Professional Documents
Culture Documents
Sieberer 2016 Rep. Prog. Phys. 79 096001
Sieberer 2016 Rep. Prog. Phys. 79 096001
This content has been downloaded from IOPscience. Please scroll down to see the full text.
View the table of contents for this issue, or go to the journal homepage for more
Download details:
IP Address: 147.122.128.194
This content was downloaded on 20/06/2017 at 09:55
Hybrid Exciton-Polaritons in a Bad Microcavity Containing the Organic and Inorganic Quantum Wells
Liu Yu-Xi and Sun Chang-Pu
Review
E-mail: lukas.sieberer@gmail.com
Abstract
Recent experimental developments in diverse areasranging from cold atomic gases to light-
driven semiconductors to microcavity arraysmove systems into the focus which are located
on the interface of quantum optics, many-body physics and statistical mechanics. They share
in common that coherent and drivendissipative quantum dynamics occur on an equal footing,
creating genuine non-equilibrium scenarios without immediate counterpart in equilibrium
condensed matter physics. This concerns both their non-thermal stationary states and their
many-body time evolution. It is a challenge to theory to identify novel instances of universal
emergent macroscopic phenomena, which are tied unambiguously and in an observable
way to the microscopic drive conditions. In this review, we discuss some recent results in
this direction. Moreover, we provide a systematic introduction to the open system Keldysh
functional integral approach, which is the proper technical tool to accomplish a merger of
quantum optics and many-body physics, and leverages the power of modern quantum field
theory to driven open quantum systems.
Keywords: non-equilibrium field theory, Keldysh functional integrals, driven open quantum
systems, ultracold atoms, exciton-polariton condensates, microcavity arrays
[9, 10], or driven Rydberg ensembles are prepared [11]; light- microscopic origin is of key importance despite a dominantly
driven semiconductor heterostructures realize BoseEinstein dissipative dynamics.
condensation of exciton-polaritons [12, 13]; coupled micro- Summarizing, this is a fledging topical area, where first
cavity arrays are pushed to a regime demonstrating both results underpin the promise of these systems to exhibit genu-
strong coupling of light and matter [14] and scalability [15]; inely new physics. In the remainder of the introduction, we will
large ensembles of trapped ions implement varieties of driven discuss in some more detail the three major challenges which
spin models [16, 17]. All those systems are genuinely made emerge in these systems. The first one concerns the identifi-
of quantum ingredients, and share in common that coher- cation of novel macroscopic many-body phenomena, which
ent and drivendissipative dynamics occur on equal footing. witness the microscopic driven open nature of such quantum
This creates close ties to typical setups in quantum optics. systems. Second, we anticipate the theoretical machinery,
But on the other hand, they exhibit a continuum of degrees which allows us to perform the transition from micro- to mac-
of freedom, characteristic for many-body physics5. Systems rophysics in a non-equilibrium context in practice. Third, we
located at this new interface are not guaranteed to thermal- describe some representative experimental platforms, which
ize, due to the absence of energy conservation and the result- motivate the theoretical efforts, and in which the predictions
ing breaking of detailed balance by the external drive at the can be further explored.
microscale. They rather converge to non-equilibrium station-
ary states of matter, creating scenarios without counterpart
1.1. New phenomena
in condensed, equilibrium matter. This rules out conventional
theoretical equilibrium concepts and techniques to be used, As pointed out above, one of the key goals of the research
and calls for the development of new theoretical tools. The reviewed here is the identification of new macroscopic many-
physical framework sparks broader theoretical questions on body phenomena, which can be uniquely traced back to the
the existence of new phases of bosonic [18, 19] and fermi- microscopic driven open nature of such systems, and do not
onic [2022] matter, the nature of phase transitions in such have an immediate counterpart in equilibrium systems.
driven systems [10, 2326], and the observable consequences The driven nature common to the systems considered
of quantum mechanics at the largest scales [27, 28]. Beyond here can always be associated to the fact that the underly-
stationary states [29], a fundamental challenge is set by the ing microscopic Hamiltonian is time dependent, with a time
time evolution of interacting quantum systems, which is cur- dependence relating to external driving fields such as lasers.
rently explored theoretically [3036] and experimentally in When such an ensemble (i) in addition has a natural parti-
cold atomic [3741] and photonic systems [42]. A key goal tion into a system and a batha continuum of modes well
is to identify universal dynamical regimes that hold beyond approximated by harmonic oscillators with short memory,
specific realizations or precise initial conditions. Combining (ii) the systembath coupling is weak compared to a typical
idealized closed system evolution with the intrinsic open sys- energy scale of the system Hamiltonian and (iii) linear in
tem character of any real world experiment takes this setting the bath creation and annihilation operators (so that they can
to the next stage, and exhibits emergent dynamics markedly be integrated out straightforwardly), then an effective (still
different from closed systems both for short [43, 44] and long microscopic) description in terms of combined Hamiltonian
evolution times [4551]. and drivendissipative Markovian quantum dynamics of the
The interplay of coherent and drivendissipative dynam- system ensues. The system dynamics obtains by tracing
ics can be a natural consequence of the driving necessary to out the bath variables.
maintain a certain many-body state. Going one step further, The resulting effective microdynamics is non-equilib-
it is possible to exploit and further develop the toolbox of rium, in a sense sharpened in section 2.4.1. This not only
quantum optics for the drivendissipative manipulation of concerns the time evolution, but also holds for the non-equi-
many-body systems. Recently, it has been recognized that the librium stationary states. More precisely, the above situation
concept of dissipative state preparation in quantum optics implies an explicit breaking of detailed balance, since the
[52, 53] can be developed into a many-body context, both system energy is not conserved due to the explicitly time-
theoretically [5457] and experimentally [5860]. Suitably dependent drive.
tailored dissipation then does not necessarily act as an adver- What do we actually mean by detailed balance and ther-
sary to subtle quantum mechanical correlations such as phase mal equilibrium? In an operational sense, the principle of
coherence or entanglement [6166]. On the contrary, it can detailed balance states that there is a partition invariance for
even create these correlations, and dissipation then represents the temperature (or, more generally, the noise level) present
the dominant resource of many-body dynamics. In particular, in the system: an arbitrary bipartition of the system can be
even topologically ordered states in spin systems [67] or of chosen, one part can be traced out, and the resulting subsys-
fermionic matter [68, 69] can be induced dissipatively ([56] tem will be at the same temperature (noise level) as the total
and [7072] respectively). These developments open up a new system. This partition invariance is the condition for a glob-
arena for many-body physics, where the quantum mechanical ally well-defined temperature characteristic for systems in
thermal equilibrium. More formally, thermal equilibrium can
5
This includes the cases of extended spatial continuum and lattice systems.
be detected by means of so-called fluctuation-dissipation rela-
In both cases, a continuum of momentum modes obtains, underlying the tions (FDRs). These connect the two fundamental observables
characteristics of many-body problems at long wavelength. in physical systemscorrelation and response functions (see
2
Rep. Prog. Phys. 79 (2016) 096001 Review
section2.2). In the case of thermal equilibrium, the connec- less developed, concerns the distinction between classical and
tion is dictated by the particle statistics alone. It is then given quantum effects. Again, although the quantum mechanical
by the Bose- and Fermi-distributions, respectively. Deviations description is necessary at a microscopic level, the persistence
from this universal form, which has only two free parameters of quantum effects at the macroscale is not guaranteed. This
(temperature and chemical potential, relating to the typical is mainly due to the Markovian noise level inherent to such
conserved quantities energy and particle number), provide a quantum systems. Nevertheless, systems with suitably engi-
necessary requirement for non-equilibrium conditions. neered drivendissipative dynamics show typical quantum
In non-equilibrium stationary states, no such general form mechanical phenomena such as phase coherence [54, 55, 61,
exists. We will encounter a concrete and simple example in 63, 64, 78], entanglement [5860, 62], or topological order
the context of the driven Dicke model (a cavity photon cou- [56, 66, 7072]. Especially, fermionic systems which do
pled to a collective spin) in section 3.1, where the form of not possess a classical limitare promising in this direction.
the FDR depends on the observable we are choosing (e.g. the
position or momentum correlations and responses). 1.2. Theoretical concepts and techniques
On the other hand, thermal FDRs can emerge at long wave-
length, even though the microscopic dynamics manifestly The development of theoretical tools needed to perform
breaks detailed balance [26]. In particular, in three dimen- the transition from micro- to macro-physics in driven open
sions and close to the critical point of drivendissipative quantum systems is still in progress, as a topic of current
Bose-condensation, a degeneracy of critical exponents indi- research. A reason for the preliminary status of the theory
cates a universal asymptotic thermalization, in the sense of an lies in the fact that two previously rather independent disci-
emergent thermal FDR. A similar phenomenology is observed plinesquantum optics and many-body physicsneed to be
in a disordered multimode extension of the Dicke model, see unified on a technical level.
section 3.2. This underlines the strongly attractive nature of Quantum optical systems are well described microscopi-
the thermal equilibrium fixed point at low frequencies. Still in cally in terms of Markovian quantum master equations, which
these systems, non-equilibrium conditions leave their traces treat coherent Hamiltonian and drivendissipative dynam-
in the dynamical response of the system, in terms of informa- ics on equal footing. To solve such equations both for their
tion that does not enter the FDR at leading order. For exam- dynamics and their stationary states, powerful techniques
ple, the critical behavior is characterized by a fine structure have been devised. This comprises efficient exact numerical
in a new and independent critical exponent, which measures techniques for small enough systems, such as the quantum
decoherence, and whose value distinguishes equilibrium and trajectories approach [48, 79, 80]. But it also includes ana-
non-equilibrium dynamics, see section4.2. lytical approaches such as perturbation theory for quantum
Instead of emergent thermal behavior indicating the fade- master equations, e.g. in the frame of the NakajimaZwanzig
out of non-equilibrium conditions upon coarse graining, the projection operator technique [81, 82], or mappings to P, W,
opposite behavior is also possible. For example, low dimen- or Q representations [83], casting the problem from a second
sional (d 2) bosonic systems at low noise level, such as quantized formulation into partial differential equations.
exciton-polaritons well above threshold, are not attracted to A characteristic feature of traditional quantum optical sys-
the equilibrium fixed point as their three dimensional counter- tems is the finite spacing of the few energy levels which play
parts, but rather flow to the non-equilibrium fixed point of the a role. When considering systems with a spatial continuum of
KardarParisiZhang [73] universality class, see section4.3. degrees of freedom instead, the energy levels become continu-
This can be interpreted as a universal and indefinite increase ous. This does not mean that the microscopic modelling in
of the non-equilibrium strength, which is triggered even if the terms of a quantum master equationis inappropriate: for the
violation of detailed balance at the microscopic level is very drivendissipative terms in such an equation, the assumption
small. of spatially independent dissipative processes (such as atomic
Universal non-equilibrium phenomena can also occur in the loss or spontaneous emission) is still valid as long as the emit-
time evolution of driven open systems. For example, intriguing ted wavelength of radiation is well below the spatial resolution
scaling laws describing algebraic decoherence [46], anoma- at the scale where the microscopic model is defined. Indeed,
lous diffusion [74], or glass-like behavior [45, 7577] have in this situation, destructive interference of radiation justifies
been identified in the long time asymptotics of driven spin the description of driven dissipation in terms of incoherent
systems close to the stationary states. Conversely, the short processes. However, under these circumstances the smallness
time behavior of driven open lattice bosons shows universal of a microscopic expansion parameter no longer guarantees
scaling laws directly witnessing the non-equilibrium drive, the smallness of the associated perturbative correction. Here
see section 5. This scaling can be related to a strongly pro- the reason is that in perturbation theory, one is summing over
nounced non-equilibrium shape of the time-dependent distri- intermediate states with propagation amplitudes down to the
bution function in the early stages of evolution, and be traced longest wavelengths. This can lead to infrared divergences
back to conservation laws of the drivendissipative generator in naive perturbation theorya circumstance that found its
of dynamics. physical interpretation and technical remedy in equilibrium
The above discussion mainly focuses on the difference in terms of the renormalization group [84, 85]. We emphasize
between equilibrium and non-equilibrium systems on the mac- that it is precisely this situation of long wavelength dominance
roscopic level of observation. Another direction, still much which underlies much of the universality, i.e. insensitivity to
3
Rep. Prog. Phys. 79 (2016) 096001 Review
microscopic details, which is encountered when moving from non-equilibrium Keldysh functional integral is increased by
the microscale to macroscopic observables in many-particle the characteristic doubling of degrees of freedom compared
systems. to thermal equilibrium. However, it should be noted that it
The modern framework to understand many-particle prob- is precisely this feature which relates the Keldysh functional
lems in thermodynamic equilibrium is in terms of the functional integral much closer to real time (von Neumann) evolution
integral formulation of quantum field theory. The spectrum familiar from quantum mechanics. We will demonstrate this
of its application covers a remarkable range of energy scales, in section 2.1, making the Markovian action a rather intui-
from ultracold atomic gases to condensed matter systems tive object to work with. Furthermore, when properly har-
with strong correlations to quantum chromodynamics and the nessing symmetries, the complexity of calculations can often
quantum theory of gravity. It provides us with a well-developed be made comparable to thermodynamic equilibrium. Most
toolbox of techniques, such as diagrammatic perturbation importantly, however, the powerful toolbox of quantum field
theory including sophisticated resummation schemes. But it theory is opened in this way. It is thus possible to leverage the
also encompasses non-perturbative approaches, which often full power of sophisticated techniques from equilibrium field
capitalize on the flexibility of the functional integral when it theory to driven open many-body systems.
comes to picking the relevant degrees of freedom for a given Relation to classical dynamical field theoriesThe
problem. This is the challenge of emergent phenomena, whose quantum mechanical Keldysh formulation reduces to the
solution typically is strongly scale dependent [86]. Familiar so-called Martin-Siggia-Rose (MSR) functional integral in
examples include an efficient description of emergent Cooper the (semi-) classical limit, in turn equivalent to a stochastic
pair or molecular degrees of freedom in interacting fermion Langevin equationformulation [102]. This statement will be
systems, or vortices which conveniently parameterize the made precise and discussed in section2.3. A large amount of
long-wavelength physics of interacting bosons in two dimen- work has been dedicated to this limit in the past. On the one
sions. The description of the change of physics with scale was hand, this concerns dynamical aspects of equilibrium statisti-
given its mathematical foundation in terms of the renormali- cal mechanics, and we refer to the classic work by Hohenberg
zation group already mentioned above, yet another tool devel- and Halperin [1] for an overview. This work shows that,
oped and most clearly formulated in a functional language. while the static universal critical behavior is determined by
Finally, the functional integral based on a single scalar quanti- the symmetries and the dimensionality of the problem, the
tythe systems action, which encodes all the dynamics dynamical critical behavior is sensitive to additional dynami-
on the microscopic scaleis a convenient framework when cal conservation laws. This leads to a fine structure, defining
it comes to the classification of symmetries and associated dynamical universality classes which are denoted by models
conservation laws, and their use in devising approximation AJ [1]. These models also provide a convenient framework
schemes respecting them. to describe the statistics of work, summarized in Jarzynskis
In short, while the driven open many-body systems are well work theorem [104, 105] and Crooks relation [106]. On the
described by microscopic master equations, the traditional other hand, non-equilibrium situations are captured as well.
techniques of quantum optics cannot be used efficientlyat Here, we highlight in particular genuine non-equilibrium uni-
least not in the case where the generic complications of many- versality classes, which are not smoothly connected to the
body systems start to play a role. Conversely, their driven open equilibrium models. Among them is the problem of reaction-
character makes it impossible to approach these problems in diffusion models [107] including directed percolation [4],
the framework of equilibrium many-body physics. This situa- which is relevant to certain chemical processes (for an imple-
tion calls for a merger of the disciplines of quantum optics and mentation of this universality class with driven Rydberg gases,
many-body physics on a technical level. On the numerical side, see [50]). Another key example is surface growth, described
progress has been made in one spatial dimension recently by by the KardarParisiZhang equation [73], giving rise to a
combining the method of quantum trajectories with powerful non-equilibrium universality class which is at the heart of
density matrix renormalization group algorithms [8790], see driven phenomena such as the growth of bacterial colonies or
[48] for an excellent review on the topic. For more analytical the spreading of fire.
approaches, the Keldysh functional or path integral [9196] In the same class of approaches ranges the so-called Doi-
is ideally suited (but see [97] for a recent systematic pertur- Peliti functional integral [108, 109], which is a functional rep-
bative approach to lattice Lindblad equationswith extensions resentation of classical master equations, and may be viewed
to sophisticated resummation schemes [98], and [99101] for as an MSR theory with a specific, highly non-linear appearance
advanced variational techniques). Conceptually, the latter cap- of the field variables. It reduces to the conventional MSR form
tures the most general situation in many-body physicsthe in a leading order Taylor expansion of the field non-linearities.
dynamics of a density matrix under an arbitrary temporally A comprehensive overview of models, methods, and physical
local generator of dynamics. We refer to [102, 103] for an phenomena in the (semi-)classical limit is provided in [5].
introduction to the Keldysh functional integral. In our context, We also note that the usual mean field theory, where cor-
it can be derived by a direct functional integral quantization relation functions are factorized into products of field ampl-
of the Markovian quantum master equation. This procedure itudes and which is often used as an approximation to the
results in a simple translation tablefrom the master equationto quantum master equation in the literature, corresponds to a
the key player in the associated Keldysh functional integral, further formal simplification of the semi-classical limit. Here
the Markovian action. At first sight, the complexity of the the effects of noise are neglected completely. Conversely, the
4
Rep. Prog. Phys. 79 (2016) 096001 Review
semi-classical limit represents a systematic extension of mean has recently been shifted towards loading more and more atoms
field theory, which includes the Markovian noise fluctuations. inside a cavity. Thereby, not only single particle dynamics in
This level of approximation is referred to as optimal path strong radiation fields can be probed, but also collective, mac-
approximation in the literature on MSR functional integrals roscopic phenomena, which are driven by light-matter interac-
[5, 102]. tions. For an excellent review on this topic, covering important
In many cases, even though the microscopic description is experimental and theoretical developments, see [10]. In the
in terms of a quantum master equation, at long distances the experiments, cold atoms are loaded inside an optical or micro-
Keldysh field theory reduces to a semi-classical MSR field wave cavity, for which the coherent interaction between the
theory. The reason is the finite Markovian noise level that such atomic internal states and a single cavity mode dominates over
systems exhibit generically, as explained in section2.3. The dissipative processes [117]. The atoms absorb and emit cavity
prefix semi refers to the fact that phase coherence may still photons, thereby changing their internal states. Due to this pro-
persist in such circumstancesthe situation is comparable to cess, the spatial modulation of the intra-cavity light field induces
a BoseEinstein condensate at finite temperature. Recently a coupling of the cavity photons to the atomic internal state as
however, situations have been identified where the drastic well as their motional degree of freedom [118]. In this way,
simplifications of the semi-classical limit do not apply. In part- cavity photons mediate an effective atomatom interaction,
icular, this occurs in systems with dark statespure quantum which leads to a back-action of a single atom on the motion of
states which are dynamical fixed points of drivendissipative other atoms inside the cavity. This cavity mediated interaction
evolution [54, 110]. In these cases, classical dynamical field is long-ranged in space and represents the source of collective
theories are inappropriate, which calls for the development of effects in cold atomic clouds within cavity QED experiments.
