natoms two-level
unmeasured case. This is different in case the measure- atoms (spin-1/2) collectively coupled to a single mode of
ment record is used as a feedback signal, which adjusts a lossy cavity described by the Hamiltonian
for instance the strength of a classical driving of the sys-
tem. Then, average measurement results are predicted by H = ΩJz − igJz (a − a† ) , (25)
a feedback master equation which deviates from the dy-
similar to the generalized Dicke Hamiltonian introduced
namics without feedback. Our formalism can be straight-
in Eq. (18). In the following we consider Natoms = 10
forwardly generalized to include feedback on the system
and g = Ω/2.
based on continuous measurement. Here, we present such
We assume that initially, the state of the atoms is a
a theory in the simplest case of instantaneous feedback
coherent spin state with all spins aligned along the direc-
applied on the atomic system, but other generalizations
tion x. The output field is measured via homodyning –
such as delayed feedback can be derived similarly, follow-
in a second step, the current is then fed back according
ing e.g. Ref. [1]. We want to consider in the following
to (23).
instantaneous feedback based on the homodyne signal of
Since the coherent state is a minimum uncertainty
mode k. This corresponds to the feedback Hamiltonian
state, the initial variances along the directions z and x
are equal to j/2 = Natoms /4. As we will see below, the
Hf b (t) = Jhom,k (t)Fk (t) (23) continuous measurement reduces the uncertainty along
z, hence generating spin squeezing, which we quantify
using the spin squeezing parameter [89, 90]
where Fk is an Hermitian operator which may be time-
dependent. We restrict ourselves to the case where the (∆Jz )2
ξz2 = Natoms . (26)
feedback is applied to the atoms, so that Fk is an oper- hJx i2 + hJy i2
ator in the atom Hilbert space. It could, for instance,
describe an external driving controlled by the experi- However, the spin squeezing of the conditioned states dis-
menter. In the Stratonovich formulation of the condi- appears after carrying out the ensemble average over all
tioned HEOM provided in supplement A, the feedback possible measurement records, because of the presence of
can be trivially included simply by modifying the atom a stochastic shift of hJz i which returns the unconditioned
Hamiltonian accordingly. This can be converted to the initial variance. Therefore, as in Ref. [13], one introduces
Ito formalism which then allows to take the ensemble a coherent feedback based on the measurement to coun-
average. As shown in appendix C, the following contri- teract this stochastic shift, in order to maintain the spin
bution to the conditioned HEOM (17) arises due to the squeezing. Our goal here is to apply this idea to achieve
 9

such spin squeezing in the good cavity limit, where the For the total stochastic shift to vanish for a single tra-
cavity mode cannot be eliminated. We will see that the jectory, we have to impose dhJz i = 0. From Eq. (30)
feedback conditioned on the measurement will then not and Eq. (33) we therefore obtain the condition on the
depend on quantities of the atoms only, but on quantities feedback strength
of both the atoms and the cavity, and more precisely on
their correlations, as we show in the following. hJz Xi
λ(t) = 2κ . (34)
From the general form of the Hamiltonian as in Eq. (1) hJx i
we can identify the correspondence
It is worth highlighting that the feedback strength in
HA = ΩJz , Eq. (34) depends dynamically on conditioned expecta-
L = iJz , (27) tion values of both a quantity of the atomic system alone,
Jx , and of a quantity that is related to the correlations
∆ = 0.
between atoms and cavity, Jz X, as shown in Fig. 5.
A continuous measurement produces a stochastic shift Note that we can directly extract the relevant expecta-
of the z component hJz i. Using the conditioned HEOM tion value from the hierarchy (17) as
equation (13), as well as (14), we find that this shift is  
(1,0) (0,1)
given by hJz Xi = i/g Tr Jz (ρA − ρA ) . (35)

dhJz i =Tr[Jz dρA ] In the bad cavity limit, where the cavity mode can be
√
= 2κ(hJz Xi − hJz ihXi)dW, (28) adiabatically eliminated, one can set optimised values for
the feedback strength under the assumptions that perfect
where we used the quadrature X = a + a† . From the measurements on the system keep it in a pure state, so
homodyne current in Eq. (3) we can extract the form of that the conditioned expectation values can be approxi-
dW and use this in Eq. (28), obtaining mated with the unconditioned ones (indicated with the
subscript u): hJx i ' hJx iu and hJz2 i ' hJz2 iu [88].
√ √
dhJz i = 2κdt(hJz Xi − hJz ihXi)(Jhom (t) − 2κhXi),
(29)
which, using the fact that hXi = 0 (as shown in Figure 5),
simplifies to
√
dhJz i = 2κhJz XiJhom (t)dt. (30)

In the absence of feedback, this stochastic shift, averaged

over all the trajectories, will return a final non squeezed
state, recovering the unconditioned variance for Jz with
the same value j/2 of the initial coherent state. This
shift induced by the measurement is hence detrimental
for the spin squeezing along z and must be counteracted
with an additional feedback term that continuously acts
on the system and induces an opposite shift of hJz i.
Following the same protocol as in [13], we add a feed- FIG. 5. Time-dependent behavior of the correlations hJz Xi
back that generates a rotation of the collective spin for two different values of the cavity loss rate κ = Ω (solid red
around the y axis: line) and 10Ω (dashed blue line) – corresponding respectively
to the good and bad cavity regimes – and of hXi (dashed
λ(t) dotted green line) for both values of κ. While the quadrature
Hf b (t) = √ Jhom (t)Jy = F Jhom (t), (31) hXi remains zero for all times in both regimes, correlations
2κ
between the atoms and the cavity mode build up. The smaller
where we defined the feedback operator κ, the larger the correlations.