quantum dynamical field theories [28]. These developments One hallmark of collective dynamics in cavity QED has been
are just in their beginnings. the observation of self-organization of a BoseEinstein conden-
In this review, we concentrate on systems composed of sate in an optical cavity. This is accompanied by a Dicke phase
bosonic degrees of freedom. However, it is also possible to transition via breaking of a discrete Z2-symmetry of the under-
address spin systems in terms of functional integrals [111], lying model [9], [119121].
and simple models systems have been analyzed in this way, Although the dominant dynamics in these systems is
see [112] and section3.1. More sophisticated approaches to coherent, there are dissipative effects which cannot be dis-
spin systems were elaborated in the context of multimode carded. These are the loss of cavity photons due to imperfec-
optical cavities in [113], and systematically for various sym- tions in the cavity mirrors, or spontaneous emission processes
metries for lattice systems in [114]. Fermi statistics is also of atoms, which emit photons into transverse modes. These
conveniently implemented in the functional integral form- effects modify the dynamics of the system on the longest time
ulation. This is relevant, e.g. for driven open Fermi gases in scales. Therefore, they become relevant for macroscopic phe-
optical cavities [21, 22, 115] or lattices [116], or dissipatively nomena, such as phase transitions and collective dynamics.
stabilized topological fermion matter [7072]. For instance, dissipative effects have been shown theoretically
[122, 123] and experimentally [124] to modify the critical
exponent of the Dicke transition compared to its zero temper-
1.3. Experimental platforms
ature value. This illustrates that for the analysis of collective
The progress in controlling, manipulating, detecting and scal- phenomena in cavity QED experiments, the dissipative nature
ing up driven open quantum systems to many-body scenarios of the system has to be taken into account properly [112, 125].
has been impressive over the last decade. Here we sketch the Typically, for a cavity field which has a very narrow spec-
basic physics of three representative platforms, and indicate trum, the atomic internal degrees of freedom can be reduced
the relevant microscopic theoretical models in the frame of the to two internal states, whose transitions are nearly resonant to
Markovian quantum master equation. In later sections, we will the photon frequency. The operators acting on these two inter-
translate this physics into the language of the Keldysh func- nal states can be represented by Pauli matrices, making them
tional integral. For each of these platforms, excellent reviews equivalent to a spin-1/2 degree of freedom. A very important
exist, which we refer to at the end of section1.4 together with model in the framework of cavity QED is the Dicke model
further literature on open systems. The purpose of this sec- [126128], which describes N atoms (i.e. two-level systems)
tionis to give an overview only, and to put the respective plat- coupled to a single quantized photon mode. This is expressed
forms into their overarching context as driven open quantum by the Hamiltonian (here and in the following we set ! = 1)
systems with many degrees of freedom.
N N
z g
1.3.1. Cold atoms in an optical cavity, and microcavity arrays:
HD = 0 aa +
2
iz + N
ix(a + a). (1)
i=1 i=1
drivendissipative spin-boson models. Cavity quantum elec-
trodynamics (cavity QED), with its focus on strong light-matter Here, 0 is the photon frequency, z = | 11| | 00| rep-
interactions, is a growing field of research, which has experi- resents the splitting of the two atomic levels with energy
enced several groundbreaking advances in the past few years. difference z. x = | 01| + | 10| describes the coherent
Historically, these systems were developed as few or single excitation and de-excitation of the atomic state proportional
atom experiments, detecting the radiation properties of atoms to the atomphoton coupling strength g. The Dicke model
that are strongly coupled to a quantized light field. The focus features a discrete Z2 Ising symmetry: it is invariant under the
5
Rep. Prog. Phys. 79 (2016) 096001 Review
transformation (a, a, ix) ! (a, a, ix). In the thermody- therefore to a dephasing of the atoms, described by additional
namic limit, for N , it features a phase transition, which Lindblad operators L i = iz [130]. However, these losses are at
spontaneously breaks the Ising symmetry. Crossing the trans- least three orders of magnitude smaller than the cavity decay
ition, the system enters a superradiant phase, characterized rate and typically not considered [9, 125]. The basic model
by finite expectation values a 0, ix 0. This describes for cavity QED with cold atoms is therefore represented by
condensation of the cavity photons, i.e. the formation of a the Dicke model with dissipation, formulated in terms of the
macroscopically occupied, coherent intra-cavity field, and a master equation (2). Its dynamics is discussed in section 3,
ferromagnetic ordering of the atoms in the x-direction. including the Dicke superradiance transition.
Although the Dicke model is a standard model for cav- The Dicke model Hamiltonian takes the form
ity QED experiments in the ultra-strong coupling limit HD = Hc + Hs + Hcs, where Hc,s,cs represent the bosonic cav-
ity, spin, and spin-boson sectors, respectively. There are many
N g > z 0, it has been realized only very recently in cold
directions to go beyond the Dicke model with single collective
atom experiments, where an entire BEC was placed inside
spin, still keeping the basic feature of coupling spin to boson
an optical cavity [9]. It has been shown that this setup maps
(cavity photon) degrees of freedom. One directionrelevant
to a Dicke model, with a collective spin degree of freedom
to future cold atom experimentsis to consider multimode
[120]. Here, the detuning of the pump laser was chosen such
cavities instead of a single one. In particular, in conjunction
that the atoms effectively remain in the internal ground state,
with quenched disorder, intriguing analogies to the physics
but acquire a characteristic recoil momentum when scatter-
of quantum glasses can be established in this way [113, 131].
ing with a cavity photon. This scattering creates a collective,
Here, the global coupling of all spins to a single mode gi g
motionally excited state, which replaces the role of an individ-
is replaced by random couplings gi,, where the index now
ual, internally excited atom. The experimental realization of a
refers to a collection of cavity modes. The many-body physics
superradiance transition in the Dicke model is usually inhib- of such an open system is discussed in section3.
ited, since the required coupling strength by far exceeds the The basic building block of the Dicke model is the spin-
available value of the atomic dipole coupling. However, for boson term of the form Hsc (a + a) x, i.e. a Rabi type non-
the BEC in the cavity, the energy scales of the excited modes linearity that preserves the Z2 symmetry of Hc,s. In the context
are much lower than the optical scale of the atomic modes. In of circuit quantum electrodynamics, a natural many-body
this way, the superradiance transition indeed became exper- generalization of Hamiltonians with a spin-boson interaction
imentally accessible. This was inspired by a theoretical pro- is to consider entire arrays of microcavities (instead of con-
posal using two balanced Raman channels between different sidering many modes within a single cavity). These cavities
internal atomic states inside an optical cavity, which reduced can be coupled to each other by single photon tunnelling pro-
the effective level splitting of the internal states to much lower cesses between adjacent cavities, giving rise to Hubbard-type
energy scales [129].
hopping terms J ai aj + h.c., where i, j now label the spatial
In addition to the unitary dynamics represented by the
index of the cavities. In cirquit QED, strong non-linearities
Dicke model, the cavity is subject to permanent photon loss
can be generated, e.g. by coupling to adjacent qubits made of
due to imperfections in the cavity mirrors. For high finesse
Cooper pair boxes [132, 133]. This gives rise to many-body
cavities, the coupling of the intra cavity photons to the sur-
variants of the Rabi model [134], whose phase diagrams have
rounding vacuum radiation field is very weak, and the latter
been studied recently [135137]. Furthermore, for the imple-
can be eliminated in a Born-Markov approximation [81]. This
mentation of lattice Dicke models with large collective spins,
results in a Markovian quantum master equationfor the sys-
the use of hybrid quantum systems consisting of supercon-
tems density matrix
ducting cavity arrays coupled to solid-state spin ensembles
t = i[HD, ] + Ld , (2) have been proposed [138]. Spontaneous collective coherence
in drivendissipative cavity arrays has been studied in [139].
where is the density matrix for the intra cavity system and In many physical situations (away from the ultra-strong
Ld adds dissipative dynamics to the coherent evolution of the coupling limit), a form of the spin-cavity interaction alter-
Dicke model. For a vacuum radiation field, it is given by native to the Rabi term is more appropriate, with non-linear
1 building block Hcs a+ + a. In fact, this form results nat-
Ld = LL {LL , }.
(3) urally from the weak coupling rotating wave approximation of
2
a driven spin-cavity problem. For a single cavity mode and
The Lindblad operator L acting on the density matrix describes spin, the resulting model is the JaynesCummings model,
pure photon loss (L = a) with an effective loss rate ; the lat- which in contrast to the Dicke model possesses a continuous
ter depends on system specific parameters, but is typically the U(1) phase rotation symmetry under a eia, ei.
smallest scale in the master equation(2) [10]. Clearly, when such systems are driven coherently via a
In generic cavity experiments, there are also atomic spon- Hamiltonian Hd = (a + a) or suitable multimode generali-
taneous emission processes. The atoms scatter a laser or cav- zations thereof, both the U(1) and even the Z2 symmetries of
ity photon out of the cavity, and this represents a source of the above models are broken explicitly. Coherent drive is usu-
decoherence. This process has been considered in [112], and ally the simplest way to compensate for unavoidable losses
leads to an effective decay rate of the atomic excited state and due to cavity leakage, although incoherent pumping schemes
6
Rep. Prog. Phys. 79 (2016) 096001 Review
7
Rep. Prog. Phys. 79 (2016) 096001 Review
value of the photon mass is orders of magnitude smaller than high-energy excitations. However, for the study of universal
the mass of the exciton, so that the dispersion of the latter long-wavelength behavior in section4, this degree of micro-
appears to be flat on the scale of figure1(b). scopic modeling is actually not required: indeed, any (pos-
Upon absorption of a photon by the semiconductor, an sibly simpler) model that possesses the relevant symmetries
exciton is generated. This process (and the reverse process of (see the discussion at the beginning of section 4) will yield
the emission of a photon upon radiative decay of an exciton) the same universal physics. Such a model can be obtained by
is described by describing incoherent pumping and losses of lower polaritons
by means of a Markovian master equation7:
dq
HC X = R (2 )2 (aX ,(q)aC,(q) + H.c.), (7) t = i[HLP, ] + Ld , (9)
where Ld encodes incoherent single-particle pumping and
where R is the rate of the coherent interconversion of pho-
losses, as well as two-body losses:
tons into excitons and vice versa. The quadratic Hamiltonian
(4) can be diagonalized by introducing new modesthe lower
and upper exciton-polaritons, LP, (q) and UP, (q) respec-
Ld = dx(pD[ (x)] + l D[(x)] + 2ud D[(x)2]),
(10)
tively, which are linear combinations of photon and exci-
ton modes. The dispersion of lower and upper polaritons is where
depicted in figure 1(b). In the regime of strong light-matter 1
D[L ] = LL {L L , } (11)
coupling, which is reached when R is larger than both the 2
rate at which photons are lost from the cavity due to mirror
reflects the Lindblad form, and p, l, 2ud are the rates of sin-
imperfections and the non-radiative decay rate of excitons, it
gle-particle pumping, single-particle loss, and two-body loss,
is appropriate to think of exciton-polaritons as the elementary
respectively. The inclusion of the non-linear loss term ensures
excitations of the system.
saturation of the pumping. An analogous mechanism is imple-
In experiments, it is often sufficient to consider only lower
mented in the above-mentioned GrossPitaevskii description.
polaritons in a specific spin state, and to approximate the
More precisely, in the spirit of universality, the above quantum
dispersion as parabolic [13]. Interactions between exciton-
master equation(9) and the abovementioned phenomenologi-
polaritons originate from various physical mechanisms, with a
cal model reduce to precisely the same low frequency model
dominant contribution stemming from the screened Coulomb
for bosonic degrees of freedom upon taking the semiclassical
interactions between electrons and holes forming the excitons.
limit in the Keldysh path integral associated to equation(9)
Again, in the low-energy scattering regime, this leads to an
(see section 2.3 for its implementation), and integrating out
effective contact interaction between lower polaritons. As a
the upper polariton reservoir in the phenomenological model.
result, the low-energy description of lower polaritons takes
the form (in the following we drop the subscript indices in
1.3.3. Cold atoms in optical lattices: heating dynamics. In
LP, ) [13]
recent years, experiments with cold atoms in optical lattices
2 have shown remarkable progress in the simulation of many-
dx (x)LP 2mLP (x) + uc(x)2(x)2 .
0 body model systems both in and out of equilibrium. A part-
HLP =
icular strength of cold atom experiments is the unprecedented
(8) tuneability of model parameters, such as the local interaction
While this Hamiltonian is quite generic and arises also, e.g. in strength and the lattice hopping amplitude. This becomes pos-
cold bosonic atoms in the absence of an external potential, the sible by, e.g. manipulation of the lattice laser and external
peculiarity of exciton-polaritons is that they are excitations magnetic fields. It comes along with a very weak coupling
with relatively short lifetime. In turn, this necessitates con- of the system to the environment, such that the dynamics can
tinuous replenishment of energy in the form of laser driving often be seen as isolated on relevant time scales for typical
in order to maintain a steady population. In figure 1(b), we measurements of static, equilibrium correlations. However,
consider the case in which the excitation laser is tuned to ener- more and more experiments start to investigate the realm of
gies well above the lower polariton band. The high-energy non-equilibrium phenomena with cold atoms, e.g. by letting
excitations thus created are deprived of their excess energy systems prepared in a non-equilibrium initial condition relax
via phononpolariton and stimulated polaritonpolariton scat- in time towards a steady state [37, 39], [163165]. With these
tering. Eventually, they accumulate at the bottom of the lower experiments, time scales are reached for which the dissipative
polariton band. As a consequence of multiple scattering pro- coupling to the environment becomes visible in experimental
cesses, the coherence of the incident laser field is quickly lost, observables. Such dissipation may even hinder the system
and the effective pumping of lower polaritons is incoherent. from relaxation towards a well defined steady state.
A phenomenological model for the dynamics of the lower A relevant example of a dissipative coupling is decoher-
polariton field, which accounts for both the coherent dynamics ence of an atomic cloud induced by spontaneous emission of
generated by the Hamiltonian (8) and the drivendissipative atoms in the lattice [166, 167]. In this way, the many-body
one described above, was introduced in [159]. It involves a 7
While this approach captures the universal behavior, we note that
dissipative GrossPitaevskii equation for the lower polari- non-Markovian effects can be of key importance for other properties
ton field that is coupled to a rate equationfor the reservoir of [160162].
8
Rep. Prog. Phys. 79 (2016) 096001 Review
system is heated up, and therefore driven away from the low
entropy state in which it was prepared initially. A detailed dis-
cussion of the microscopic physics and its long time dynam-
ics can be found in [4547], see also the review article [48].
For bosonic atoms in optical lattices, the coherent dynamics is
described by the BoseHubbard Hamiltonian [168, 169]
U Figure 2. Illustration of the coherent (a) and the incoherent (b)
HBH = J bl bm + 2
nl(nl 1), (12) contribution to the dynamics stemming from the coupling of the
l , m l
atoms to a laser with amplitude || and large detuning . (a) Via
which models bosonic atoms in terms of the creation and anni- the AC Stark effect, stimulated absorption and emission lead to
||2
hilation operators [bl , bm] = lm in the lowest band of a lattice a coherent periodic potential with amplitude 4 . (b) Stimulated
with site indices l, m. The atoms hop between neighboring lat- absorption and subsequent spontaneous emission lead to effective
||2
tice sites with an amplitude J, and experience an on-site repul- decoherence of the atomic state with rate 4 , where is the
sion U. The lattice potential V(x), which leads to the second microscopic spontaneous emission rate.
quantized form of the BoseHubbard model, is created by the
superposition of counter-propagating laser beams in each spa- of the atomlaser coupling to the dynamics is illustrated in
tial dimension. The coherent laser field couples two internal figure2.
atomic states via stimulated absorption and emission, which The dissipative term in the master equation (15) leads to
leads to the single particle Hamiltonian an energy increase HBH(t ) t linear in time, and thus to
p 2 ()
x heating. Furthermore, it introduces decoherence in the num-
Hatom = z x , (13) ber state basis, i.e. it projects the local density matrix on its
2m 2 2
diagonal in Fock space and leads to a decrease of the coher-
where p is the atomic momentum operator, is the detuning ences in time [4548]. Starting from a low entropy state at
of the laser from the atomic transition frequency, and ()
x is t = 0, the heating leads to a crossover from coherence domi-
the laser field at the atomic position. For large detuning, the nated dynamics at short and intermediate times [43] to a
excited state of the atom can be traced out, which leads to the decoherence dominated dynamics at long times [4547]. In
lattice Hamiltonian one dimension, both regimes have been analyzed extensively
both numerically (with a focus on decoherence dominated
eff p 2 | ()
x |2
H atom = + . (14) dynamics) and analytically, and display several aspects of
2m 4 non-equilibrium universality, see section 5. Therefore, heat-
This describes a lattice potential for the ground state atoms ing in interacting lattice systems represents a crucial example
generated by the spatially modulated AC Stark shift. Adding for universality in out-of-equilibrium dynamics, which can be
an atomic interaction potential and expanding both the sin- probed by cold atom experiments.
gle particle Hamiltonian (14) and the interaction in terms of
Wannier states, the leading order Hamiltonian is the Bose 1.4. Outline and scope of this review
Hubbard model (12) [169].
In the semi-classical treatment of the atom-laser interac- The remainder of this review is split into two parts.
tion, spontaneous emission events are neglected, as their prob- Part I develops the theoretical framework for the efficient
ability is typically very small (see below for a more precise description of driven open many-body quantum systems. In
statement). They can be taken into account on the basis of section2, we begin with a direct derivation of the open sys-
optical Bloch equations[166], which leads, after elimination tem Keldysh functional integral from the many-body quantum
of the excited atomic state, to an additional, drivendissipative master equation(section 2.1). We then discuss in section2.2.1
term in the atomic dynamics. It describes the decoherence of in detail a simple example: the damped and driven optical
the atomic state due to spontaneous emission, i.e. position cavity. This allows the reader to familiarize themself with the
dependent random light scattering. The leading order contrib- functional formalism. In particular, the key players in terms
ution to the dynamics in the basis of lowest band Wannier of observablescorrelation and response functionsare
states is captured by the master equation described. We also point out a number of exact structural prop-
erties of Keldysh field theories, which hold beyond the spe-
t = i[HBH, ] [nl , [nl , ]], (15) cific example. Another example is introduced in section2.2.2:
l
there we discuss the mean field theory of condensation in a
where nl = bl bl is the local atomic density, and is the many- bosonic many-body system with particle losses and pump-
||2 ing. The semi-classical limit of this model and its validity are
body density matrix. For a red detuned laser the rate = 42
the content of section 2.3. This is followed by a discussion
is proportional to the microscopic spontaneous emission rate of symmetries and conservation laws in the Keldysh formal-
and the laser amplitude ||. Note the suppression of the scale ism in section2.4. In particular, we point out a symmetry that
by a factor / 1 for large detuning, compared to the allows one to distinguish equilibrium from non-equilibrium
eff
strength of H atom . The coherent and incoherent contribution conditions. Finally, an advanced field theoretical toolthe
9
Rep. Prog. Phys. 79 (2016) 096001 Review
open system functional renormalization groupis introduced of driven open quantum systems, such as Rydberg atoms
in section2.5. [172] and opto-nanomechanical settings [173]. For a recent
Part II harnesses this formalism to generate an understanding exposition of the physics of quantum master equationsand to
of the physics in different experimental platforms. We begin in efficient numerical techniques for their solution, see [48].
section3 with simple but paradigmatic spin models with discrete
Ising Z2 symmetry in driven non-equilibrium stationary states. Part I. Theoretical background
In particular, we discuss the physics of the driven open Dicke
model in section3.1. This is followed by an extended variant 2. Keldysh functional integral for driven open
of the latter in the presence of disorder and a multimode cav- systems
ity, which hosts an interesting spin and photon glass phase in
section 3.2. Section 4 is devoted to the non-equilibrium sta- In this part, we will be mainly concerned with a Keldysh field
tionary states of bosons with a characteristic U(1) phase rota- theoretical reformulation of the stationary state of Markovian
tion symmetry: the drivendissipative condensates introduced many-body quantum master equations. As we also demon-
in section2.2.2. After some additional technical developments strate, this opens up the powerful toolbox of modern quantum
relating to U(1) symmetry in section 4.1, we discuss critical field theory for the understanding of such systems.
behavior at the Bose condensation transition in three dimensions. The quantum master equation, examples of which we have
In particular, we show the decrease of a parameter quantifying already encountered in section 1.3, describes the time evol-
non-equilibrium strength in this case (section 4.2). The oppo- ution of a reduced system density matrix and reads [81, 82]
site behavior is observed in two (section 4.3) and one (section
1
4.4) dimensions. Finally, leaving the realm of stationary states, t = L = i[H , ] + L L {LL, },
in section5 we discuss an application of the Keldysh formalism
2 (16)
to the time evolution of open bosonic systems in one dimen- where the operator L acts on the density matrix from both
sion, which undergo number conserving heating processes. We sides and is often referred to as the Liouville superoperator
set up the model in section5.1, derive the kinetic equationfor or Liouvillian (sometimes this term is reserved for the second
the distribution function in section5.2, and discuss the relevant contribution on the RHS of equation(16) alone). There are two
approximations and physical results in sections 5.3 and 5.4, contributions to the Liouvillian: first, the commutator term,
respectively. Conclusions are drawn in section6. Finally, brief which is familiar from the von Neumann equation, describes
introductions to functional differentiation and Gaussian func- the coherent dynamics generated by a system Hamiltonian H;
tional integration are given in appendices A and B. the second part, which we will refer to as the dissipator D8,
Reflecting the bipartition of this review, the scope of it is describes the dissipative dynamics resulting from the interac-
twofold. On the one hand, it develops the Keldysh functional tion of the system with an environment, or bath. It is defined
integral approach to driven open quantum systems from in terms of a set of so-called Lindblad operators (or quantum
scratch, in a systematic and coherent way. It starts from the jump operators) L, which model the coupling to that bath.
Markovian quantum master equationrepresentation of driven The dissipator has a characteristic Lindblad form [174, 175]:
dissipative quantum dynamics [81, 82, 170], and introduces an it contains an anticommutator term which describes dissipa-
equivalent Keldysh functional integral representation. Direct tion; in order to conserve the normalization tr( ) = 1 of the
contact is made to the language and typical observables of system density matrix, this term must be accompanied by
quantum optics. It does not require prior knowledge of quantum fluctuations. The corresponding term, where the Lindblad
field theory, and we hope that it will find the interest ofand operators act from both sides onto the density matrix, is
be useful forresearchers working on quantum optical sys- referred to as the recycling or quantum jump term. Dissipation
tems with many degrees of freedom. On the other hand, this occurs at rates which are non-negative, so that the density
review documents some recent theoretical progresses made in matrix evolution is completely positive, i.e. the eigenvalues of
this conceptual framework in a more pedagogical way than the remain positive under the combined dynamics generated by
original literature. We believe that this not only exposes some H and D [176]. If the index is the site index in an optical
interesting physics, but also demonstrates the power and flex- lattice or in a microcavity array, or even a continuous posi-
ibility of the Keldysh approach to open quantum systems. tion label (in which case the sum is replaced by an integral),
This work is complementary to excellent reviews putting in a translation invariant situation there is just a single scale
more emphasis on the specific experimental platforms par- = for all associated to the dissipator.
tially mentioned above: The physics of driven BoseEinstein The quantum master equation (16) provides an accurate
condensates in optical cavities is reviewed in [10]. A general description of a systembath setting with a strong separation of
overview of driven ultracold atomic systems, specifically in scales. This is generically the case in quantum optical systems,
optical lattices, is provided in [48], and systems with engi- which are strongly driven by external classical fields. More
neered dissipation are described in [57]. Detailed accounts precisely, there must be a large energy scale in the bath
for exciton-polariton systems are given in [13, 147, 148]; spe- (as compared to the systembath coupling), which justifies
cifically, we refer to the review [149] working in the Keldysh
formalism. The physics of microcavity arrays is discussed in 8
We use the term dissipation here for all kinds of environmental influences
[14, 24, 134, 141], and trapped ions are treated in [17, 171]. We on the system which can be captured in Lindblad form, including effects of
also refer to recent reviews on additional upcoming platforms decay and of dephasing/decoherence.
10
Rep. Prog. Phys. 79 (2016) 096001 Review
10
Of course, the evolution of an M M matrix, where M is the dimension
9
Davies prescription [177] allows one to also describe equilibrium systems of the Hilbert space, can be formally recast into the evolution of a vector of
in terms of operatorial master equations, however, with collective Lindblad length M M . While such a strategy is often pursued in numerical
operators ensuring detailed balance conditions (see also [82]). approaches [48], it does in general not allow for a physical interpretation.
11
Rep. Prog. Phys. 79 (2016) 096001 Review
branches respectively (see figure3(b)). Indeed, this is an intui- states, and we first collect a few important properties of
tive and natural feature of evolving matrices instead of vectors. those. Coherent states are defined as (for simplicity, in the
So far, we have concentrated on closed systems which present discussion we restrict ourselves to a single bosonic
evolve according to purely Hamiltonian dynamics. However, mode b) | = exp(b) |, where | represents the vac-
we can allow for a more general generator of dynamics and uum in Fock space. (Note that according to this definition,
still proceed along the two-branch strategy. The most gen- which is usually adopted in the discussion of field integrals
eral (time local) evolution of a density matrix is given by the [180], the state | is not normalized.) A key property of
quantum master equation(16). Its formal solution reads coherent states is that they are eigenstates of the annihilation
(t ) = e(t t0 )L(t0 ) lim (1 + t L)N (t0 ). (19) operator, i.e. b| = | , with the complex eigenvalue .
N Clearly, this implies the conjugate relation | b = | 12.
The last equality gives a meaning to the formal solution in terms The overlap of two non-normalized coherent states is
of the Trotter decomposition: at each infinitesimal time step, the
given by | = e , and the completeness relation reads
exponential can be expanded to first order, such that the action d d
1 = e | |.
of the Liouvillian superoperator is just given by the RHS of the
quantum master equation(16); At finite times, the evolved state The starting point of the derivation is equation (19), and
is given by the concatenation of the infinitesimal Trotter steps. we focus first on a single time step, as in the usual deriva-
If we restrict ourselves to the stationary state of the system11, tion of the coherent state functional integral [180]. That is, we
but want to evaluate correlations at arbitrary time differences, decompose the time evolution from t0 to tf into a sequence of
we should extend the time branch from an initial time in the small steps of duration t = (t f t0 )/N , and denote the den-
distant past t0 to the distant future, t f +. In anal- sity matrix after the nth step, i.e. at the time tn = t0 + tn, by
ogy to thermodynamics, we are then interested in the so-called n = (tn ). We then have
Keldysh partition function Z = tr(t ) = 1. The trace operation n + 1 = et Ln = (1 + t L)n + O( t2). (20)
contracts the indices of the time evolution operator as depicted
in figure 3(c), giving rise to the closed time path or Keldysh As anticipated above, we proceed to represent the density
contour. Conservation of probability in the quantum mechani- matrix in the basis of coherent states. For instance, n at the
cal system is reflected in the time-independent normalization of time tn can be written as
the partition function. In order to extract physical information, d+, n d+, n d, n d
,n
+, n + 1| L( | +, n, n| ) | , n + 1 = i(+, n + 1| H| +, n, n| , n + 1 +, n + 1| +, n, n| H| , n + 1)
1
(
+ +, n + 1| L| +, n, n| L| , n + 1 +, n + 1| LL| +, n, n| , n + 1 + +, n + 1| +, n, n| LL| , n + 1
2
).
(22)
Without loss of generality, we assume that the Hamiltonian is normal ordered. Then, a matrix element | H| of the
Hamiltonian between coherent states can be obtained simply by replacing the creation operators by and the annihilation
operators by . The same is true for matrix elements of L , L, and LL, after performing the commutations which are necessary
to bring these operators to the form of sums of normal ordered expressions (see [181] for a detailed discussion of subtleties
related to normal ordering). Then we obtain by re-exponentiation
d+, n d+, n d, n d
,n
+, n + 1| n + 1| , n + 1 = eit (+,n it+,n ,nit,n iL(+,n+1, +,n, ,n+1, ,n ))+, n| n| , n + O( t2), (23)
12
Note that the creation operator cannot have eigenstates due to the fact
that there is a minimal occupation number of a bosonic state. In part-
11
We assume that it exists. We thus exclude scenarios with dynamical limit icular, the coherent states are not eigenstates. We rather have the relations
cycles, for simplicity. b| = (/ ) | , | b = (/)|.
12
Rep. Prog. Phys. 79 (2016) 096001 Review
The coherent state representation of L in the exponent in equa- denote spinors of a field and its complex conjugate by capital
tion(23) comes with a prefactor t, so that to leading order for letters),
t 0 it is consistent to ignore the difference stemming from
the bra vector at n + 1 and the ket vector at n in equation(24). iS + i
t,x(J ++ J)
Assuming all operators are normally ordered in the sense dis-
Z [J+, J] = D [+, ] e
cussed above, we obtain (J ++ J
e t,x
i )
= , (31)
L(+, +,
, ) = i(H+ H)
1 where we abbreviate (switching now to a spatial continuum of
+ L, + L, (L, +L, + + L, L, ) , (28)
2 fields) = dt dd x, the average is taken with respect to the
t, x
where H = H (, ) contains fields on the contour action S, and we have the normalization Z [J+ = 0, J = 0] = 1
only, and the same is true for L, . We clearly recognize the in the absence of sources. Physically, sources can be realized,
Lindblad superoperator structure of equation(16): operators e.g. by coherent external fields such as lasers; this will be made
acting on the density matrix from the left (right) reside on more concrete in the following section. The source terms can
the forward, + (backward, ), contour. This gives a simple be thought of as shifts of the original Hamiltonian operator,
and direct translation tablefrom the bosonic quantum master justifying the von Neumann structure indicated above.
equationto the Markovian Keldysh action (28), with the cru- Keldysh rotationWith these preparations, arbitrary cor-
cial caveat of normal ordering to be taken into account before relation functions can be computed by taking variational
performing the translation. derivatives with respect to the sources. However, while the
Keldysh partition function for stationary statesWhen we representation in terms of fields residing on the forward and
are interested in a stationary state, but would like to obtain backward branches allows for a direct contact to the second
information on temporal correlation functions at arbitrarily quantized operator formalism, it is not ideally suited for prac-
long time differences, it is useful to perform the limit t0 , tical calculations. In fact, the above description contains a
t f + in equation(25). In an open system coupled to sev- redundancy which is related to the conservation of probability
eral external baths, it is typically a useful assumption that the (this statement and the origin of the redundancy is detailed
initial state in the infinite past does not affect the stationary below in section 2.2.1). This can be avoided by performing
statein other words, there is a complete loss of memory of the so-called Keldysh rotation, a unitary transformation in the
the initial state. Under this physical assumption, we can ignore contour index or Keldysh space according to
the boundary term, i.e. the matrix element +(t0 )| (t0 )| (t0 )
of the initial density matrix in equation(25), and obtain for the 1 1
c = (+ + ), q = (+ ), (32)
final expression of the Keldysh partition function 2 2
13
Rep. Prog. Phys. 79 (2016) 096001 Review
and analogously for the source terms. The index c (q) stands values of lower order averages are subtracted; for example,
for classical (quantum) fields. This terminology signals c(t , x)c(t , x) = c(t , x)c(t , x)d c(t , x)d c(t , x)d
that the symmetric combination of fields can acquire a (clas- describes the density of particles which are not condensed. In
sical) field expectation value, while the antisymmetric one a compact notation, where W(2) denotes the second variation
cannot. In terms of classical and quantum fields, the Keldysh as in equation(34),
partition function takes the form
W (2)(t t , x x) J = J = 0
iS + i t,x (J qc + J cq ) c q
Z [Jc, Jq ] = D [c, q]e G (t t , x x) GR(t t , x x)
K
= A . (36)
=
e t,x q c c q ;
i (J + J ) (33) G (t t , x x) 0
The effective action is obtained from the generating func-
note in particular the coupling of the classical field c = (c, c) tional W by a change of active variables. More precisely, the
to the quantum source Jq = ( jq , jq ), and vice versa. Apart from active variables of the functional W, which are the external
removing the redundancy mentioned above, a further key sources Jc,q, i.e. W = W [Jc, Jq ], are replaced by the field expec-
advantage of this choice of basis is that taking variational tation values = ( , ) where = c, q, i.e. = [ c,
q ].
derivatives with respect to the sources produces the two basic (We remind the reader that capital letters denote spinors of
types of observables in many-body systems: correlation and fields and their complex conjugates.) The field expectation
response functions. Of particular importance is the single par- values are defined as (the notation indicates that these expec-
ticle Greens function, which has the following matrix struc- tation values are taken in the presence of the sources taking
ture in Keldysh space (for a brief introduction to functional non-zero values)
differentiation see appendix A; note that here we are talking = | = W / J = (W / j , W /j ), (37)
functional derivatives with respect to the components of the Jc, Jq
spinors of sources J = ( j , j ) where = c, q), where = q for = c and vice versa. Switching to these new
(t , x)(t , x) (t , x) (t , x)
variables is accomplished by means of a Legendre transform
c c d c q d
familiar from classical mechanics,
(t , x)(t , x) (t , x) (t , x)
q c d q q d
c,
q ] = W [Jc, Jq ]
t,x (J c q + J q c).
[ (38)
2Z 2Z
jq (t , x)jq (t , x) jq (t , x)jc (t , x)
We now proceed to show that the difference between the
= action S and the effective action lies in the inclusion of both
2Z 2Z
j (t , x)j (t , x) j (t , x)j (t , x) statistical and quantum fluctuations in the latter. To this end,
c q c c Jc = Jq = 0 instead of working with equation (38), we represent as a
G K (t t , x x) G R(t t , x x) functional integral [182], with the result
= i dA d . (34)
d
q + q ] i q i c
c + c ,
iS[
G ( t t , x x ) 0 c,
q] T T
e i[
= D [ c, q] e . c
q
(39)
In the last equality, in addition to stationarity (time translation
invariance) we have assumed spatial translation invariance. On the right hand side, we have used the property of the
GR/A/K are called retarded, advanced, and Keldysh Greens Legendre transformation
function, andin the terminology of statistical mechanics c = J q, q = J c.
/ / (40)
the index d stands for disconnected averages obtained from
differentiating the partition function Z; the zero component Equation (39) is obtained by exponentiating (i times) equa-
is an exact property and reflects the elimination of redundant tion (38), and using the explicit functional integral repre-
information (see section2.2.1 below). We anticipate that the sentation of the Keldysh partition function in W = i ln Z ,
retarded and advanced components describe responses, and equation(33). We have introduced the notation = + .
the Keldysh component the correlations. The physical mean-
This implies |Jc, Jq = 0 by construction (the average is taken
ing of the Greens function is discussed in the subsequent sub-
in the presence of non-vanishing sources), and moreover
section by means of a concrete example.
allows us to use D [c , q ] = D [ c , q ] in the functional
Keldysh effective action At this point we need one last
measure. Note the appearance of the field fluctuation in
technical ingredient: the effective action, which is an alterna-
the last term in the functional integral (39). This is due to the
tive way of encoding the correlation and response information
explicit subtraction of the source term in equation(38).
of a non-equilibrium field theory [182, 183]. We first intro-
Equation (39) simplifies when we consider vanishing exter-
duce the new generating functional
nal sources Jc = Jq = 0. This case is particularly intuitive, as it
W [Jc, Jq ] = i ln Z [Jc, Jq ]. (35) shows that the effective action [ c,
q ] corresponds to supple-
Differentiation of the functional W generates the hierarchy menting the action S[ c,
q ] by all possible fluctuation contrib-
of so-called connected field averages, in which expectation utions, expressed by the functional integration over c and
q.