λ(t) In the good cavity limit, however, the state becomes

F = √ Jy , (32)
2κ mixed, and we determine the optimal values of the feed-
back strength numerically. While we could choose a time-
and where λ is the feedback strength. As the feedback dependent feedback, determining this would not be prac-
Hamiltonian adds the extra terms (24) to the conditioned tical in an experimental setup. We therefore consider the
HEOM, the feedback induces another shift on hJz i simpler situation of constant feedback applied through-
out the dynamics, using different values of the feedback
λ(t) strength λ within a given range, to see which values opti-
df b hJz i = − √ Jhom (t)hJx idt. (33)
2κ mize the spin squeezing along z, Eq. (26). This is shown
 10

times, by using a sequence of the optimal constant feed-

back strengths with different signs, switching from one to
another at given times. The result is shown by the solid
lines in Fig. 7.
Note that while our approach provides a simple phys-
ical picture of the mechanism leading to a better spin
squeezing, one could potentially improve further these
results by optimizing a continuous time-dependent feed-
back strength λ(t) via optimal control schemes. However,
from an experimental point of view, this would be much
harder to implement.

FIG. 6. Minimum spin squeezing parameter ξz2 obtained over

time as a function of constant feedback strength for two dif-
ferent values of κ that corresponds to the good (κ = Ω,
red solid line) and bad (κ = 10Ω, blue dashed line) cavity
regimes. Better spin squeezing is achieved in the good cav-
ity regime. The brown dashed line indicates the feedback
strength λ∗ = 2κ max{t} hJz Xi/hJx (0)i, which is evaluated
using the maximum of hJz Xi over time (shown in Fig. 5), and
takes the same value for both κ = Ω and κ = 10Ω. The values
(κ/Ω)
λ−,+ of the feedback strength corresponding to the negative
and positive minima, in the good and bad cavity limit, are:
(1) (10) (1) (10)
λ− = λ− = −0.22Ω, λ+ = 0.23Ω and λ+ = 0.55Ω.

in Fig. 6 for both the good (red line) and bad (blue line)
cavity limits.
First, we can observe that the good cavity case, in
this regime of parameters, produces better spin squeez-
ing. Second, while for the bad cavity regime, considering
a feedback strength λ∗ set by the maximum of the corre-
lations hJz Xi is a good approximation, this is not true in
the good cavity limit. In fact, if we use this approxima-
FIG. 7. Spin squeezing obtained applying the feedback of
tion to set the value of the feedback strength, we do not
Eq. (31) with different feedback strengths, for the good [top,
obtain any spin squeezing, and this is due to two factors: panel (a)] and bad [bottom, panel (b)] cavity regimes κ =
first, the correlations are stronger and grow more slowly Ω and κ = 10Ω respectively. The green (red) dashed line
(as shown in Figure 5), introducing a limitation of the is obtained when we apply the feedback with the optimal
approximation to a constant value; second, in the good (κ/Ω)
positive (negative) feedback parameter λ+(−) coherently from
cavity limit the state of the system gets mixed quickly t = 0. The maximum spin squeezing is obtained at Ωt = 4
and the conditioned values obtained from the measure- and Ωt ∼ 1.8 for the good and bad cavity limits respectively.
ment differ from the unconditioned ones that we use to The blue solid line is the spin squeezing parameter where
define the constant feedback strength. Fig. 6 also shows we alternate the positive and negative feedback following the
(κ/Ω) (κ/Ω) (κ/Ω)
that in both the good and bad cavity limits, when we sequence: λ+ → λ− → λ+ , where the switching
consider the full model, we have two local minima corre- occurs at times t1
(κ/Ω)∗ (κ/Ω)∗
and t2 . Here we have Ωt1
(1)∗
=
sponding to values of λ having different signs. It is worth Ωt1
(10)∗ (1)∗
= 1.6, Ωt2 = 4.66 and Ωt2
(10)∗
= 4.81.
pointing out that if we adiabatically eliminate the cavity
mode, for the same values of the parameters used here,
we would obtain only one of the two minima and the
physics of the second minimum would not be captured in
this approximation. VI. CONCLUSIONS AND OUTLOOK
The constant feedback applied with strengths set by
the different values of the minima in Fig. 6 generates spin In this article we have developed a general theory de-
squeezing at different points in time, as can be observed scribing the time evolution of a non-Markovian subsys-
from the dashed lines in Fig. 7. We can exploit this to tem coupled to damped bosonic modes which are mon-
contrast the increase of the the spin squeezing parameter itored continuously. This appears in many scenarios,
after a transient [13] and decrease it further for longer including advanced cavity QED where multiple cavity
 11