14
Rep. Prog. Phys. 79 (2016) 096001 Review
The variational condition on , equation(40) (technically discussed in section 2.5. Examples in which such strategies
reflecting the change of active variables via equation (38)), were put into practice are discussed in section2.5 of this review.
precisely takes the form of the equation of motion derived
from the principle of least action (in the presence of an exter-
2.2. Examples
nal source or force term) familiar from classical mechanics.
Here, however, it governs the dynamics of the full effective In this section, we bring the rather formal considerations of
action. In light of the above interpretation of the effective the previous one to life by considering explicit examples for
action, equation(40) thus promotes the conventional classical drivendissipative systems. To be specific, in section2.2.1 we
action principle to full quantum and statistical status. consider the decay of a single-mode cavity. This is arguably
Conversely, neglecting the quantum and statistical fluctua- one of the simplest examples that combines coherent and dissi-
tions in equation(39), we directly arrive at [ c, q ] = S[ c, q ]. pative dynamics: the system itself consists of a single bosonic
This approximation is appropriate at intermediate distances13 mode (the cavity photon), and has a quadratic Hamiltonian
in the case where there is a macroscopically occupied con- and linear Lindblad operators. Hence, the Keldysh action is
densate: we may then count c = O( N ), with N the exten- quadratic and the functional integral can be solved exactly.
Hence, we are able to obtain an explicit expression for the
sive number of particles in the condensate, while fluctuations
q = O(1) for the noise field, generating functional defined in equation(33), which allows
c , q = O(1), as well as
us to conveniently study several properties of the Keldysh for-
which cannot acquire a non-vanishing expectation value. This
malism: what is known as the causality structure, the analyti-
crude approximation of the full effective action reproduces the
city properties of the Greens functions, and the intuitive and
standard GrossPitaevski mean-field theory, if we consider a
transparent way in which spectral and statistical properties are
generic Hamiltonian with kinetic energy and local two-body
encoded in the formalism. Moreover, we compare these find-
collisions, and drop the dissipative contributions to the action.
ings with the case of a bosonic mode in equilibrium. The pres-
The functions generated by the effective action functional
ence of the properties discussed in this sectionis not restricted
(via taking variational derivatives with respect to its variables
to the non-interacting case. Much rather, they prevail also in
c,
q) are called the one-particle irreducible (1PI) or ampu-
the presence of interactions, and are thus exact properties of
tated vertex functions [182]. A crucial and useful relation non-equilibrium field theories. We point this out alongside the
between the 1PI vertex functions and connected correlation examples.
functions (generated by W ) is In section2.2.2, we consider an example for an interacting
2 bosonic many-body system, which contains non-linearities
0 not only due to particleparticle interactions, but also in the
c(t , x)q(t , x)
t , x
dissipative contribution to the dynamics. These features are
2 2 realized experimentally in exciton-polariton systems. In the
(t , x) (t , x) (t , x) (t , x) present section, we restrict ourselves to a mean-field analysis
q c q q of this system and defer a discussion of the role of fluctuations
2W 2W to section4.
jq (t , x)jq (t , x) jq (t , x)jc (t , x)
2.2.1. Single-mode cavity, and some exact properties. The
2W
j (t , x)j (t , x)
0 master equationdescribing the decay of photons in a single-
c q mode cavity takes the general form of equation (16), with
H = 0 aa, where a and a are creation and annihilation
= (t t ) (x x)1. (41) operators of photons, and 0 is the frequency of the cavity
This relation follows directly from equations (37) and (40), mode. Assuming the external electromagnetic field to be in
using the chain rule. Combined with equation (36), it states the vacuum state, there is only a single term in the sum over
that the second variation of the effective action is precisely the with L = a, describing the decay of the cavity field at a rate 2
full inverse Greens function. (the factor of 2 is chosen for convenience). The corresponding
Typically, an exact evaluation of the effective action is not Keldysh action is given by [112]
possibleit would constitute the full solution of the interact-
t {a+(it0)a+ a(it0)a
ing non-equilibrium many-body problem. However, powerful S=
analytical tools have been developed for the analysis of such
i [2a+a
(a+a+ + a a)]}, (42)
problems, ranging from systematic diagrammatic perturba-
tion theory over the efficient introduction of emergent degrees where a, a represent the complex photon field. Performing
of freedom to genuine non-perturbative approaches such as the basis rotation to classical and quantum fields as in equa-
the functional renormalization group. The latter technique is tion(32) in section2.1, and going to Fourier space, the action
becomes
0 P A( ) ac( )
(ac( ), aq( )) R
13
,
At very long wavelengths, gapless fluctuations of the Goldstone mode
lead to infrared divergences in perturbation theory and invalidate this power
counting argument [184, 185].
S= P ( ) P K aq( )
(43)
15
Rep. Prog. Phys. 79 (2016) 096001 Review
16
Rep. Prog. Phys. 79 (2016) 096001 Review
GR(t , t ) = iac(t )aq(t ) Responses: The retarded Greens function, often called syn-
onymously response function, describes the linear response of
i
= (a+(t ) + a(t ))(a+(t ) a
(t )) a system which is perturbed by a weak external source field.
2 As an illustrative example, consider a driven cavity system
i (t )a(t ))
= (Ta(t )a(t ) + [a(t ), a(t )] Ta consisting of atoms and photons, which is considered to be in
2 a stationary state. Due to imperfections in the cavity mirrors,
= i(t t )[a(t ), a(t )], (53) there is a finite rate with which a photon is escaping the cav-
ity, or an external photon is entering the cavity. This process is
where T , T are the time-ordering, anti-time-ordering opera- expressed by the Hamiltonian
tors, which lead to a cancellation of the commutator for t > t
(see [102] for a more detailed discussion of time ordering on Hp = 2 (ba + ab ),
(58)
the Keldysh contour). The Keldysh Greens function in the where the operators b, b represent photons outside the cavity.
operator formalism is obtained the same way, Shining a laser through the cavity mirror, the operators b, b
can be replaced by the coherent laser field j(t), j(t), which
GK (t , t ) = iac(t )ac(t ) oscillates with the laser frequency j. The corresponding
i Hamiltonian, describing the complete system, is (for the pres-
= (a+(t ) + a(t ))(a+(t ) + a(t ))
2 ent purposes we can leave the Hamiltonian H of photons and
i (t )a (t )) atoms in the cavity unspecified)
= (Ta(t )a (t ) + {a(t ), a (t )} + Ta
2
Hj = H + 2 ( j (t )a + j(t )a). (59)
= i{a(t ), a(t )}. (54)
Since the fields j, j are classical external fields, they are equal
Response versus correlation functionsIn order to gen- on the plus and minus contour and the Keldysh action in the
erate response and correlation functions directly from the presence of the laser is
partition function, it is convenient to introduce source fields
j+ , j and express the partition function as the average (see
equation(31)) t
Sj = S ( j (t )aq(t ) + j(t )aq(t )). (60)
tion with respect to the source fields (see appendix A). For
example
= ac(t )j = 0 +
t
GR(t t ) j(t ).
The retarded Greens function is therefore a measure of the
Z ( jc , jq ) systems response to an external perturbation. An exper-
ac(t ) = i ,
jq (t ) imental setup, with which one can measure the coherent light
jc = jq = 0
field and therefore the response function of the cavity via so-
2Z ( jc , jq ) called homodyne detection, is illustrated in figure4. A closely
GK (t , t ) = iac(t )ac(t ) = i , related function of interest is the spectral function
jq (t )jq (t )
jc = jq = 0
A( ) = 2Im GR( ). (61)
2Z ( jc , jq )
GR(t , t ) = iac(t )aq(t ) =i . (57) It is the distribution of excitation levels of the system, i.e.
jq (t )jc (t ) when adding a single photon with frequency to the system,
jc = jq = 0
A( ) is the probability to hit the system at resonance. Indeed,
Due to causality, the field jq has to be zero in any physical it can be shown [178] that the spectral function is positive,
setup and the introduction of this field only serves as a techni- A( ) > 0, and fulfills the sum rule
cal tool to compute expectation values via derivatives. On the
other hand, the field jc can, in principle, be different from zero
and we will see in the following what the physical meaning of
A( ) = [a, a] = 1. (62)
this classical source field is, and in which respect it generates For our example of a single-mode cavity, the spectral function
the response function. is given by
17
Rep. Prog. Phys. 79 (2016) 096001 Review
(67)
i.e. it is a Lorentzian, which is centered at the cavity frequency
showing that the cavity is empty in the steady statewhich is
0 and has a half-width at half-maximum given by . Note
not surprising in the absence of external pumping.
that for 0, the photon number states become exact eigen-
We note that in the master equationformalism, information
states and the spectral density reduces to a -function peaked
on the static or spatial correlations is most easily accessible
at 0, A( ) = 2 ( 0 ). As these considerations illustrate,
only the state (density matrix) has to be known, but not the
the retarded Greens function contains essential information
dynamics acting on it. Temporal correlation functions can be
on the systems response towards an external perturbation,
extracted using the quantum regression theorem [186]. In the
and on the spectral properties.
Keldysh formalism, spatial and temporal correlation functions
Correlations: The Keldysh Greens function contains
are treated on an equal footing.
elementary information on the systems correlations and the
We can concisely summarize the above discussion of the
occupation of the individual quantum mechanical modes.
response and Keldysh Greens functions of the system, at the
A prominent example of a correlation function in quantum
risk of oversimplifying:
optics is the photonic g(2) correlation function. It is defined as
the four-point correlator Responses GR / A spectral information on excitations:
a(t )a(t + )a(t + )a(t ) which excitations are there?
g(2)(t , ) = , (64)
|a(t )a(t )|2 Correlations GK statistical information on excitations:
how are the excitations occupied?
and it is proportional to the intensity fluctuations of the intra-
cavity radiation field g(2)(t , ) I (t )I (t + ). In the limit
of 0, the g(2) correlation function reveals the statistics Relation to thermodynamic equilibriumBefore we move
of the cavity photons, i.e. it demonstrates super-Poissonian on, let us briefly discuss how we have to modify the formal-
(g(2)(0) > 1) or sub-Poissonian statistics (g(2)(0) < 1) as an ism in order to describe a cavity in thermodynamic equilib-
effect of the light-matter interactions. In the absence of inter- rium. In general, a system is in thermodynamic equilibrium
actions (as in our example), it is straightforward to show that at a temperature T = 1/ , if it is in a Gibbs state = eH /Z
where Z = treH , and its dynamics is coherent and gener-
|GK (t , + t ) + GR(t , + t ) GA(t , + t )|2 ated by the Hamiltonian H. (In particular, dissipative dynam-
g(2)(t , ) = 1 + .
|iGK (t , t ) 1|2 ics described by a term in the Lindblad form in equation(16)
(65) is not compatible with equilibrium conditions: see [187189],
For equal times, the retarded and advanced Greens func- and the discussion in section 2.4.1.) Specifying the thermal
tions satisfy GR(t , t ) GA(t , t ) = i, indicating the expected density matrix at t0 in the Keldysh functional integral in equa-
photon-bunching at 0. On the other hand, in the pres- tion (25) explicitly in terms of its matrix elements is rather
ence of interactions, equation(64) is modified and contains inconvenient, especially if one is interested in steady state
additional higher order terms, as well as off-diagonal contrib- properties and wants to take the limit t0 . An alternative
utions. However, it remains a measure of the photonic statis- 15
The same structural argument based on operator ordering insensitivity
tics in the cavity due to the physical relation to the intensity of the functional representation demonstrates why an ordinary Euclidean
fluctuations. functional integral yields the time ordered Greens functions, see, e.g. [180].
18
Rep. Prog. Phys. 79 (2016) 096001 Review
is suggested by the observation that thermodynamic equi- case of a time-evolving system, for which time translation
librium can be established in the system if it is weakly cou- invariance is absent, the Wigner transform of F corresponds to
pled to a thermal bath. Then, the finite decay rate in the the instantaneous local distribution function. For the important
retarded Greens function in equation(46) should be replaced case of thermodynamic equilibrium (of a bosonic system), it
by an infinitesimally weak systembath coupling 0. is F ( ) = coth( /2) = 2n( ) + 1, and equation(72) reduces
Additionally, the Keldysh Greens function has to be modified to (71). For the case where the bosonic Greens function has
in such a way that the integral in equation(67) yields the Bose a matrix structure in Nambu space, a subtlety concering the
distribution function n( 0 ) implying thermal occupations of preservation of the symplectic structure of that space arises,
the cavity mode, see section5.2. There, also, a time-dependent variant of the
non-equilibrium fluctuation-dissipation relations is discussed.
1
aa = n( 0 ) = . (68)
e 0 1 2.2.2. Driven-dissipative condensate. In the previous sec-
This can be achieved by replacing the Keldysh comp- tion, we discussed the simple case of a quadratic Keldysh
onent of the inverse Greens function in equation (43) by action, which allowed us to perform the Keldysh functional
P K ( ) = i2 (2n( ) + 1). We emphasize the key structural integral explicitly. An additional simplification resulted from
difference of the thermal Keldysh component, which is the fact that we were considering only a single bosonic mode.
strongly frequency dependent, to the Markovian case dis- Let us now consider a genuine many-body problem, which is
cussed previously, where this entry is frequency indepen- non-linear and in which the system consists of a continuum
dentthis gives a strong hint that Markovian systems can of modes. To be specific, in this sectionwe discuss the model
behave quite differently from systems in thermal equilibrium. introduced in section1.3.2 for a bosonic many-body system
With P R( ) = P A( ) = 0 i we obtain the equilib- with interactions and non-linear loss processes (i.e. with
rium Greens functions a loss rate that is proportional to the density) in addition to
the linear dissipative terms which were already present in the
1 example of the single-mode cavity. Then, for a specific value
GR( ) = GA( ) = , (69)
0 + i of the mean-field density 0, non-linear loss and linear pump
exactly balance each other and the system reaches a station-
GK ( ) = i2 ( 0 )(2n( ) + 1). (70) ary state. If 0 is different from zero, this signals the presence
Note that the Greens functions obey a thermal fluctuation- of a condensate, which is accompanied by the breaking of a
dissipation relation (FDR), which for the present example of specific phase rotation symmetry as will be discussed in detail
a single bosonic mode reads in section2.4, and the establishment of long-range order. In
section4, we give a detailed account of the influence of fluc-
GK ( ) = (2n( ) + 1)(GR( ) GA( )). (71)
tuations on this drivendissipative condensation transition.
This is discussed in more detail in section2.4.1. For the time Here, we content ourselves with a mean-field analysis, which
being, let us mention that this construction to describe ther- serves to illustrate some of the field theoretical concepts intro-
modynamic equilibrium by adding infinitesimal dissipative duced in section 2.1in particular, the effective action and
terms not only works in the present case of a quadratic action field equationsin a simple setting.
but can also be applied in the interacting case. Then, the con- In the basis of classical and quantum fields, the Keldysh
struction ensures that the free Greens functions (i.e. the ones action associated with the quantum master equation(9) reads
obtained by ignoring the interactions) obey an FDR. If the
t,x {q(it + Kc2 rc + ird )c + c.c.
non-linear terms in the action obey the equilibrium symme- S=
try discussed in section 2.4.1 (which is the case for generic
interaction terms), this property is shared by the full Greens
functions of the non-linear system.
( )
(uc iud ) qcc2 + qc2q + c.c.
(or: bare, microscopic) action S discarding fluctuations, in momenta we find purely incoherent diffusive, non-propagat-
contrast to the field equationsin equation(40), which involve ing modes 1R i
Kcuc 2
q and 2R i2ud 0. In particular, for
ud
the effective action)
q = 0 we have 1R = 0: this is a dissipative Goldstone mode
S S [159, 191, 192], associated with the spontaneous breaking
= 0, = 0. (74)
c q of the global U(1) symmetry in the ordered phase. The exis-
As already mentioned above in section2.2.1, there is no term tence of such a mode is not bound to the mean-field approx-
imation, but rather is an exact property guaranteed by the U(1)
in the action equation (38) with zero power of either q or
invariance of the effective action, even in the present case of
q, and the same is clearly true for S /c. Therefore, the first a drivendissipative condensate. We will come back to this
equationin (74) is solved by q = 0. Inserting this condition point in section2.4.6.
into the second equation, we have
[rc + ird (uc iud ) |0|2 ] 0 = 0. (75) 2.3. Semiclassical limit of the Keldysh action
The solution c = 0 is determined by the imaginary part of The theoretical description of a many-body system depends
equation(75): for rd 0, in the so-called symmetric phase, the crucially on the scale (this could be a length, time, or energy
classical field expectation value is zero, 0 = |0|2 = 0, whereas scalein practice these scales can be expressed in terms of
for rd < 0 we have a finite condensate density 0 = rd /ud. each other) on which it is observed and, in particular, on the
Taking the real part of equation(75), we obtain for the para- relation between this observation scale and the intrinsic scales
meter rc the relation rc = ucrd /ud. This condition can always be of the system. For example, at finite temperature T above a
satisfied by proper choice of a rotating frame, i.e. by perform- quantum critical point, non-trivial quantum critical behavior
ing a gauge transformation c ! c eit such that rc ! rc . can be observed at moderate energy scales which are larger
In the original (laboratory) frame this simply means that the than T but smaller than the Ginzburg scale where fluctuations
condensate amplitude oscillates at a finite frequency. start to dominate over the mean field effects [111]. On the
In a first step beyond mean field theory, quadratic fluctua- other hand, classical thermal critical behavior, which is per-
tions around the mean-field order parameter can be inves- fectly described by taking the semiclassical limit of the under-
tigated within a Bogoliubov or tree-level expansion: we lying quantum theory, sets in at energy scales below T. Out of
thermodynamic equilibrium, the analogue of a finite temper-
set c = 0 + c, q = q in the action equation (73) and
ature is Markovian noise. In the Keldysh action, this corre-
expand the resulting expression to second order in the fluc-
sponds to a constant term in the Keldysh sector of the inverse
tuations c, q. The inverse retarded, advanced and Keldysh Greens function, i.e. a constant noise vertex. Such a term is
Greens functions now become 2 2 matrices in the space of present in the action given in equation(73) and, therefore, we
Nambu spinors = ( , ). In particular, we have in the expect that in the long-wavelength limit this action can actu-
frequency and momentum domain ally be simplified by taking the semiclassical limit [102, 178].
Kcq 2 (uc iud ) ( u c i u d ) 0 We refer to it as the semiclassical limit, as, e.g. effects of
P R(, q) = 0
, quantum mechanical phase coherence are not necessarily sup-
(uc + i u d ) 0 Kcq (uc + iud )
2
pressed in this limit, as we will see. A useful equilibrium anal-
P A(, q)) = P R(, q), ogy is the physics of BoseEinstein condensates at finite but
P K = i 1. (76) low temperatures, where phase coherence still persists.
Formally, the suitability of the semiclassical limit can be
The excitation spectrum is obtained from the condition understood in terms of ideas that lie at the basis of the renor-
det P R(, q) = 0. Indeed, this is the condition for the field malization group (RG). The latter provides a recipe for finding
equation of the fluctuations P R(, q) c (, q) = 0 to have an effective description that is valid on large length scales,
nontrivial solutions. This yields [159] starting from a microscopic theory. Various RG schemes exist,
R
which allow us to systematically eliminate fluctuations on
1,2 = iud 0 Kcq 2(Kcq 2 + 2uc0 ) (ud 0 )2 . (77) short scales and infer their influence on the effective descrip-
We note that due to the tree-level shifts 0 the above- tion of the system on large scales; see section2.5. The most
described instability for rd < 0 is lifted: both poles are con- basic but still informative RG approach however consists in
sistently located in the lower complex half-plane, indicating simply ignoring the effect of short-scale fluctuations: one
a physically stable situation with decaying single-particle subdivides momentum space into a slow and a fast region,
excitations. For ud = 0, equation(77) reduces to the standard with qs [0, /b] and qf [/b, ] where is the UV cutoff
Bogoliubov result [190], where for q 0 the dispersion is (given by the inverse of some microscopic length scale below
R
phononic, 1,2 = cq (c = 2Kcuc0 the speed of sound), which the theoretical description is not valid any more, e.g.
R a lattice spacing) and b > 1, and omits all contributions to
whereas particle-like behavior 1,2 Kcq 2 is obtained at
the action which involve fluctuations with fast momenta. In
high momenta. Here, due to the presence of two-body loss a second step, one rescales all momenta with b to restore the
ud 0, the dispersion is qualitatively modified: while at high original range of momenta q [0, ] , and the thus obtained
R
momenta the dominant behavior is still 1,2 Kcq 2, at low effective long-wavelength description can be compared to the
20
Rep. Prog. Phys. 79 (2016) 096001 Review
original one. As a result, due to this simple RG transforma- fluctuations [26, 181]. Moreover, a complex prefactor of the
tion couplings in the action are rescaled as g ! gb dg, where dg term involving the derivative with respect to time, which also
is known as the canonical scaling dimension of g. Evidently, emerges upon renormalization, can be absorbed into a redefi-
for dg > 0, g grows under renormalization and hence g is a nition of the fields.
relevant coupling, while it shrinks under renormalization Strictly speaking, the above analysis is valid in the vicin-
and, hence, is irrelevant at long wavelength for dg < 0. For ity of the critical point only, where the inverse retarded and
the case of a marginal coupling with dg = 0, this level of advanced Greens functions show scaling. However, as long
approximation is not conclusive as to whether the coupling as one is close enough to threshold |l p| /(l + p ) 1, the
becomes larger or smaller under renormalization. Classifying canonical power counting is expected to give a useful orienta-
the coupling parameters of an action according to this scheme tion in the problem.
is known as canonical scaling analysis, or power counting. Apart from providing a significant simplification, taking
A useful approximation consists in neglecting all irrelevant the semiclassical limit also allows us to establish the con-
couplings. On the other hand, relevant couplings which are nection [193, 194] between the Keldysh functional integral
compatible with the symmetries of the modeleven those, formalism and the more traditional formulation of dynamics
which are not present in the microscopic descriptionshould close to a continuous phase transition in terms of Langevin
be included in the long-wavelength effective action. equations [1, 5]. In fact, the action in equation (79) is fully
Let us perform such an analysis for the Keldysh action equivalent to the following Langevin equation:
in equation (73). We can anticipate the canonical scaling
dimensions by noting that they are just the physical dimen- itc = [(Kc iKd ) 2 + rc ird + (uc iud ) |c|2 ] c + ,
sions, measured in powers of the momentum. We first focus (80)
on the vicinity of the threshold for condensation, i.e. when where is a Markovian Gaussian noise source with
rd l p 0. Then the retarded and advanced inverse zero mean, (t , x) = 0, and second moment (t , x) (t , x) =
Greens functions scale as P R / A q 2 (note that q 2 for 2 (t t ) (x x). This equivalence can be established by
low-momentum excitations). The canonical dimension is means of a HubbardStratonovich transformation of the noise
thus positive, d P R/A = 2. On the other hand, as anticipated vertex [102, 178], i.e. the term i2qq. To wit, in the Keldysh
above, Markovian noise analogous to a finite temperature functional integral
corresponds to a momentum-independent noise vertex, i.e. iS[c, c, q, q]
D [c, c, q, q] e
Z= , (81)
the term qq in the Keldysh action. In other words, the
Keldysh component of the inverse Greens function has van- where S is the semiclassical action in equation(79), we write
ishing canonical scaling dimension and classifies as mar- the noise vertex as a Gaussian integral over an auxiliary vari-
ginal, P K = i2 q 0. We furthermore use the natural scaling able (see appendix B),
dd x qd and dt 1/ q2, and the condition of scale
2 t,x qq =
1 t,x i t,x(q q). (82)
invariance of the action, S q 0this is a requirement stem- e D [, ] e 2
ming from the fact that the action appears in the exponent of
the functional integral (25), and thus must be dimensionless. As a result, the exponent in the functional integral (81)
We then find the scaling dimensions of the fields from the becomes linear in the quantum field, and the corresponding
Gaussian, quadratic part of the action to be integration can be performed, resulting in a -functional,
This in turn allows us to derive the scaling dimensions of ( [it +(Kc iKd )2 rc + ird (uc iud ) |c|2 ] c )
the quartic terms, and to classify their degree of relevance as
( [it +(Kc + iKd )2 rc ird (uc + iud ) |c|2 ] c ).
pointed out above. In particular, we find that in three spatial
dimensions, any quartic term that includes more than a single (83)
quantum field is irrelevant; the only non-irrelevant term of This expression can be interpreted as follows: for a given realiza-
higher order in the noise field is the quadratic noise vertex dis- tion of the noise field , the -functional restricts the functional
cussed above. Therefore, omitting irrelevant terms amounts to integral over c to the manifold of solutions to the Langevin
keeping only the classical vertex in the action equation(73), equation(80). The statistics of the noise field is determined by
i.e. to taking the semiclassical limit [102, 178]. Then, the the Gaussian weight factor in equation(83). Correlation func-
Keldysh action takes the form tions of classical fields can then be calculated by picking a ran-
dom realization of and calculating the corresponding c that
t,x {q[it+(Kc iKd )2 rc + ird ]c + c.c.
S= solves the Langevin equation, evaluating the correlation func-
tion for this solution and finally averaging the result according
(uc iud )qcc2 + c.c.+ i2qq , } (79)
to the Gaussian distribution of the noise field. Equation (80)
is, therefore, equivalent to the functional integral (83) in that it
where in addition to omitting irrelevant contributions we allows for the evaluation of arbitrary correlation functions.
have added an effective diffusion term iKd 2. Such a term Originally, Langevin equations like equation (80) have
can and will be generated upon integrating out short-scale been introduced as a phenomenological description of the
21
Rep. Prog. Phys. 79 (2016) 096001 Review
Figure 5. Equivalent descriptions on varying length scales. The quantum and classical Langevin equationsare stochastic equationsof motions
for the field operators, or for classical field variables. In contrast, the descriptions in the middle column are deterministic equationsof motion,
where either a density operator or a probability distribution (diagonals of a density matrix) are evolved. In the functional integral formulation,
the basic object in both quantum and classical cases is an action, which is averaged over all possible realizations of field configurations. The
semiclassical limit is valid at mesoscopic scales as discussed in section2.3. An effective description at macroscopic scales can be obtained by
means of renormalization group methods (see section2.5). Generically in Markovian systems, in order to reduce the complexity of the problem
it is useful to first perform the semiclassical limit before doing a renormalization group computation. However, in driven open quantum
systems there are also circumstances where this is inappropriate, and the full quantum problem has to be analyzed, see [28].
coarse-grained dynamics of the order parameter, and only In the semiclassical limit, we obtain a simple and intuitive geo-
later has a functional integral approachknown as the MSR metric interpretation of the symmetry and its absence under
approachbeen developed [195198]. The action derived non-equilibrium drive in terms of the location of the coupling
in these references is formally equivalent to the one in equa- constants of the effective action in the complex plane.
tion(79), with c taking the role of the order parameter field, A second fundamental consequence of the presence of
while the field that corresponds to q is known as the response symmetriesnow, continous global symmetriesis the
field. Noether theorem, stating that such symmetries imply con-
In addition to the descriptions of semiclassical dynamical served charges. In section2.4.3, we discuss the Noether theo-
models in terms of a semiclassical Keldysh (or MSR) functional rem in the context of the Keldysh formalism. Working on the
integral or a Langevin equation, there exists yet another equiva- Keldysh contour, we have to distinguish between two sym-
lent approach, in which the stochastic Langevin equationfor the metry transformations, for both the forward and backward
classical field or order parameter field is replaced by a deter- branches of the closed time path. In the basis of classical and
ministic evolution equationfor the probability distribution of the quantum fields, we can identify classical symmetry transfor-
latter, known as the FokkerPlanck equation. The derivation of mations, which act in the same way on fields on the forward
the FokkerPlanck equationcan be found, e.g. in [5, 102, 178, and backward branches, and quantum transformations, for
199]. Figure5 illustrates the equivalence of these approaches, which the transformation of the fields on the backward branch
which are valid on a (coarse-grained) mesoscopic scale, and the is the inverse of the transformation on the forward branch.
relations to microscopic and macroscopic descriptions. Non-trivial conservation laws follow only from the symme-
try of the Keldysh action under quantum transformations. We
illustrate this point with the example of the symmetry of the
2.4. Symmetries of the Keldysh action
action of a closed system with respect to space-time transla-
tions and phase rotations in section2.4.4, which entail con-
Symmetries take center stage in field theories. Translating the servation of energy and momentum, and of the number of
physics of the quantum master equationto the Keldysh func- particles, respectively. On the other hand, in open systems, in
tional integral allows us to leverage the power of symmetry which the number of particles is not conserved, only classical
considerations over to the context of open systems. In this sec- phase rotations are a symmetry of the action, and the continu-
tion, we discuss three different aspects of symmetries. ity equationthat is implied by particle number conservation
The presence of a first, discrete symmetry, considered in sec- has to be extended as detailed in section2.4.5.
tion2.4.1, allows us to conclude whether or not a quantum or Finally, another consequence of a global continuous sym-
classical system is situated in the realm of thermodynamic equi- metry is the existence of gapless modes in cases where this
librium. This symmetry is equivalent to the validity of thermal symmetry is broken spontaneously, as in a Bose condensation
fluctuation-dissipation relations (see section2.2.1) for correla- transition. In a drivendissipative condensate in which the
tion functions of any order and can be connected to energy con- number of particles is not conserved (see section 2.2.2), we
servation. Thermal equilibrium can thus be diagnosed by means show that the symmetry which is broken spontaneously at the
of a simple symmetry test on the Keldysh action. In particular, condensation transition is the classical phase rotation symme-
we argue that Markovian quantum master equationsexplicitly try, and we work out the Goldstone theorem, which guarantees
violate this symmetry, indicating non-equilibrium conditions. the existence of a massless mode, for this case in section2.4.6.
22
Rep. Prog. Phys. 79 (2016) 096001 Review
2.4.1. Thermodynamic equilibrium as a symmetry of the Keldysh also any higher order correlation function is determined by
action. A system is in thermodynamic equilibrium at a corresponding higher order response functions. Then, the
temperature T = 1/ , if (i) the density matrix is given by questionwhich we answer in the affirmative belowis,
= eH /Z where Z = tr eH , and (ii) the very same Ham- whether this infinite hierarchy of FDRs expressing thermal
iltonian operator H appearing in generates the unitary time equilibrium is related to a structural property of the theory.
evolution of the system, U = eiHt. Condition (ii) implies that Indeed, this structural property is a symmetry of the Keldysh
static correlations are in general not sufficient to prove that action, i.e. a transformation of the fields ! T such that
a system is in thermal equilibrium; however, dynamical cor-
relations are: this can be inferred from the fact that the static S[] = S[T ]. (85)
correlations of a physical system can always be encoded in Here, we denote = (+, +, ,
),
and we specify the pre-
a density matrix , and the latter can always be parameter- cise form of T below in equation(88). The tilde on the RHS
ized formally as an equilibrium density matrix, = eH/Z , of the equationindicates that all external fields appearing in
with a Hermitian operator H. (In other words, any state can S have to be replaced by their corresponding time-reversed
be thought of as a thermal state with respect to some Hamil- values. To name an example, the signs of magnetic fields have
tonian.) On the other hand, static (i.e. purely momentum- or to be inverted [189]. Evidently, discussing a single symmetry
space-dependent) properties do not allow us to discriminate instead of an infinite hierarchy of equationsis much more ele-
whether the generator of dynamics coincides with H. In sharp gant and practical: checking whether a given Keldysh action
contrast, any dynamical (i.e. frequency- or time-dependent) obeys equation (85) is straightforward and can be accom-
observable is manifestly governed by this generator, as is eas- plished in finite time, in contrast to verifying the full set of
ily seen in the Heisenberg picture (or a suitable generalization FDRs.
to open systems). Response functions at finite frequency are It is easily seen how these FDRs can be deduced as a
such dynamical observables, and the equilibrium conditions consequence of the symmetry of the Keldysh action (85). To
formulated above are reflected in the fact that in thermody- this end, we note that by a Keldysh rotation (32) and after
namic equilibrium the response of the system to an external Fourier transformation, the Greens functions appearing in
perturbation at a frequency is related to thermal fluctuations equation (84) can be expressed as a sum of averages of the
within the system at the same frequency by what is known as form (t , x)(t , x). Writing these explicitly as field int-
a fluctuation-dissipation relation [200] (FDR; for a discussion egrals, and anticipating that a change of integration variables
in the context of non-equilibrium BoseEinstein condensation T leaves the functional measure D [] invariant [189],
see [201]). Such a relation, which is equivalent to the com- we find
bination of the KuboMartinSchwinger (KMS) condition
[202, 203] with time reversal [204], is valid for any pair of
D [] (t, x)(t , x)eiS[]
(t , x)(t , x) =
operators. In particular, for the case that these are the basic
field operators of a single-component Bose system at the = D [T ] T (t , x)T (t , x)eiS[T ]
23
Rep. Prog. Phys. 79 (2016) 096001 Review
A(t )B(t ) = B(t )A(t + i ). (87) of the dissipative Keldysh action in section 2.1 started from
equation(16) with time-independent Hamiltonian. Even then,
Comparing this with equation(86) indicates that T involves the thermal symmetry can be used to diagnose non-equilibrium
a translation of t into the complex plane by an amount pro- conditions [189]. Again, it is sufficient to study only the time-
portional to the inverse temperature . Moreover, the order translation part of T : for time-independent H and L, in
of operators on the RHS of the KMS condition is reversed as equation(28), the Keldysh action is still invariant under time
compared to the LHS. The original time order can be restored translations of the form (t , x) ! (t + s, x). Importantly,
by means of a time reversal transformation [210]this step is this differs from the time translation that occurs in T by the
necessary in order to obtain a time-ordered expression which absence of a factor of , meaning that t is shifted by the same
can be expressed as a Keldysh field integral (by construction, amount on the forward and backward branches. Then, by
field integrals yield time-ordered averages [180]). A careful means of a simple shift of the integration variable t in equa-
analysis [189] leads to the precise form of the thermal sym- tion(30) the original form of the action can be restored. This
metry transformation ( = +() for fields on the forward strategy fails for the dissipative contributions in equation(28)
(backward) branch): that couple forward and backward branches, when the trans-
T (t , x) = (t + i /2, x), formation involves the contour index 18. Again we reach the
conclusion that quantum master equations describe genuine
T (t , x) = (t + i /2, x) (88)
non-equilibrium conditions, even though by a slightly different
(in the presence of a chemical potential , we have to multi- argument than in the driven but purely Hamiltonian setting.
ply the RHS of the first (second) line by e / 2(e / 2)). The In some cases, the Markov and rotating-wave approx-
transformation T is a composition of complex conjugation imations leading to a description in terms of a quantum mas-
of the fields and inversion of the sign of the time tboth ter equation or equivalent Keldysh action are applied in the
originating from the time reversal transformationand trans- absence of external driving fields. Then, these approximations
lation of the value of t by an amount i /2. The latter is might still be justified to study the behavior of specific observ-
induced by the KMS condition equation(87). We note that the ables, even though they explicitly break the symmetry [189].
translation of t in equation(88) takes opposite signs depend- To give an example, if one attempts to study thermalization of
ing on whether a field on the forward or on the backward a system due to the coupling with a heat bath, and one inte-
branch is being transformed. As we show in section 2.4.4 grates out the bath using the abovementioned approximations,
below, a similar form of time translations is connected to con- the resulting dynamics of the system will still lead to a ther-
servation of energy: indeed, if the Keldysh action is invariant mal stationary state. Correspondingly, all static properties of
under time translations (t , x) ! (t + s, x) with s R, the system will appear thermal. However, as discussed at the
the total energy in the system is conserved. A crucial differ- beginning of the present section, to unambiguously prove the
ence from the time translation that is part of T is that energy presence of thermal equilibrium conditions, one has to con-
conservation requires invariance under shifts by an arbitrary sider dynamical signatures such as FDRs. Then, as a conse-
real value s, whereas T involves a shift by the purely imagi- quence of the explicit violation of the thermal symmetry by
nary value i /2, where = 1/T is fixed and determined by the the Markov and rotating-wave approximations, the descrip-
temperature. tion of the system dynamics in terms of a quantum master
Under which conditions does the Keldysh action (30) equationwill lead to the wrong prediction that fluctuations in
with L defined in equation(28) have the thermal symmetry, the system do not obey an FDR.
i.e. under which conditions does it describe a system in ther- A simple example that illustrates the above discussion is
mal equilibrium? Let us first consider the parts of the action given by a single bosonic mode a with energy 0, driven coher-
corresponding to unitary time evolution, i.e. the first two ently at a frequency l, and coupled to a bath of harmonic oscil-
terms in equation(30) and the first line in equation(28). It is lators b in thermal equilibrium. The associated Keldysh action
straightforward to check [189] that these terms are symmet- can be decomposed as S = Ss + Ssb + Sb, where the action for
ric if the Hamiltonian densities H do not explicitly depend the system consists of two parts, Ss = S0 + Sl which read
on time. On the other hand, adding an external classical driv-
a( )( 0)a( ),
ing field such as a laser, and thus breaking time translational S0 = (89)
invariance by making H time dependent, the thermal sym-
metry is also broken17. The violation of the thermal symme- and
try by classical driving fields on the level of a microscopic
dt (a(t )ei t + a(t )ei t)
Hamiltonian description indicates that quantum master equa- Sl = l l
189]. Physically, this is due to the fact that a system for which = (a (l ) + a(l )). (90)
such a description is appropriate is necessarily driven.