modes couple to an interacting many-body quantum sys- quantum gates between photonic qubits [105, 106]), or
tem (atoms) embedded in the cavity. We address the the potential of using the feedback formalism to imple-
problem which information about the atoms can be in- ment error correction protocols. Finally, it is important
ferred from continuous measurement of the leaking out- to stress again that while we use the language of opti-
put light of the various cavity modes, and how that cal cavity QED, the underlying model is universal and
knowledge can be used to manipulate the many-body can equally be applied to plasmonic cavities [41], cold
state using feedback. We refrain from using any of the atoms reservoirs [38–40], electron-phonon systems [37],
usual approximations but succeed in formulating a the- or circuit QED [34, 36]. Given the high cooperativity
ory that allows us to determine directly the exact, non- achievable in this latter platform, we expect our formal-
Markovian atomic (mixed) state dynamics, conditioned ism to be indeed particularly useful for exploring control
on the measurement record. We can thus study in de- and readout of superconducting qubits.
tail the gradual information gain arising from monitoring
more and more cavity modes. We show how the contin-
uous observation of spatially selective cavity modes re-
veals information about (quantum) correlations between
groups of atoms at different locations inside the cavity. ACKNOWLEDGMENTS
Moreover, we can determine non-Markovian feedback hi-
erarchical equations of motion as a starting point for con-
trol theory, allowing us to drive the many-body quantum It is a pleasure to thank Richard Hartmann and Stu-
system in desired states. Specifically, we consider a feed- art Flannigan for various helpful discussions in connec-
back scenario which aims to generate spin squeezing in a tion with this work. Work at the University of Strath-
collective ensemble of atoms. Here, the mixedness of the clyde was supported by the EPSRC Programme Grant
conditioned atom state invalidates the previously known DesOEQ (EP/P009565/1), the European Union’s Hori-
optimal feedback protocols of the adiabatic regime. In zon 2020 research and innovation program under grant
fact, with a more coherent cavity it is possible to gener- agreement No. 817482 PASQuanS, and AFOSR grant
ate stronger spin squeezing for significantly longer times. number FA9550-18-1-0064. V. L. , F. D., W. S. and
Crucially, in our approach, no restrictions on coupling A. D. thank KITP for hospitality during this work, sup-
strengths or cavity qualities are required. Instead, our re- ported by the National Science Foundation under Grant
sult takes the form of conditioned hierarchical equations No. NSF PHY-1748958.
of motion (cHEOM, Eq. (17)), proving to be an efficient
scheme for tackling pressing issues in the highly complex
quantum dynamics of strong atom-cavity coupling.
Our result paves the way for further exciting routes
Appendix A: Homodyne cHEOM in Stratonovich
of study: our exact cHEOM could be naturally coupled
convention
to approximated schemes such as mean-field or cumu-
lant expansion, to easily go towards larger many-body
atomic systems while keeping the system-reservoir cou- To derive the Stratonovich from of equation (13) we
pling exact. It could also be advantageous to formulate start from the conditioned atom and cavity evolution in
cHEOM with matrix product operators. In connection Stratonovich convention [1, 67]
with the hierarchy of pure states method, recent works
have demonstrated that such an ansatz can be used both
to describe many-body dynamics in non-Markovian envi- ∂t |ψi = −iHA − ig(L† a + La† ) − (κ + i∆)a† a
ronments [37] and very strong system-bath coupling [91]. √  (A1)
−κa2 + 2κJhom (t)a + N (t) |ψi,
Our theory constitutes an ideal framework for the ex-
ploration of a wide range of many-body phenomena on
the level of the reduced atomic non-Markovian dynam-
where N (t) = κ(ha† ai + 1/2ha2 + (a† )2 i) −
ics in the strong-coupling regime, such as phase transi- p †
tions [92], measurement-induced phase transitions [93, κ/2Jhom (t)ha + a i is a factor which ensures nor-
94], neural network like behaviors such as associative malization. The measured homodyne current is given
memories [28, 95], cavity-enhanced transport [96–100] as
and superconductivity [101], continuous measurement of
√
transport [102], or cavity cooling with higher capture Jhom (t) = 2κha + a† iJ (t) + ζ(t) , (A2)
range [103] and its monitoring [104]. New schemes for
quantum information processing and production of en-
tanglement [32] could also be investigated, exploiting where ζ(t) is a Gaussian white noise process in the
the higher coherence achievable in the strong-coupling Stratonovich sense with statistics E [ζ(t)] = 0 and
regime, the use of different cavity modes to realize quan- E [ζ(t)ζ(s)] = δ(t − s). In order to obtain the hierar-
tum gates (or conversely the use of the atoms to realize chy we follow the same derivation as in section III, to
 12

arrive at the Stratonovich version of Eq. (13) 0 and dWc dWc∗ = dt. The HEOM in this case reads

∂t ρ(n,m) (t) = dρ(n,m)
a = − i[Ha , ρ(n,m)
a ] − ((m − n)∆ + (m + n)κ)ρ(n,m)
a
a
2 (n−1,m) (n,m−1) †
− i[HA , ρ(n,m) ] − [(n − m)i∆ + (m + n)κ] ρ(n,m)


a a + g nLρa + mρa L
 
2
+ g nLρa (n−1,m)
+ mρa(n,m−1) †
L + [ρ(n+1,m)
a , L† ] + [L, ρ(n,m+1)
a ]
√
+ 2κ −haidW − ha† idW ∗ ρ(n,m)
h i h i

a
+ ρa(n+1,m) , L† + L, ρ(n,m+1)
a √ √
i 2κ (n,m+1) i 2κ (n+1,m)
i√   + ρa dW ρa dW ∗ .
+ 2κJhom (t) ρ(n,m+1)
a − ρ(n+1,m)
a g g
g (B3)
κ  
+ 2 ρ(n+2,m)
a + ρ(n,m+2)
a − 2ρ(n+1,m+1)
a
g Another well known unraveling of master equation (16)
− 2 Re (N (t)) ρ(n,m)
a . are quantum jumps which are related to continuous direct
(A3) photodetection [1, 66]. In this case the conditioned evo-
lution is piecewise deterministic until at a random time a
photon is detected and the state jumps discontinuously.
The advantage of the Stratonovich formulation is that it An evolution equation describing this can be formulated
becomes obvious that the state depends on the homodyne as follows
current (A2) only. Similarly, the second order perturba-
tion theory equation (5) is equivalent to the following d|ψ(t)i = −iHA |ψ(t)idt
Stratonovich equation  
− ig(L† a + La† ) − (κ + i∆)a† a − κha† aiN |ψ(t)idt
∂t ρa = − i[HA , ρa ] + L̄ρa , L† + L, ρa L̄†
    !
a
+ p − 1 |ψ(t)idN (t).
i√ ha† aiN (t)
2κJhom (t) ρa L̄† − L̄ρa (t)