In an effective description of a drivendissipative system in
terms of a Markovian master equationin a rotating frame, there 18
We note that also dissipative contributions corresponding to a system
is often no explicit time dependenceindeed, our derivation in thermal equilibrium, as described below equation(68) in section2.2.1,
couple the forward and backward branches. However, the special form of
17
The more precise statement is that there exists no rotating frame in which these terms conspire with the fact that the temporal shift in T is determined
the explicit time dependence fully disappears from the problem. by the inverse temperature = 1/T to make them invariant under T .
24
Rep. Prog. Phys. 79 (2016) 096001 Review
is the amplitude of the driving field, and due to the harmonic action is not invariant under the equilibrium transformation.
time dependence of the drive it affects only the component No frame exists in which the reference to the driving scale l
a(l ) in frequency space. The action of the bath is expressed were eliminated.
most conveniently on the basis of classical and quantum fields While this simple example demonstrates explicitly that
(see equation(32)). In addition to the coherent part stemming generically external driving takes a system out of thermal
from the oscillator frequencies, it involves infinitesimal dissi- equilibrium, and how this is manifest in the symmetry proper-
pative regularization terms specifying the thermal equilibrium ties of the Keldysh action, it is interesting to note that there are
state of the bath at inverse temperature = 1/T as discussed surprising exceptions to this rule, emerging as limiting cases.
in section2.2.1, One of them has been identified in [211]. Along the lines of
this reference, we discard the driving term Sl and instead con-
(b,c( ), b,q( )) sider a parametric systembath coupling of the form19
Sb =
i b, c( ) dt (a(t )b,(t )ei t + a(t )b,(t )ei t)
0 Ssb =
.
p p
(91)
+ i i2 coth( /2) b, q( )
Sl = (a (l )el / 2 + a(l )el / 2), (94) 2.4.2. Semiclassical limit of the thermal symmetry. For many
applications, as discussed in section2.3, the Keldysh action
where the appearance of the exponentials shows that the sym- in the semiclassical limit (79) is appropriate. Correspondingly
metry is violated. One might wonder whether it is possible to we should consider the semiclassical limit of the transforma-
restore the symmetry in a rotating frame in which the explicit tion (88). In the limit of high temperatures = 1/T 0, we
time dependence of the action is eliminated, i.e. by introduc- can perform an expansion of the transformed fields, with their
ing new variables a (t ) and b (t ) rotating at the frequency of arguments shifted by i /2, in terms of derivatives. To leading
the driving field, a (t ) = a (t )eilt and analogously for b (t ). In order20, this yields
terms of these variables, the driving part of the action becomes
T c (t , x) = x c (t , x),
Sl =
dt (a(t ) + a(t ))
T q (t , x) = x q (t , x) +
i
t c (t , x),
(97)
2T
= (a (0) + a(0)), (95)
where x is the usual Pauli matrix. High temperatures thus
means that the typical frequency scale of the field is much
i.e. the drive couples to the zero-frequency component of
smaller than temperature; indeed this recovers the intuition of
a ( ). Clearly, applying again the same transformation equa-
a semiclassical limit. If we replace the quantum field q by the
tion (93) with = 0 to the new variables, the driving term,
response field = iz q, equation(97) takes the form of the
as well as S0 in equation(89) and the systembath coupling
Ssb in equation(92) are invariant. However, in the action for classical symmetry introduced in [207]. The FDR in the semi-
the bath (91), as a consequence of the transformation to the classical limit, which may be derived as a WardTakahashi of
rotating frame the distribution function acquires an effective
chemical potential and becomes coth( ( l )/2). Hence, 19
In realistic systems the systembath coupling usually also contains terms of
the form a (t )b (t ) that are neglected in the rotating-wave approximation [211].
to leave this part of the action invariant, the transformation 20
Note that a contribution t q in the first line is suppressed according to
equation(93) has to be applied with = l, and again the full canonical power counting, see section2.3, equation(78).
25
Rep. Prog. Phys. 79 (2016) 096001 Review
26
Rep. Prog. Phys. 79 (2016) 096001 Review
T = + G + O(2 ), (103) L
where G is called the generator of the transformation. In general,
Q= x j 0 = x tT G f 0 . (110)
G is a 4 4 matrix with derivative operators as entries. We con- While this derivation is quite general, a more concrete expres-
sider specific examples below in section2.4.4. In the following sion for the Noether current can be obtained by restricting the
we assume that as in equation(30) the action S can be written form of the generator G in equation(103). In particular, in the
in terms of a Lagrangian density L as S = L , and that the next section we consider the specific cases of classical and
t, x
Lagrangian density is a local function of the fields and their first quantum transformations.
derivatives with respect to time and space. This assumption is
appropriate for most practical purposes. 2.4.4. Closed systems: energy, momentum and particle number
In an expansion of the LHS of equation(102) in powers of conservation. Energy and momentum conservationThis
, each coefficient has to vanish individually. The first-order property is expected for systems whose classical action is
contribution yields the relation determined by a translationally invariant Hamiltonian alone.
L Indeed, the construction of the Keldysh action for a Markov-
T G + = 0,
L
t , x T
G
(104) ian master equation in section 2.1 shows that terms which
couple the forward and backward branches are only contained
T in the dissipative parts of the action (30). In other words, they
where L/ ( L/+, L/+, L/, L/) .
arise upon coupling the system to a bath and integrating out
In the cases we consider, equation (104) holds true because
the latter. On the other hand, the Keldysh action for a closed
the integrand can be written as the divergence of a vector field system with unitary dynamics does not contain such terms and
f , i.e. we have can be written as
L L
f =
T
G +
T
G . (105) S =
t,x L , (111)
To proceed we consider local transformations, i.e. we con- where L+ and L are the Lagrangian densities evaluated with
sider T with = (t , x). We perform a change of integration fields on the forward and backward branches respectively.
variables T in the partition function. Then, assuming Given this structure of the action, let us consider a transfor-
that the functional measure is invariant with respect to the mation T which does not mix fields on the forward and back-
local transformation, we have ward branches. Then, the generator G has the block-diagonal
structure
Z= D []eiS[] = D []eiS[T ].
(106)
g 0
G= + . (112)
As before, we expand the RHS of this equality in a power 0 g
series in . Since by assumption the latter is a function of
(t , x), equation(102) does not hold true any more. Instead, The cases of physical interest are the classical and quantum
we find transformations mentioned above, corresponding to the
choices g+ = g and g+ = g respectively. In both cases, the
f + O(2 ), (107)
L Noether current can be represented as superposition of cur-
S[T ] = S[] + t, x T
G
rents on the forward and backward branches, which we define
in terms of g g+ as
where we used equation (105) and integration by parts to
write the RHS in a form that does not contain derivatives of L
j = T
g f , (113)
explicitly. Inserting equation(107) in the exponential on the
RHS of equation(106), and expanding the latter to first order
in , we obtain the condition where
L L
L f = g + g . (114)
t,x f
T
G = 0. (108) T T
With this definition, j can be obtained from j+ simply by
Since there are no restrictions on the choice of , this equa- replacing all instances of + appearing in j+ by . Then, the
tionimplies that the divergence of the expectation value van- symmetry of the action under a quantum transformation yields
ishes. The latter is just the quantum Noether current, a classical Noether current and vice versa. These currents are
L given by (as noted above we are free to choose a convenient
j = G f . (109) multiplicative normalization of the currents)
T
Note that j is not unique: for constant a, b, the combination 1
jc = ( j+ + j ), jq = j+ j . (115)
aj + b is also a conserved current. The associated Noether 2
charge Q corresponds to the integral over space of the zeroth As pointed out in section2.2.2, due to causality only the classi-
component of the Noether current, cal component of the field can acquire a finite expectation value.
27
Rep. Prog. Phys. 79 (2016) 096001 Review
The same is true for the currents defined in equation (115): the symmetry with regard to quantum phase rotations implies
in the presence of an external field that induces a current, we the classical symmetry. The generators of Uq are given by
have j+ = j = jc 0, whereas jq = 0. Hence, as g = iz. Then, considering particles with quadratic disper-
anticipated above, the continuity equation jq = 0, which sion relation and inserting the representation of the Lagrangian
follows from the symmetry under a classical transformation, density in equation(117) in the expression for the Noether cur-
does not entail a non-trivial Noether charge. In the next sec- rent (113), we find that the latter can be written as j = (, j )T ,
tion, we show that such a symmetry nevertheless does have where = ||2 is the density on the contour with index and
physical consequences, as it can be broken spontaneously in a the spatial components of the current are given by
condensation transition. 1
First, let us derive the Noether charge associated with the j = ( ), (119)
i 2m
symmetry of a Hamiltonian Keldysh action under contour-
dependenti.e. quantumtranslations in space-time. To be which is just the ordinary quantum mechanical current, eval-
specific, we consider the transformation (X ) ! (X + e ), uated on the closed time path. What is the physical meaning
where X = (t , x), and e is the basis vector in the direction of of the classical and quantum components, which are defined
the th coordinate of d + 1-dimensional space-time. This is a in equation(115), of the current given in equation(119)? If
symmetry of the action (111): the shift of the coordinates can we introduce in the Hamiltonian an externally imposed gauge
be absorbed by performing a change of integration variables field, in the Keldysh action it would couple to the quantum
X X e in the integral over the Lagrangian density L. current, by analogy to the source terms in the generating
Clearly, the symmetry is violated in the presence of dissipa- functional equation (33). However, the observable effect of
tive terms that couple the forward and backward branches. such a gauge field is that it can induce a classical current
For space-time translations, the generators g in equa- jc 0.
tion (112) acquire an additional index and are given by We close this sectionwith a comment on the status of the
g, = . It follows that the vector field f in equation(114) Uq(1) symmetry. It will be present on the microscopic level for
,
an action corresponding to a Hamiltonian which commutes
takes the form f , = L, and the classical Noether current,
with the number operator. Since the symmetry considerations
which we denote as Tc, , is given by
are properly applied to the microscopic action (and the func-
1 L tional measure), the conservation law ensues. However, this
T c, = T , , T , = L , (116)
2 T does not imply that the effective action must manifestly pre-
serve a Uq(1) symmetry. In fact, classical models of number
where T , are the components of the energy-momentum tensor conserving dynamics [1] do not exhibit this symmetry. This
[86, 205] on the forward and backward branches. In particular, leads to the picture that this symmetry is broken spontane-
the component T 0 0
,0 is the energy density, whereas T , i is the
ously under RG. The precise workings of such a mechanism
density of momentum in the spatial direction i = 1, 2, , d . are, to the best of our knowledge, not settled so far.
The spatial integrals over these densities yield the associated
Noether charges, e.g. if we express the Lagrangian density L 2.4.5. Extended continuity equation in open systems. The
in terms of a Hamiltonian density (as in equation(30), where conservation of particle number in a closed system follows
L is given by equation(28) with = 0) as from the continuity equationfor the classical Noether current
that is associated with the symmetry under quantum phase
1
L = izt H, (117) rotations. In an open system, this symmetry is absent, and
2 the continuity equationhas to be extended to account for the
we obtain the conserved energy density E , which is as exchange of particles with the environment, as we discuss in
expected given by the sum of the Hamiltonian densities on the the following.
forward and backward branches, To be specific, we consider the Keldysh action (73) for a
1 1 system with single-particle pump as well as single-particle
E =
2
x T 0 ,0 = 2 x (H+ + H). (118) and two-body loss. In order to derive the extended continuity
equation, as in equation(106) we perform a change of integra-
Along the same lines, conservation of angular momentum fol- tion variables Uq , where Uq is a local quantum phase
lows from the quantum symmetry of the Keldysh action with rotation, in the Keldysh partition function. Since the partition
respect to rotations of the spatial coordinates. function is invariant under this transformation, the coefficients
Number conservationAs another example, we consider in an expansion of the partition function in powers of the
phase rotations and their relation to particle number con- phase shift must vanish. In linear order, we find the condition
servation. In this case the classical transformation reads
! Uc = ei and the quantum transformation is jc p+ + l
+ + 4ud ( +)2 = 0. (120)
! Uq = ei. Classical phase rotations are a symme- This should be compared to equation(108): if Uq were a sym-
try of the Keldysh action, if the fields appear only in the combi- metry of the action, we would have found an ordinary continu-
nations . The quantum transformation is more restrictive: ity equation. This would have been the case in the absence of
in this case only products with = are allowed. Hence, the pumping and loss terms proportional to p, l , and ud. The
28
Rep. Prog. Phys. 79 (2016) 096001 Review
interpretation of these terms is as follows: in a system that is Greens functions correspond to roots of the determinant of
perfectly isolated from its environment, the temporal change inverse propagator (see section2.1). Hence, the presence of
of the density at a given point is due to the motion of particles a massless mode can be detected by checking whether the
toward or away from this point, i.e. the rate of change of den- determinant of the mass matrix M vanishes. The latter is deter-
sity is a source or sink for the mass flow as measured by the mined by the inverse propagator
divergence of the current jc . In an open system, on the
P A(, q)
P(, q) = R = G1(, q), (122)
other hand, there is in addition a dissipative current [215, 216] 0
due to the exchange of particles between the system and its P (, q) P K (, q)
environment. In particular, in equation(120) this dissipative
current has contributions from the pumping and loss terms. given exemplarily in equation(76), at = q = 0, i.e. M =
It is interesting to note that equation (120) can also be P(0, 0). Therefore, it is sufficient to consider frequency- and
obtained from the equation of motion of the local density momentum-independent field configurations or, in other
n(x) = (x)(x) in the operator formalism. From the master words, homogeneous field configurations, so that instead of the
equationt = L with Liouvillian L in Lindblad form given full effective action equation(39) we only have to deal with
in equation(9), it follows that the time evolution of n(t , x) in the the effective potential U = hom. /, where is the quanti-
Heisenberg representation is given by t n(t , x) = Ln(t , x) with zation volume in space-time. Since the effective potential U
the adjoint Liouvillian defined by tr(ALB ) = tr(BLA). We find inherits the symmetries of the effective action, it is a function
of precisely these combinations of fields which are invariant
tn(t , x) = j(t , x) + p(t , x)(t , x) under classical or quantum phase rotations. In the basis of clas-
ln(t , x) 4ud (t , x)2(t , x)2, (121) sical and quantum fields, these Uc-symmetric combinations are
= , where the indices and can take the values c
where the first term encodes coherent dynamics and cor- and q. In the following, we show that Uc invariance is sufficient
responds to the Heisenberg commutator i[HLP, n(t , x)], to guarantee the existence of a massless mode.
whereas the remaining contributions incorporate the dissipa- For convenience we switch to a basis of real fields
tive parts. Taking the average of this relation and expressing = (c,1, c,2, q,1, q,2 ) which correspond to the real and
the expectation values as Keldysh functional integrals yields imaginary parts of the complex classical and quantum fields,
equation(120). 1
i.e. = ( ,1 + i ,2 ) for = c, q. In this basis, the mass
Finally, we can obtain some intuition on the dissipative 2
correction to the current expectation value in mean field matrix reads
theory, where the correlators in equation(120) factorize into 2 U
= + a b U ,
2U a
2
products of single field expectation values + = = 0. Mij =
ij a
ij a ss a, b i i ab ss
We then recognize in the non-derivative terms (0 times) the ss
(123)
LHS of equation(75) (note that due to the factor of 1/ 2 in
where the subscript ss indicates that the fields should be set
the Keldysh rotation (32) the expectation values are related as
to their average values in the stationary state. The indices i
c = 0 = 2 , and that rd = (l p )/2), which equals and j label the components of the four-vector defined above
zero in a homogeneous situation within mean field theory. In (i.e. 1 = c,1, 2 = c,2 etc.), and a and b are double indices,
this way, we see that there is no particle number current on taking the values cc, cq, qc, qq. Let us consider the first term
average in a homogeneous drivendissipative system in sta- on the RHS of equation(123): in the ordered phase, the clas-
tionary state, as expected. sical field has a finite expectation value in the stationary state.
Without loss of generality we assume that this value is real.
2.4.6. Spontaneous symmetry breaking and the Goldstone Then, the field equations
U
theorem. Even though the pumping and loss terms in the Kel-
= a = 0,
dysh action in equation(73) break the symmetry with respect U
a i a ss
(124)
to quantum phase rotations Uq, the action is still symmetric i ss
under Uc transformations. As a result, even in the absence of
which actually determine the average value of the fields i in
particle number conservation there is a possibility of sponta-
neous symmetry breaking. In particular, a finite average value stationary state, have the solution i|ss = 20 i,1. Performing
c 0 breaks the classical phase rotation symmetry in a the derivatives a /i in equation(124) explicitly and insert-
non-equilibrium condensation transition. Here we show that ing i|ss, we obtain the following conditions:
this spontaneous symmetry breaking is accompanied by the
appearance of a massless mode, i.e. a mode with vanishing U U U
= = = 0. (125)
frequency and decay rate for zero momentum, which is known cc ss
cq qc
ss ss
as the Goldstone boson [217219]. We obtain this result by
carrying over the corresponding derivation from the equilib- Therefore, only the derivative uqq = [U /qq ]ss contributes
rium formalism (see, e.g. [182]) to the Keldysh framework. to the first term on the RHS of equation (123). Denoting
The single-particle excitation spectrum is encoded in the mixed second derivatives of the effective potential as
poles of retarded and advanced Greens functions, and such uab = uba = [ 2U /ab]ss, we find that the mass matrix can
a pole at = q = 0 is dubbed a massless mode. Poles of the be written as
29
Rep. Prog. Phys. 79 (2016) 096001 Review
An entry ucc, cc is not ruled out by symmetry; however, it must The functional renormalization group equation (for
vanish due to conservation of probability, see section 2.2.1. reviews see [229233]) constitutes an exact reformulation
Crucially, the retarded and advanced sectors feature one zero of the functional integral representation of a quantum many-
eigenvalue. This proves the existence of a massless mode as a body problem in terms of a functional differential equation. In
consequence of spontaneous symmetry breaking for a theory this, it is strongly distinct from, e.g. perturbative field theor-
with Uc invariance. etical renormalization group approaches, which concentrate
While this analysis guarantees the existence of a zero exclusively on the critical surface of a given problem. Instead,
mode (i.e. the complex dispersion relation has the property it may be viewed as an alternative and potentially more trac-
(q = 0) = 0), it does not provide a statement on the func- table tool for the analysis of the complete many-body prob-
tional dependence (q ). The latter could be inferred along the lem, also on length scales well below the correlation length
lines of [220224]. As we found in section2.2.2 in Bogoliubov near criticality. Indeed, it has proven a very versatile tool in
approximation, in an open system without (i.e. broken) Uq many different physical contexts, ranging from quantum dots
symmetry and spontaneously broken Uc symmetry, the lead- [234236], ultracold atoms [233], strongly correlated elec-
ing behavior at low momenta is diffusive, (q ) iDq 2, trons [237], classical stochastic models [238, 239], quantum
with a real diffusion coefficient D. This has to be contrasted chromodynamics [232, 240], to quantum gravity [241]. Here
to closed systems, in which microscopically both Uq and Uc we give a brief overview of the general concept adapted to
symmetry are present. There, the leading behavior in a phase non-equilibrium systems [26, 181, 234236, 242254]. It is
with spontaneously broken Uc symmetry is that of coherent used for the discussion of critical behavior in driven open
sound waves, (q) cq, with real speed of sound c. A phe- quantum systems in section4, with applications to a broader
nomenological justification of this behavior is given in [1], non-equilibrium many-body context left for future work.
where this can be understood as a consequence of a coupling The transition from the action S to the effective action
to additional slow modes relating to particle number conser- consists in the inclusion of both statistical and quantum fluc-
vation. In these phenomenological models, Uq symmetry is tuations into the latter (see equation (39)). In the functional
absent. This suggests a scenario of an additional spontaneous renormalization group approach based on the Wetterich equa-
breakdown of Uq symmetry upon coarse graining, but this tion[228], the functional integral over fluctuations is carried
issue seems not to be settled to date. out stepwise. To this end, an infrared regulator is introduced,
which suppresses the fluctuations with momenta less than
an infrared cutoff scale . This is achieved by adding to the
2.5. Open system functional renormalization group
action in (33) a term
0 R c
In section4.2, we investigate dynamical criticality of the Bose
condensation transition in driven open systems, motivated S = (c, q) ,
R 0 q
(127)
by many-body ensembles such as exciton-polariton conden- t, x
sates [12, 13, 159, 192, 225]. There again, coherent dynamics
with a cutoff or regulator function R. Some key structural
naturally competes with dissipation in the form of incoher-
properties are indicated below, but apart from these properties
ent particle losses and pumping. The situation parallels a laser
the choice of the cutoff is flexible and problem-specific. The
threshold [226, 227], however with a continuum of spatial
resulting cutoff-dependent generating functional and its loga-
degrees of freedom. This ingredient, however, causes the char-
rithm (see equation(35)) are denoted by, respectively, Z and
acteristic long-wavelength divergences of many-body prob-
W. Then, the scaled-dependent effective action is defined
lems in their symmetry-broken phase, or at a critical point.
by modifying the Legendre transform in equation(38) accord-
It implies that perturbation theory necessarily breaks down,
ing to
even when the interaction constants are small, due to a con-
tinuum of modes without a gap, which form the intermediate c,
[ q ] = W[Jc, Jq ]
states and are summed over in many-body perturbation theory.
This calls for the development of efficient functional integral t,x (Jc q + Jq c) S[ c, q]. (128)
techniques able to cope with these problems. Our method of
choice is the functional renormalization group based on the The subtraction of the S on the RHS guarantees that the
Wetterich equation[228], which we briefly introduce here in only difference between the functional integral representa-
its Keldysh formulation. This approach offers the particularly tions for and is the inclusion of the cutoff term in the
attractive feature of not being restricted to the critical point. latter,
30
Rep. Prog. Phys. 79 (2016) 096001 Review
R(q 2 ) 20, 0, As for the equilibrium path integral, the Keldysh field theory
(131) for spin models which obey the standard Ising Z2 symmetry,
R(q 2 ) 0, 0.
is formulated in terms of real fields, fluctuating in time and
Under the condition that 0 exceeds all energy scales in the space. These models have become rather important in the field
action, for 0 we may evaluate the functional integral of quantum optics, where the typical situation consists of a
(129) in the stationary phase approximation. Then, we find set of atoms, modeled as two-level systems, coupled to the
to leading order 0 Sin the absence of fluctuations (sup- radiation field of a high finesse cavity. One important model in
pressed by the cutoff mass gap 20), the effective action this context is the Dicke model, which has been introduced in
approaches the classical, or microscopic one. The evolution of equation(1) as the generic model for cavity QED experiments
from this starting point in the ultraviolet to the full effective with strong lightmatter coupling. Its Hamiltonian is
action in the infrared for 0 is described by an exact flow z N
g N
equationthe Wetterich equation[228]which was adapted H = p aa +
2
iz + N
ix(a + a). (133)
to the Keldysh framework in [234, 235, 244, 246]. It reads i=1 i=1
i 1
The Dicke Hamiltonian represents an effective model, which
= Tr [((2) + R ) R] , (132) describes the dynamics of a set of two-level atoms, labeled
2
by the index i, inside a cavity, which is pumped by a trans-
where (2) and R denote respectively the second variations verse laser. In this scenario, the coupling constant g describes
of the effective action and the cutoff action S. As antici- the scattering of laser photons into the cavity, as well as the
pated above, the flow equation provides us with an alterna- reverse process, and therefore rotates in time with the laser
tive but fully equivalent formulation of the functional integral frequency l. In a rotating frame, this time dependence is
(129) as a functional differential equation. Like the functional gauged away in the Hamiltonian, resulting in an effective shift
integral, the flow equationcannot be solved exactly for most p = c l of the bare cavity frequency c. In addition to
interesting problems. It is, however, amenable to numerous this external drive, the cavity is subject to permanent photon
systematic approximation strategies. For example, in the loss, due to imperfections in the cavity mirrors. The photon
vicinity of a critical point it is possible to perform an expan- loss is described by a weak coupling of the intra-cavity pho-
sion of the effective action in terms of canonical scaling tons to the environment, i.e. the vacuum radiation field outside
dimensions, keeping only those couplings which arein the the cavity. For typical experimental parameters, this coupling,
sense of the renormalization grouprelevant or marginal at which represents the rate with which the intra-cavity field and
the phase transition, see the discussion in section2.3, and see the environment exchange photons, is much smaller than the
section4.2 for applications. typical relaxation time of the environment and the latter can
31
Rep. Prog. Phys. 79 (2016) 096001 Review
32
Rep. Prog. Phys. 79 (2016) 096001 Review
The mode structure in the vicinity of the transition has been The exponent x = 1 found for the present non-equilibrium
analyzed in [112], where it has been found that in the presence transition coincides with the classical, finite temperature
of photon decay, the critical mode becomes purely imaginary exponent for the equilibrium Dicke model in line with the
before the transition happens, and the corresponding critical discussion above. On the other hand, at zero temperature, the
dynamics corresponds to a classical finite temperature trans- exponent of the corresponding quantum phase transition is
ition, see also figure7. This is mirrored by the fact that the found to be x = 1/2.
photons effectively thermalize in the low energy regime and The scaling behavior of the critical mode at the transition
their correlations can be described by a low energy effective is expressed in terms of the dynamical critical exponent t.
temperature Teff . The effective temperature can be obtained In the presence of dissipation, the excitations in the system
via a fluctuation dissipation relation, as discussed in equa- decay exponentially in time, which results in an asymptotic
tion (72) but promoted to Nambu space (see section 5.2 exponential decay of real-time correlation functions
for a discussion of the FDR in Nambu space). Solving this
equation yields the photon distribution function in Nambu {a(t ), a(0)} = G Kph(t ) et / t . (146)
representation
The characteristic decay time t is determined by the slowest
2 decaying mode in the system, i.e. by the mode with the small-
2g 2 z
Fph = z + x . (144) est imaginary part in the frequency spectrum. In the vicinity
z 2 2z
of the phase transition, this is the critical mode , which shows
g2
For large frequencies , this corresponds to the zero scaling
z
temperature distribution of the non-interacting system 1
t , (147)
F = z, which is diagonal in the photon modes. The approach |g gc|
to this limit is, however, algebraic 3, in contrast to an i.e. a dynamical critical exponent t = 1. This behavior is in
exponential approach according to large frequency behavior stark contrast to a zero temperature quantum phase transition,
of the Bose-distribution function in equilibrium. On the other where correlation functions do not decay over time but oscil-
g2
hand, for small frequencies , the occupation becomes late with characteristic frequencies c, the smallest of which
z
essentially thermal and purely off-diagonal F =
2Teff x
, with indicates the distance to the phase transition and encodes the
critical scaling behavior. The fact that in the present setting the
g2
low energy effective temperature Teff = ,
see equation(84).
z critical mode becomes purely imaginary is a generic feature
In this low frequency limit, the elementary excitations are of dissipative phase transitions, which is discussed further in
strongly hybridized polariton modes, which are diagonal section4.
in the photon quadratures x and p. The absence of a global All of the above-discussed results are encoded in the quad-
temperature scale in the present system reflects the fact that ratic Keldysh action, described by equations(138)(141). The
due to the Markovian dissipation, this interacting system with resulting critical behavior is then that of a non-interacting sys-
a discrete set of degrees of freedom is not able to achieve tem, and is associated accordingly to a Gaussian fixed point
detailed balance between the particular modes, i.e. the x and p of the renormalization group flow. This is in contrast to criti-
quadratures, for all energy scales. cal behavior in interacting systems, which are described by
33
Rep. Prog. Phys. 79 (2016) 096001 Review
35
Rep. Prog. Phys. 79 (2016) 096001 Review
Here, the soft spin approximation has been performed and the correlated locally in space, has infinite correlation in time.
scaling of the coupling constants gsl in the thermodynamic The latter is expressed by the fact that the variables Jsi are
1
limit is implicit, i.e. gsl N 2 . The thermodynamic limit is time-independent.
reached by taking an extensive number of atoms N but Averaging over all realizations of the couplings is done by
leaving the number of photon modes M < finite since the a Gaussian integration (see appendix B), and transforms the
consideration of an extensive number of photon modes is disorder part, i.e. the second line of equation(162), to a time
physically unrealistic. In order to obtain a large-N effective non-local quartic contribution
action, the photon degrees of freedom are integrated out, lead- iK
,
ing to the atomic action in frequency space Sat at = l, m ( )m, l ( ), (168)
N
c 2 cs
l, m
s (cs, qs)
1 0 with
S=
z c
2
0 qs
0 A( ) cm
0 A( ) ci l, m ( ) = (cl , ql ) R .
(cs, qs ) R .
(169)
+ Jsi (162) ( ) K ( )qm
si
( ) K ( )qi
The double sum in (168) is decoupled via a Hubbard
The frequency dependent terms are the symmetrized photon
Stratonovich transformation, which introduces the macro-
Greens functions
scopic fields Q, with , = c, q being Keldysh indices
p [113, 131]. This transformation is in general not unique but
R( ) = ,
( + i )2 2p can be made unique by requiring the equivalence
2i( 2 + 2 + 2p) 1
( ) =
K
2
(163)
Q
Z = 0 Q(, ) =
N
l( )l( ). (170)
( + i )2 2p l
sufficiently large M distributed according to a Gaussian dis- Here and Q are matrices in Keldysh space and G is defined
tribution function. We set their average value J Jsi = 0 as
as it lifts the frustration in the system caused by fluctuating
1
couplings but does not modify the universal behavior of the G ( ) = (G0 1( ) 2K ( )Q( )( )) . (173)
system at the glass transition [113, 131].
The averaged partition function of the system, including In the thermodynamic limit, the partition function is deter-
the probability distribution for the couplings Jsi is mined by the saddle point value of the action, i.e. by the
condition
Z= D [s]D [Jsi]eiS+iS , P
(165) S !
=0
where Q
(174)
for all , . This yields the values of the fields Q( ) at the
N J 2si
iSP =
2
K
(166) saddle-point, which can be identified with the atomic response
s, i and correlation functions. The saddle-point equationsfor the
is the action of the random couplings Jsi, with correlations atomic response and correlation functions are
K 2( 2 ) 1
JsiJlm = (s, li, m + s, mi, l ) . (167) Q R( ) = 4K (R( ))2Q R( ) (175)
N z
In this sense, the coupling of the atomic degrees of freedom to
and
the random variables Jsi has the same structure as the coupling
of the atoms to an external bath. However, the significant dif-
4K| Q R |2 K (Q AA + Q R R)
ference to a Markovian bathwhich would be -correlated Q K ( ) = . (176)
in space and timeis that the quenched disorder, while 1 4K| Q R R|2
36
Rep. Prog. Phys. 79 (2016) 096001 Review
Additionally, due to equation (170), the soft-spin constraint the spectrum reveals the breaking of time reversal symmetry
maps to the Q-fields according to by the dissipation and formally corresponds to the dynamical
universality class of dissipative quantum glasses, other examples
2
i Q K ( ) = N l cl( )cl( ) = 2. (177) of which are spin glasses coupled to an ohmic bath [261263]
or a bath of metallic electrons [257, 259]. The crossover
In the glass phase, the spins attain temporally frozen con- frequency is
2 + 1 2 (2 + 2 )2 1
figurations, expressed by an infinite correlation time of the
c = 2 2 2 + .
atomic correlator, which is expressed by a non-zero Edwards p p
(180)
2
Anderson parameter + 2 K
p z
1
qEA lim
t N
lx(t )lx(0). (178) and the modifications of the spectrum below this scale are
l
due to the damping introduced by the Markovian bath. On the
Consequently, the correlation function Q K ( ) consists of a normal or disordered side of the phase diagram, the corresp-
regular part, describing the short time dynamics and a non- onding low frequency propagator is
vanishing contribution at = 0. It can be expressed via the
modified fluctuation dissipation relation [261] QnR( ) = Z (( + i )2 2 )1, (181)
Q K ( ) = 4iqEA ( ) + QrK ( ) (179) describing the Ising spins as damped harmonic oscillators
with decay rate > 0 and frequency , with the physical
in terms of the order parameter and a regular contribution meaning of an inverse lifetime and energy gap of the atomic
QrK ( ). The soft-spin condition (177) together with equa- excitations. When the glass transition is approached, the gap,
tion (176) fixes the value of the Lagrange parameter and the inverse lifetime and the residue Z scale to zero simultane-
therefore fully determines the spectrum of the system. Similar ously and the atomic response in the glass phase
to the superradiant transition, the EdwardsAnderson para-
1
meter becomes non-zero when the poles of the system become QgR = Z ( 2 + | | ) 2 (182)
critical (approach the real axis) and its emergence is a mech-
anism to stabilize the critical modes in the ordered phase. is non-analytic and can no longer be interpreted as a harmonic
oscillator. The broken time reversal invariance manifests itself
3.2.2. Non-equilibrium glass transition. The variance of the in both parameters , 0 being non-zero, which modifies
effective, long-range atomatom interaction K is a measure of the low frequency dynamics in the entire glass phase towards
the frustration in the system and for sufficiently strong frustra- a dissipative quantum glass. This is for instance expressed by
tion, the system enters a glass phase, described by an infinite a spectral density which features an anomalous square root
autocorrelation time of the spins and the emergence of a non- behavior A ( ) = 2Im(Q R( )) || for small frequencies
zero EdwardsAnderson parameter qEA > 0. This goes hand in and therefore has a non-analytic response at zero frequency,
hand with a critical continuum of modes reaching zero, which see figure8. The latter leads to a characteristic algebraic decay
is the characteristic feature of a critical phase of matter. The of the photon correlation function, as discussed below. The
phase diagram for the fully coherent model has been analyzed universality class of the present dissipative quantum glass
in [131] and in [113] it was shown that the glass phase persists transition is determined by the critical exponents at the trans-
in the presence of dissipation. However, the dissipative model ition, which describe the scaling of the parameters , , Z as
shows new universal features of the glass phase, which corre- a function of = | K Kc|. It is summarized in the equations
spond neither to a zero nor to a finite temperature equilibrium
transition. This is attributed to the fact that the critical modes 3
2 (2p + 2 ) 2
are described by poles in the complex plane which are neither =
purely real as in the zero temperature (or quantum) case nor 8 K3 log( )
purely imaginary as for the finite temperature transition. 2p + 2
2
In the presence of dissipation, a photon that is emitted from = , (183)
16K2 log( )
an atom can either be absorbed by another atom, leading to an
effective atomatom interaction, which is of infinite range in p(2p + 2 )
3
37
Rep. Prog. Phys. 79 (2016) 096001 Review
38
Rep. Prog. Phys. 79 (2016) 096001 Review
glass phase can be performed by different but standard photon which makes continuous driving necessary to maintain a
output measurements, such as homodyne detection and inten- steady population.
sity correlation measurements. Formally, the photon Greens What makes these systems genuinely drivendissipative,
functions can be obtained from the Keldysh action formalism is that they are coupled to several baths, or to time-dependent
by introducing source fields in the microscopic action, driving fields, with which they exchange particles and energy.
which couple to the photon variables. After integrating out In the case of exciton-polaritons, which are hybrid quasipar-
the photon variables in the action, the photon Greens func- ticles composed of a WannierMott exciton and a photon, the
tions are then obtained via functional derivatives with respect photonic component can leak out through the mirrors forming
to these source fields. This standard field theory technique the cavity. Thus, the electromagnetic vacuum outside the cav-
relates the photon Greens functions to the solution for the ity serves as a reservoir into which particles are lost. These
atomic response and correlation functions [113]. losses have to be compensated by laser driving. In many exper-
iments, the driving laser is far blue-detuned from the bottom
3.2.4. Quenched and Markovian bath coupling in the Keldysh of the lower polariton band, and thus coherently creates high-
formalism. The Keldysh path integral formalism represents a energy excitations. During the relaxation of the latter, which is
theoretical approach which incorporates in a very straightfor- caused by phononpolariton and stimulated polaritonpolariton
ward way the coupling of the system to a non-thermal bath. scattering, coherence is quickly lost. In other words, lower
Here we compare quenched disorder and Markovian baths in polaritons can be regarded as being pumped not directly by the
more detail. As shown above in the present section, the cou- laser but rather by a reservoir of high energy excitons. Several
pling of the system variables to quenched disorder, realized by approaches have been used to model this effectively incoher-
the variables Jlm, is equivalent to the coupling to a bath with ent pumping mechanism [159, 192, 264]. In this context, the
infinite correlation time, description of a drivendissipative condensate introduced in
section2.2.2 might be considered as a toy model, which hides
K all the microscopic details associated with coherent excitation
Jlm(t )Jl m(t ) = (ll mm + lmml ). (187)
N and subsequent relaxation of high energy excitons in the cou-
pling of the system to several independent baths.