+
g (A4) (B4)
κ Here, dN (t) is the increment of a realization of a Poisson
+ 2 L̄2 ρa (t) + ρa (L̄† )2 − 2L̄ρa (t)L̄†


g process obeying
− 2 Re (N (t)) ρa ,
E [dN ] = 2κha† aiN dt , (dN )2 = dN. (B5)
which reduces to a Redfield master equation on average. Clearly, dN can be either one or zero, depending on
whether a jump does or does not occur. To realize this
in a numerical implementation one first draws a random
number r between 0 and 1 and then evolves the state
Appendix B: Different Detection Schemes according to the deterministic part of Eq. (B4). Then a
Rt
jump occurs when 0 ds 2κha† aiN (s) = − ln r is satisfied
p
The derivations in section III can be straightforwardly and the state looses one photon |ψi → a|ψi/ ha† aiN .
applied to different measurement schemes. For exam- For the direct photodetection the hierarchy of equations
ple the formally very similar continuous heterodyne de- of motion reads
tection of the cavity output field yields the stochastic
Schrödinger equation [1, 66] dρ(n,m)
a =

 − i[HA , ρ(n,m)
a ] − ((n − m)i∆ + (m + n)κ) ρ(n,m) a
d|ψ(t)i = − iHA − ig(L† a + La† ) 2

(n−1,m) (n,m−1) †

+ g nLρa + mρa L


− κ ha† ihai − 2ha† ia − a† a |ψ(t)idt (B1) h
†
i h i
√ + ρ(n+1,m)
a , L + L, ρ (n,m+1)
a dt
+ 2κ (a − hai) |ψ(t)idWc (t).  
1 (n+1,m+1)
+ 2κha† aiN (t)ρ(n,m) a − 2κ ρ dt
g2 a
This describes the evolution of atom and cavity state  
conditioned on the heterodyne current 1 (n+1,m+1) (n,m)
+ ρ − ρ dN (t).
g 2 ha† aiN (t) a a
√ (B6)
Jhet (t)dt = 2κhaidt + dWc , (B2)
One can check immediately that the last two lines drop
with the now complex Gaussian increment dWc , whose out after taking the average with relations (B5) and the
stochastic properties are given by E [dWc ] = dWc dWc = standard HEOM is recovered again. If a jump occurs
 13

then the hierarchy has to be modified according to This Stratonovich equation is then translated to the Ito
equation
(n+1,m+1)
ρa (t)
ρ(n,m)
a (t) → , (B7)
g 2 ha† aiN (t)
 
dt X ∂ ∂
dψj = αj dt + βj dζ + βl βj + βl∗ ∗ βj .
2 ∂ψl ∂ψl
i.e. the entire hierarchy is moving down one level. l

Applied to (C1) we arrive at the following SSE with feed-

back in Ito form:
Appendix C: Homodyne cHEOM with feedback
  

gj (L† aj + La†j ) + ∆j a†j aj 

X
To derive a HEOM that includes feedback we start d|ψi = −i HA +
from the SSE in Stratonovich convention as described j
in appendix A. There the rules for stochastic integration  
−κl a†l al − ha†l + al ial + hal + a†l i2 /4
X
allow to simply add the feedback Hamiltonian (31), which +
describes linear feedback based on the measurement of a l
!
single mode k. We obtain p † √
 − i κk /2hak + ak iFk − i 2κk Fk ak − Fk2 dt|ψi

gj (L† aj + La†j )
X
∂t |ψi = −i(HA + Jhom,k Fk ) − i
 
 
2κj aj + haj + a†j i/2 − iFk  dζ|ψi.
Xp
j +
j
−(κj + i∆j )a†j aj −κj a2j
X
+
j
(C2)
 Starting from the equation above we now follow the
Xp same lines as in Sec. III, i.e. we project the equation
+ 2κj Jhom,j (t)aj + N (t) |ψi. onto a Bargmann coherent state |yi, define the auxil-
n
j iary states |ψ (n) i = (ig∂y∗ ) hy|ψi and obtain the HEOM
(C1) from ρA
(n,m)  (n) (m) 
= Ey |ψ ihψ | . This leads us to
As we consider only feedback on the atoms FK is an
(n,m) (n,m) (n,m)
operator in the system Hilbert space. From Eq. (A2) dρA = dmsmt. ρA + d f b ρA , (C3)
we see that the above SSE depends on the real noise
ζ(t). To convert it into Ito formalism one can then take (n,m)
an arbitrary basis, define ψj = hj|ψi, Lj,k = hj|L|ki and where dmsmt. ρA is given as the left hand side of
make use of the conversion formula [67] for a Stratonovich Eq. (17) and describes the change of the state due to the
equation of the form continuously measured evolution. The new terms intro-
(n,m)
duced through the feedback are captured by df b ρA ,
∂t ψj = αj + βj ζ(t). which is given in Eq. (24).

[1] H. M. Wiseman and G. J. Milburn, Quantum Measure- 4363 (1992).

ment and Control (Cambridge University Press, 2009). [7] N. Gisin and I. C. Percival, The quantum-state diffusion
[2] A. Barchielli and V. P. Belavkin, Measurements contin- model applied to open systems, Journal of Physics A:
uous in time and a posteriori states in quantum mechan- Mathematical and General 25, 5677 (1992).
ics, Journal of Physics A: Mathematical and General 24, [8] H. J. Carmichael, An Open Systems Approach to Quan-
1495 (1991). tum Optics (Springer, Berlin, Germany, 1993).
[3] H. M. Wiseman and G. J. Milburn, Interpretation of [9] M. B. Plenio and P. L. Knight, The quantum-jump ap-
quantum jump and diffusion processes illustrated on the proach to dissipative dynamics in quantum optics, Rev.
Bloch sphere, Phys. Rev. A 47, 1652 (1993). Mod. Phys. 70, 101 (1998).
[4] J. Dalibard, Y. Castin, and K. Mølmer, Wave-function [10] K. Jacobs, Quantum Measurement Theory and its Ap-
approach to dissipative processes in quantum optics, plications (Cambridge University Press, 2014).
Phys. Rev. Lett. 68, 580 (1992). [11] A. J. Daley, Quantum trajectories and open many-body
[5] K. Mølmer, Y. Castin, and J. Dalibard, Monte Carlo quantum systems, Adv. Phys. 63, 77 (2014).
wave-function method in quantum optics, J. Opt. Soc. [12] J. Zhang, Y.-x. Liu, R.-B. Wu, K. Jacobs, and F. Nori,
Am. B 10, 524 (1993). Quantum feedback: Theory, experiments, and applica-
[6] C. W. Gardiner, A. S. Parkins, and P. Zoller, Wave- tions, Phys. Rep. 679, 1 (2017).
function quantum stochastic differential equations and [13] L. K. Thomsen, S. Mancini, and H. M. Wiseman, Spin
quantum-jump simulation methods, Phys. Rev. A 46, squeezing via quantum feedback, Phys. Rev. A 65,
 14

061801 (2002). [31] F. Ferri, R. Rosa-Medina, F. Finger, N. Dogra, M. Sori-

[14] C. Sayrin, I. Dotsenko, X. Zhou, B. Peaudecerf, ente, O. Zilberberg, T. Donner, and T. Esslinger,
T. Rybarczyk, S. Gleyzes, P. Rouchon, M. Mirrahimi, Emerging Dissipative Phases in a Superradiant Quan-
H. Amini, M. Brune, J.-M. Raimond, and S. Haroche, tum Gas with Tunable Decay, Phys. Rev. X 11, 041046
Real-time quantum feedback prepares and stabilizes (2021).
photon number states, Nature 477, 73 (2011). [32] S. J. Masson and S. Parkins, Rapid Production of Many-
[15] D. Vitali, P. Tombesi, and G. J. Milburn, Controlling Body Entanglement in Spin-1 Atoms via Cavity Output
the Decoherence of a ”Meter” via Stroboscopic Feed- Photon Counting, Phys. Rev. Lett. 122, 103601 (2019).
back, Phys. Rev. Lett. 79, 2442 (1997). [33] S. P. Kelly, A. M. Rey, and J. Marino, Effect of Ac-
[16] J. Wang, H. M. Wiseman, and G. J. Milburn, Dynamical tive Photons on Dynamical Frustration in Cavity QED,
creation of entanglement by homodyne-mediated feed- Phys. Rev. Lett. 126, 133603 (2021).
back, Phys. Rev. A 71, 042309 (2005). [34] S. Schmidt and J. Koch, Circuit QED lattices: To-
[17] J.-G. Li, J. Zou, B. Shao, and J.-F. Cai, Steady atomic wards quantum simulation with superconducting cir-
entanglement with different quantum feedbacks, Phys. cuits, Ann. Phys. 525, 395 (2013).
Rev. A 77, 012339 (2008). [35] C. Noh and D. G. Angelakis, Quantum simulations and
[18] S. Gopalakrishnan, B. L. Lev, and P. M. Gold- many-body physics with light, Reports on Progress in
bart, Emergent crystallinity and frustration with Physics 80, 016401 (2016).
Bose–Einstein condensates in multimode cavities, Na- [36] A. Blais, A. L. Grimsmo, S. M. Girvin, and A. Wallraff,
ture Physics 5, 845 (2009). Circuit quantum electrodynamics, Rev. Mod. Phys. 93,
[19] S. Gopalakrishnan, B. L. Lev, and P. M. Goldbart, 025005 (2021).
Atom-light crystallization of Bose-Einstein condensates [37] S. Flannigan, F. Damanet, and A. J. Daley, Many-body
in multimode cavities: Nonequilibrium classical and quantum state diffusion for non-Markovian dynamics in
quantum phase transitions, emergent lattices, superso- strongly interacting systems, arXiv (2021).
lidity, and frustration, Phys. Rev. A 82, 043612 (2010). [38] R. G. Lena and A. J. Daley, Dissipative dynamics and
[20] S. Gopalakrishnan, B. L. Lev, and P. M. Goldbart, cooling rates of trapped impurity atoms immersed in a
Frustration and Glassiness in Spin Models with Cavity- reservoir gas, Phys. Rev. A 101, 033612 (2020).
Mediated Interactions, Phys. Rev. Lett. 107, 277201 [39] I. de Vega, D. Porras, and J. I. Cirac, Matter-Wave
(2011). Emission in Optical Lattices: Single Particle and Col-
[21] A. T. Black, H. W. Chan, and V. Vuletić, Observa- lective Effects, Phys. Rev. Lett. 101, 260404 (2008).
tion of Collective Friction Forces due to Spatial Self- [40] C. Navarrete-Benlloch, I. de Vega, D. Porras, and J. I.
Organization of Atoms: From Rayleigh to Bragg Scat- Cirac, Simulating quantum-optical phenomena with
tering, Phys. Rev. Lett. 91, 203001 (2003). cold atoms in optical lattices, New Journal of Physics
[22] H. Ritsch, P. Domokos, F. Brennecke, and T. Esslinger, 13, 023024 (2011).
Cold atoms in cavity-generated dynamical optical po- [41] K. Santhosh, O. Bitton, L. Chuntonov, and G. Haran,
tentials, Rev. Mod. Phys. 85, 553 (2013). Vacuum Rabi splitting in a plasmonic cavity at the sin-
[23] A. J. Kollár, A. T. Papageorge, K. Baumann, M. A. gle quantum emitter limit, Nature Communications 7,
Armen, and B. L. Lev, An adjustable-length cavity and ncomms11823 (2016).
Bose–Einstein condensate apparatus for multimode cav- [42] S. Franke, S. Hughes, M. K. Dezfouli, P. T. Kristensen,
ity QED, New Journal of Physics 17, 043012 (2015). K. Busch, A. Knorr, and M. Richter, Quantization of
[24] A. J. Kollár, A. T. Papageorge, V. D. Vaidya, Y. Guo, Quasinormal Modes for Open Cavities and Plasmonic
J. Keeling, and B. L. Lev, Supermode-density-wave- Cavity Quantum Electrodynamics, Phys. Rev. Lett.
polariton condensation with a Bose–Einstein conden- 122, 213901 (2019).
sate in a multimode cavity, Nat. Commun. 8, 1 (2017). [43] N. Megier, W. T. Strunz, and K. Luoma, Continuous
[25] V. D. Vaidya, Y. Guo, R. M. Kroeze, K. E. Ballan- quantum measurement for general Gaussian unravelings
tine, A. J. Kollár, J. Keeling, and B. L. Lev, Tunable- can exist, Phys. Rev. Research 2, 043376 (2020).
Range, Photon-Mediated Atomic Interactions in Multi- [44] A. Imamoglu, Stochastic wave-function approach to
mode Cavity QED, Phys. Rev. X 8, 011002 (2018). non-Markovian systems, Phys. Rev. A 50, 3650 (1994).
[26] F. Mivehvar, F. Piazza, T. Donner, and H. Ritsch, Cav- [45] B. J. Dalton, S. M. Barnett, and B. M. Garraway, The-
ity QED with quantum gases: new paradigms in many- ory of pseudomodes in quantum optical processes, Phys.
body physics, Advances in Physics 70, 1 (2021). Rev. A 64, 053813 (2001).
[27] A. Wickenbrock, M. Hemmerling, G. R. M. Robb, [46] B. M. Garraway, Nonperturbative decay of an atomic
C. Emary, and F. Renzoni, Collective strong coupling system in a cavity, Phys. Rev. A 55, 2290 (1997).
in multimode cavity QED, Phys. Rev. A 87, 043817 [47] G. Pleasance, B. M. Garraway, and F. Petruccione, Gen-
(2013). eralized theory of pseudomodes for exact descriptions
[28] B. P. Marsh, Y. Guo, R. M. Kroeze, S. Gopalakrish- of non-Markovian quantum processes, Phys. Rev. Re-
nan, S. Ganguli, J. Keeling, and B. L. Lev, Enhancing search 2, 043058 (2020).
Associative Memory Recall and Storage Capacity Using [48] L. Mazzola, S. Maniscalco, J. Piilo, K.-A. Suominen,
Confocal Cavity QED, Phys. Rev. X 11, 021048 (2021). and B. M. Garraway, Pseudomodes as an effective de-
[29] V. Torggler, P. Aumann, H. Ritsch, and W. Lechner, A scription of memory: Non-Markovian dynamics of two-
Quantum N-Queens Solver, Quantum 3, 149 (2019). state systems in structured reservoirs, Phys. Rev. A 80,
[30] K. Baumann, C. Guerlin, F. Brennecke, and 012104 (2009).
T. Esslinger, Dicke quantum phase transition with a [49] H. Yang, H. Miao, and Y. Chen, Nonadiabatic elimi-
superfluid gas in an optical cavity, Nature 464, 1301 nation of auxiliary modes in continuous quantum mea-
(2010). surements, Phys. Rev. A 85, 040101 (2012).
 15

[50] H.-P. Breuer, Genuine quantum trajectories for non- [69] K. Luoma, W. T. Strunz, and J. Piilo, Diffusive Limit
Markovian processes, Phys. Rev. A 70, 012106 (2004). of Non-Markovian Quantum Jumps, Phys. Rev. Lett.
[51] A. Barchielli, C. Pellegrini, and F. Petruccione, Stochas- 125, 150403 (2020).
tic Schrödinger equations with coloured noise, EPL (Eu- [70] L. Diósi and W. T. Strunz, The non-Markovian stochas-
rophysics Letters) 91, 24001 (2010). tic Schrödinger equation for open systems, Physics Let-
[52] Y. Guo, R. M. Kroeze, V. D. Vaidya, J. Keeling, and ters A 235, 569 (1997).
B. L. Lev, Sign-Changing Photon-Mediated Atom Inter- [71] L. Diósi, N. Gisin, and W. T. Strunz, Non-Markovian
actions in Multimode Cavity Quantum Electrodynam- quantum state diffusion, Phys. Rev. A 58, 1699 (1998).
ics, Phys. Rev. Lett. 122, 193601 (2019). [72] W. T. Strunz, L. Diosi, and N. Gisin, Open system
[53] Y. Guo, V. D. Vaidya, R. M. Kroeze, R. A. Lunney, dynamics with non-Markovian quantum trajectories,
B. L. Lev, and J. Keeling, Emergent and broken sym- Phys. Rev. Lett. 82, 1801 (1999).
metries of atomic self-organization arising from Gouy [73] L. Diósi, Non-Markovian Continuous Quantum Mea-
phase shifts in multimode cavity QED, Phys. Rev. A surement of Retarded Observables, Phys. Rev. Lett.
99, 053818 (2019). 100, 080401 (2008).
[54] E. T. Jaynes and F. W. Cummings, Comparison of [74] L. Diósi, Erratum: Non-Markovian Continuous Quan-
quantum and semiclassical radiation theories with ap- tum Measurement of Retarded Observables [Phys. Rev.
plication to the beam maser, Proceedings of the IEEE Lett. 100, 080401 (2008)], Phys. Rev. Lett. 101, 149902
51, 89 (1963). (2008).
[55] V. Gorini, A. Kossakowski, and E. C. G. Sudarshan, [75] M. W. Jack and M. J. Collett, Continuous measurement
Completely positive dynamical semigroups of N-level and non-Markovian quantum trajectories, Phys. Rev. A
systems, J. Math. Phys. 17, 821 (1976). 61, 062106 (2000).
[56] G. Lindblad, On the generators of quantum dynamical [76] H. M. Wiseman and J. M. Gambetta, Pure-State Quan-
semigroups, Commun. Math. Phys. 48, 119 (1976). tum Trajectories for General Non-Markovian Systems
[57] R. Palacino and J. Keeling, Atom-only theories for u(1) Do Not Exist, Phys. Rev. Lett. 101, 140401 (2008).
symmetric cavity-qed models, Phys. Rev. Research 3, [77] J. Gambetta and H. M. Wiseman, Interpretation of
L032016 (2021). non-Markovian stochastic Schrödinger equations as a
[58] F. Damanet, A. J. Daley, and J. Keeling, Atom-only hidden-variable theory, Phys. Rev. A 68, 062104 (2003).
descriptions of the driven-dissipative Dicke model, Phys. [78] S. Krönke and W. T. Strunz, Non-Markovian quantum
Rev. A 99, 033845 (2019). trajectories, instruments and time-continuous measure-
[59] Y. Tanimura, Stochastic Liouville, Langevin, ments, Journal of Physics A: Mathematical and Theo-
Fokker–Planck, and Master Equation Approaches retical 45, 055305 (2012).
to Quantum Dissipative Systems, Journal of the [79] D. Suess, A. Eisfeld, and W. T. Strunz, Hierarchy of
Physical Society of Japan 75, 082001 (2006). Stochastic Pure States for Open Quantum System Dy-
[60] Y. Tanimura, Reduced hierarchical equations of motion namics, Phys. Rev. Lett. 113, 150403 (2014).
in real and imaginary time: Correlated initial states [80] R. Hartmann and W. T. Strunz, Exact Open Quantum
and thermodynamic quantities, The Journal of Chemi- System Dynamics Using the Hierarchy of Pure States
cal Physics 141, 044114 (2014). (HOPS), Journal of Chemical Theory and Computation
[61] Y. Tanimura and R. Kubo, Two-Time Correlation Func- 13, 5834 (2017), pMID: 29016126.
tions of a System Coupled to a Heat Bath with a [81] D. Suess, W. T. Strunz, and A. Eisfeld, Hierarchical
Gaussian-Markoffian Interaction, Journal of the Phys- Equations for Open System Dynamics in Fermionic and
ical Society of Japan 58, 1199 (1989). Bosonic Environments, J. Stat. Phys. 159, 1408 (2015).
[62] Y. Tanimura, Numerically “exact” approach to open [82] Z. Tang, X. Ouyang, Z. Gong, H. Wang, and J. Wu, Ex-
quantum dynamics: The hierarchical equations of mo- tended hierarchy equation of motion for the spin-boson
tion (HEOM), The Journal of Chemical Physics 153, model, The Journal of Chemical Physics 143, 224112
020901 (2020). (2015).
[63] K. Nakamura and Y. Tanimura, Hierarchical [83] F. Dimer, B. Estienne, A. S. Parkins, and H. J.
Schrödinger equations of motion for open quantum Carmichael, Proposed realization of the Dicke-model
dynamics, Phys. Rev. A 98, 012109 (2018). quantum phase transition in an optical cavity QED sys-
[64] A. Shabani, J. Roden, and K. B. Whaley, Continuous tem, Phys. Rev. A 75, 013804 (2007).
Measurement of a Non-Markovian Open Quantum Sys- [84] H. J. Groenewold, A problem of information gain by
tem, Phys. Rev. Lett. 112, 113601 (2014). quantal measurements, International Journal of Theo-
[65] W. Jiang, F.-Z. Wu, and G.-J. Yang, Non-Markovian retical Physics 4, 327 (1971).
entanglement dynamics of open quantum systems with [85] G. Vidal and R. F. Werner, Computable measure of
continuous measurement feedback, Phys. Rev. A 98, entanglement, Phys. Rev. A 65, 032314 (2002).
052134 (2018). [86] P.-P. Zhang, C. D. B. Bentley, and A. Eisfeld, Flexi-
[66] D. Walls and G. Milburn, Quantum Optics (Springer ble scheme to truncate the hierarchy of pure states, J.
Berlin Heidelberg, 2008). Chem. Phys. 148, 134103 (2018).
[67] J. Gambetta and H. M. Wiseman, Non-Markovian [87] P. Warszawski and H. M. Wiseman, Adiabatic elimi-
stochastic Schrödinger equations: Generalization to nation in compound quantum systems with feedback,
real-valued noise using quantum-measurement theory, Phys. Rev. A 63, 013803 (2000).
Phys. Rev. A 66, 012108 (2002). [88] L. K. Thomsen, S. Mancini, and H. M. Wiseman, Con-
[68] J. Piilo, S. Maniscalco, K. Härkönen, and K.-A. Suomi- tinuous quantum nondemolition feedback and uncondi-
nen, Non-Markovian Quantum Jumps, Phys. Rev. Lett. tional atomic spin, J. Phys. B: At. Mol. Opt. Phys. 35,
100, 180402 (2008). 4937 (2002).
 16

[89] A. Sørensen, L.-M. Duan, J. I. Cirac, and P. Zoller, ics, Phys. Rev. B 97, 205303 (2018).
Many-particle entanglement with Bose–Einstein con- [98] J. Schachenmayer, C. Genes, E. Tignone, and
densates, Nature 409, 63 (2001). G. Pupillo, Cavity-Enhanced Transport of Excitons,
[90] X. Wang, Spin squeezing in nonlinear spin-coherent Phys. Rev. Lett. 114, 196403 (2015).
states, J. Opt. B: Quantum Semiclassical Opt. 3, 93 [99] D. Hagenmüller, J. Schachenmayer, S. Schütz,
(2001). C. Genes, and G. Pupillo, Cavity-Enhanced Transport
[91] X. Gao, J. Ren, A. Eisfeld, and Z. Shuai, Non- of Charge, Phys. Rev. Lett. 119, 223601 (2017).
Markovian Stochastic Schrödinger Equation: Matrix [100] C. Maier, T. Brydges, P. Jurcevic, N. Trautmann,
Product State Approach to the Hierarchy of Pure C. Hempel, B. P. Lanyon, P. Hauke, R. Blatt, and C. F.
States, arXiv (2021). Roos, Environment-Assisted Quantum Transport in a
[92] A. V. Bezvershenko, C.-M. Halati, A. Sheikhan, C. Kol- 10-qubit Network, Phys. Rev. Lett. 122, 050501 (2019).
lath, and A. Rosch, Dicke Transition in Open Many- [101] J. B. Curtis, Z. M. Raines, A. A. Allocca, M. Hafezi,
Body Systems Determined by Fluctuation Effects, Phys. and V. M. Galitski, Cavity Quantum Eliashberg En-
Rev. Lett. 127, 173606 (2021). hancement of Superconductivity, Phys. Rev. Lett. 122,
[93] M. Buchhold, Y. Minoguchi, A. Altland, and S. Diehl, 167002 (2019).
Effective Theory for the Measurement-Induced Phase [102] S. Uchino, M. Ueda, and J.-P. Brantut, Universal noise
Transition of Dirac Fermions, Phys. Rev. X 11, 041004 in continuous transport measurements of interacting
(2021). fermions, Phys. Rev. A 98, 063619 (2018).
[94] T. Müller, S. Diehl, and M. Buchhold, Measurement- [103] V. Vuletić, H. W. Chan, and A. T. Black, Three-
induced dark state phase transitions in long-ranged dimensional cavity Doppler cooling and cavity sideband
fermion systems, arXiv (2021). cooling by coherent scattering, Phys. Rev. A 64, 033405
[95] E. Fiorelli, M. Marcuzzi, P. Rotondo, F. Carollo, and (2001).
I. Lesanovsky, Signatures of associative memory behav- [104] J. Zeiher, J. Wolf, J. A. Isaacs, J. Kohler, and
ior in a multimode dicke model, Phys. Rev. Lett. 125, D. M. Stamper-Kurn, Tracking Evaporative Cooling of
070604 (2020). a Mesoscopic Atomic Quantum Gas in Real Time, Phys.
[96] S. Schütz, J. Schachenmayer, D. Hagenmüller, G. K. Rev. X 11, 041017 (2021).
Brennen, T. Volz, V. Sandoghdar, T. W. Ebbesen, [105] F. O. Prado, F. S. Luiz, J. M. Villas-Bôas, A. M. Al-
C. Genes, and G. Pupillo, Ensemble-Induced Strong calde, E. I. Duzzioni, and L. Sanz, Atom-mediated ef-
Light-Matter Coupling of a Single Quantum Emitter, fective interactions between modes of a bimodal cavity,
Phys. Rev. Lett. 124, 113602 (2020). Phys. Rev. A 84, 053839 (2011).
[97] D. Hagenmüller, S. Schütz, J. Schachenmayer, [106] Y.-L. Dong, X.-B. Zou, S.-L. Zhang, S. Yang, C.-F. Li,
C. Genes, and G. Pupillo, Cavity-assisted mesoscopic and G.-C. Guo, Cavity-QED-based phase gate for pho-
transport of fermions: Coherent and dissipative dynam- tonic qubits, Journal of Modern Optics 56, 1230 (2009).