This represents the opposite limit of a Markovian bath, which
In the following, we review recent investigations of the uni-
has a typical correlation time that is much shorter than the
versal long-wavelength scaling properties of drivendissipative
system correlation time. As a consequence, at each single
condensates [19, 26, 181, 265267]. Then, the precise details
systembath interaction event, the Markovian bath immedi-
of the chosen model cease to matter: according to the power-
ately equilibrates and loses all its memory on the interaction.
counting arguments given in section2.3, for any choice of a
On the other hand, the quenched bath equilibrates infinitely
microscopic model the effective long-wavelength description
slowly and keeps memory on each single systembath interac-
is given by the semiclassical Langevin equationfor the con-
tion. As a consequence, it introduces effective temporally long
densate field (80). In fact, the latter can be derived in the spirit
range interactions between the system variables. These tend to
of GinzburgLandau theory for continuous phase transitions
slow down the system, pronouncing its = 0 correlations as
[190] by writing down the most general equationthat is com-
expressed by the EdwardsAnderson parameter.
patible with the symmetries of the problem. The reasoning
In an equilibrium formalism, disordered systems have to
behind such an approach is that the universal properties are
be dealt with by a computationally demanding replica method
fully determined by the spatial dimensionality and symme-
[256, 257], which is furthermore physically far less transparent
tries of a physical system [84, 86, 182].
than the Keldysh formalism, which thus represents a conveni-
A comparison in terms of symmetries of systems exhibit-
ent approach to disordered systems. In open quantum systems,
ing drivendissipative BoseEinstein condensation with sys-
where disorder, i.e. the coupling to a quenched bath, comes
tems in thermal equilibrium can be given most conveniently
together with the dissipation introduced by a Markovian bath,
on the basis of the semiclassical limit of section2.3: indeed,
both types of bath couplings compete with each other, lead-
the equation of motion (80) for the classical field takes the
ing in the present glassy system to strongly modified spec-
form of the Langevin equationsthat are used to model univer-
tral properties, which mirror the effect of the Markovian bath
sal dynamics in thermal equilibrium phenomenologically [1].
by a crossover from non-equilibrium to equilibrium scaling
Structurally, equation (80) is similar to model A for a non-
behavior.
conserved order parameter, with the additional inclusion of
reversible mode couplings [268]. However, the equilibrium
4. Non-equilibrium stationary states: bosonic models symmetry discussed in section2.4.1, which is present in all
models of [1], is violated in drivendissipative systems. This
Recent years have seen tremendous progress in experiments violation is due to pumping and loss terms, which moreover
on exciton-polaritons in semiconductor microcavities (for lead to the absence of particle number conservation. The lat-
reviews see, e.g. [13, 148]), making these systems the prime ter typically is associated with the symmetry under quantum
candidates for studying condensation phenomena under non- phase rotations (see section 2.4.4). This is another crucial
equilibrium conditions. As we have already mentioned in difference to BoseEinstein condensation in equilibrium:
section 1.3.2, the fundamental difference from conventional there, particle number conservation entails the existence of a
condensates is due to the finite life-time of exciton-polaritons, slow dynamical mode that modifies the universal properties,
39
Rep. Prog. Phys. 79 (2016) 096001 Review
and is taken into account in model F of [1]. To summarize, universal behavior is realized in drivendissipative conden-
drivendissipative condensation differs from BoseEinstein sates is encoded in the RG flow of g.
condensation in equilibrium by the absence of the equilibrium In the condensed phase, the RG flow of g is driven domi-
symmetry and the symmetry under quantum phase rotations. nantly by fluctuations of the gapless Goldstone mode dis-
This difference lies at the heart of the novel universal behavior cussed in section 2.4.6, i.e. by fluctuations of the phase of
out of equilibrium discussed in the following. the condensate field. The latter were shown [269272] to be
In section 2.2.2, we analyzed drivendissipative conden- governed by the KardarParisiZhang (KPZ) equation [73],
sates within mean-field theory, disregarding fluctuations in which , defined in equation(188), appears as the coeffi-
around the homogeneous condensate mode. However, in order cient of the characteristic non-linear term, see equation(199).
to access universal aspects such as the behavior of correlations Below in section 4.1, we present an alternative mapping of
of the condensate field at large distances or dynamical critical the long-wavelength condensate dynamics to the KPZ equa-
phenomena at the condensation transition, one has to carefully tion, starting from the Keldysh action in equation (73) and
include the effect of fluctuations. This can be done gradually integrating out the gapped density mode within the Keldysh
first integrating out short-scale fluctuations and moving on to functional integral. As a consequence of the mapping to the
account for fluctuations on larger and larger scalesby means KPZ equation, the RG flow of g in d spatial dimensions is at
of RG methods, such as the FRG discussed in section 2.5. the one-loop level given by [5]
Then, an intriguing question is whether the non-equilibrium
(2d 3)Cd 2
nature of drivendissipative condensates at the microscopic g = (d 2)g + g, (191)
scale becomes more or less pronounced under renormaliza- 2d
tion. In the former case, effective equilibrium is established where = ln(/ 0), is the running momentum cutoff, and
at large scales, while in the latter case, the universal physics Cd = 21 d d / 2(2 d /2) is a geometric factor. The key role
is expected to be profoundly different from its equilibrium that is played by spatial dimensionality becomes manifest in
counterpart. Which of these scenarios is realized depends cru- the canonical scaling of g, which is encoded in the first term
cially on the spatial dimensionality of the system: in 3D the on the RHS of the flow equation: to wit, g is relevant in 1D
long-wavelength regime is effectively thermal, whereas in one- where d 2 < 0, marginal in 2D, and irrelevant in 3D since
and two-dimensional systems the deviation from equilibrium then d 2 > 0. In 2D, the loop correctionthe second term
is relevant in the RG sense, and these systems are governed by on the RHS of equation (191)is positive, making g mar-
strongly non-equilibrium RG fixed points. ginally-relevant. This has far-reaching consequences for a
To see how this comes about, we require a quantitative drivendissipative condensate in which the microscopic value
measure for the deviation from equilibrium conditions in a of g is small, i.e. which is close to equilibrium: upon increasing
drivendissipative condensate. In section2.4.1 we discussed the scale at which the system is observed, the non-equilibrium
that thermal equilibrium requires the ratios of coherent to nature is more pronounced in one- and two-dimensional sys-
corresponding dissipative couplings to take a common value. tems, whereas effective equilibrium is established on large
This is illustrated in figure6 and formalized in equation(101). scales in three-dimensional systems. In 1D the canonical
In turn, it implies that any deviation Kc /Kd uc /ud in the val- scaling towards strong coupling is balanced at an attractive
ues of these ratios directly indicates a deviation from equilib- strong-coupling fixed point (SCFP) g by the loop correc-
rium conditions. Hence, the quantity defined by [19] tion. This term vanishes at d = 3/2, and for d > 3/2 the one-
Ku
loop equationdoes not have a stable SCFP, which, however,
= 2Kc1 d c is recovered in a non-perturbative FRG approach [252254].
Kcud
(188)
The RG flow of g that is found within this approach is illus-
can serve as a quantitative measure for the departure from trated qualitatively in figure10, which shows that also in 2D
thermal equilibrium. Note that = 0 in equilibrium, and that the flow is out of the shaded close-to-equilibrium regime with
is well-defined also for Kd = 0, which is the microscopic g < 1, and toward a strong-coupling, non-equilibrium fixed
value of the diffusion constant Kd in the model for driven point. The situation is quite different in 3D: in this case, if
dissipative condensates introduced in section2.2.2. It turns out the microscopic value of g is small, at large scales an effec-
to be most convenient to combine with the quantities [19] tive equilibrium with a renormalized value g 0 is reached.
However, for d > 2, there exists a critical line of unstable
K u u2 fixed points gc, separating the basins of attraction of the
D = K c d + c , = 1 + 2c
Kc ud
(189)
20 ud equilibrium and non-equilibrium fixed points, for g < gc and
g > gc respectively. Thus, in addition to the effective equilib-
(see section2.2.2 for the definition of the noise strength and rium phase, a true non-equilibrium phase may be reached in
the mean-field condensate density 0) to define the dimension- systems that are far from equilibrium even at the microscopic
less non-equilibrium strength as [5]: level also in 3D [273]. The properties of this phase have not
been explored so far.
2
g = 0d 2 , (190) The rest of this section is organized as follows: in sec-
D3 tion 4.2, we review the dynamical critical behavior at the
where 0 is the UV momentum cutoff. Thus, the answer to drivendissipative condensation transition in 3D [26, 181, 274],
the question of whether equilibrium versus non-equilibrium which, according to the above discussion, is governed by
40
Rep. Prog. Phys. 79 (2016) 096001 Review
41
Rep. Prog. Phys. 79 (2016) 096001 Review
Note that the quantum vertex in the action equation(73) that normal to the surface and at a rate ds /dt = ; If an increment
is cubic in the quantum fields does not contribute at this order. ds = dt is added along the normal, the corresponding change
Denoting the density fluctuations as = 0 we find of the surface height is
S=2 t,x ( 0 {1 [t + Kc( )2] + Kc22 dh = ( dt )2 + ( dt h )2 + (h )2 dt .
2 (200)
(195)
(uc1 ud 2 )} + i( + 2ud 0 ) 2 ) , Removing from dh /dt the average deposition rate by a
where 1 and 2 are, respectively, real and imaginary parts of . transformation to a co-moving frame, h(t , x) ! h(t , x) + t,
Here, of all terms involving the products of fluctuations 1 and and adding a term D 2h that describes surface tension, we
2 we only keep the dominant ones in the long-wavelength obtain the (deterministic part of the) growth equation(199).
limit, i.e. we neglect contributions containing temporal deriv- Intuitively, it seems clear that the growth of a surface rep-
resents a genuine non-equilibrium process. More formally,
atives 2t , 1 t , and spatial derivatives 12, 2 ,
this can be seen by noting that the non-linear term in the KPZ
and 1 ( )2 which are both small as compared to the mass-
equation does not in general satisfy a potential condition
like contributions uc1 and ud 2 in equation (195) for the [5]21.
Goldstone mode , in the regime 0 for q 0. Note that As pointed out above, an alternative derivation of the KPZ
terms of higher order in and are contained in the original equation(199) as the effective long-wavelength description of
action equation (73) both in contributions involving deriva- drivendissipative condensates starts from the Langevin equa-
tives and in the coherent and dissipative vertices. The valid- tion(80) [19]. Then, the coefficients in the KPZ equationare
ity of the saddle-point approximation, therefore, is restricted slightly different, and they are given by equations(188) and
to the low-frequency and low-momentum sector in a weakly (189) instead of equation (198). To be specific, the differ-
interacting system. ences are (i) the absence in equation (189) of the tree-level
The action (195) resulting from this expansion is linear in shift ud 0 of the noise strength in equation(198), and (ii)
and hence integration over this variable is trivial and yields the absence in equation(198) of the terms proportional to Kd
a -functional, which in turn facilitates integration over the in equations(188) and (189). (i) is due to the fact that in the
imaginary part 2 of , Langevin equation (80), which is valid in the semiclassical
limit (see section2.3), the quantum vertex that is proportional
Z= D [, 1, 2] [uc1 ud2]eiS = D [, ] eiS KPZ
,
to ccqq in the action in equation (73) is neglected. (ii)
(196)
results from the absence of the diffusion term Kd in the micro-
where at each step a normalization factor is implicitly included
scopic model (73); on the other hand, in the Langevin equa-
in the integration measure, ensuring Z = 1 [102, 178]. In
tion(80) this term is included, as it is generated by integrating
the last equality we replaced 1 by the KPZ response field
out fluctuations with wavelengths below the mesoscopic scale
= i2 0 1, and the KPZ action SKPZ is given by [5, 102] on which the Langevin equationis valid (see figure5).
Starting from the effective long-wavelength description of
SKPZ = t,x t D2 2 ()2 , (197) the condensate dynamics derived in this section, the universal
scaling properties can be obtained from an RG analysis. For
where the diffusion constant, non-linear coupling, and noise the KPZ equation, this procedure, which leads at lowest order
strength, respectively, are expressed in terms of the micro- in a perturbative expansion in to the one-loop flow equa-
scopic parameters in the original action equation(73) as tion(191)22, is described, e.g. in [5, 102].
+ 2ud 0 u2
For completeness, we note that in the absence of drive and
1 + 2c . (198)
uc
D = Kc , = 2Kc, = dissipation, i.e. for Kd = rd = ud = 0, an analogous derivation
ud 20 ud for a weakly interacting Bose gas in thermal equilibrium leads
Along the lines of the derivation of the Langevin equation(80) to an effective action for the phase alone that is given by
from the action in the semiclassical limit in equation(79), the
t,x (t c22),
2
KPZ action can be seen to be equivalent to the KPZ equation, S= (201)
which reads
21
The most general Langevin equationdescribing (near-) equilibrium
t = D2 + ( )2 + , (199) dynamics contains both (i) relaxational and (ii) reversible contributions to
2 the deterministic dynamics [5]. The linear diffusion term in equation(199)
D
where the stochastic noise has zero mean, (t , x) = 0, is of type (i): it can be written as H/, where H = 2 ( )2, and this
x
term alone would correspond to relaxation to an equilibrium stationary
and is Gaussian with second moment (t , x) (t , x) =
distribution eH / . The non-linear term, on the other hand, can not be
2 (t t ) (x x). Originally [2, 73], equation (199) was represented as the derivative of a Hamiltonian functional and is hence of
suggested by Kardar, Parisi, and Zhang as a model to describe type (ii). However, for reversible contributions to be compatible with a ther-
the growth of a surface, e.g. due to the random deposition of mal stationary state, they have to satisfy a potential condition. This is not the
atoms. In this context, h = is the height of the surface, and case for the non-linear term in the KPZ equationin dimensions d > 1.
22
Due to symmetries of the KPZ equation(for a comprehensive discussion
the origin of the non-linear term is purely geometric [73]: see [277]), the RG flow is described by the single parameter g defined in
the growth is assumed to occur in a direction that is locally equation(190).
42
Rep. Prog. Phys. 79 (2016) 096001 Review
where c = 2Kcuc0 is the speed of sound, and describes the of a free energy is not meaningful in a non-equilibrium con-
dissipationless propagation of sound waves with linear dis- text, correlations and responses can be considered also in the
persion = cq. To actually describe thermal equilibrium, this latter case. Then, universality entails that the algebraic long-
action has to be supplemented by infinitesimal regularization wavelength, long-time decay of any correlation or response
terms, discussed at the end of section2.2.1. function can be characterized in terms of a set of critical
Finally, let us comment on the relation of the approach exponents, which do not depend on the microscopic details
presented here to a Bogoliubov expansion in fluctuations of the problem but rather on symmetries and dimensionality.
of the complex fields around the mean field average values, The critical point is fully characterized by this set of critical
exponents.
c = c 0 and q = q. The latter can be recovered for-
We emphasize a crucial difference between the critical
mally by expanding the transformation equation(192) around
behavior in strictly non-interacting theories described by
the arbitrarily chosen value = 0, which yields
quadratic Hamiltonians or quantum master equations, and the
more genericbut also much more complexcase of inter-
c = 0 + + i 0 , q = , (202)
2 0 acting problems (for an example where a Gaussian fixed point
is physically important, see section3.1). The non-interacting
and interpreting and as the real and imaginary parts of
critical theories are described by a so-called Gaussian fixed
c. In the KPZ action in equation(197), such an expansion
point of the RG flow, while the interacting ones are associ-
in would amount to neglecting the non-linearity. Without the
ated to a so-called WilsonFisher fixed point. The structural
non-linearity, however, the KPZ action describes a free field
difference between these fixed points is reflected by the fact
coupled to a thermal bath. In other words, by performing a
that the values of the critical exponents differ; in particular,
Bogoliubov approximation the non-equilibrium character that
at a Gaussian fixed point, all exponents are rational numbers,
is intrinsic to the microscopic model with action (73) is lost
reflecting the validity of canonical power counting (see sec-
in the long-wavelength limit. Formally, this failure is due to
tion2.3), while one obtains non-trivial rational or non-rational
the fact that the saddle-point approximation is not valid for
numbers mirroring the importance of strong long-wavelength
low-momentum modes in the Goldstone direction since they
fluctuation corrections at a WilsonFisher fixed point. The lat-
are not gapped.
ter is more stable than the Gaussian one (where, in fact, all
couplings are relevant in the sense of the RG), and thus the
4.2. Critical dynamics in 3D physically relevant one even if interactions are small on the
microscopic scale. This is intuitive, taking into account that
In three spatial dimensions, the deviation from thermody- critical behavior deals with the longest distances and time-
namic equilibrium conditions, which is quantified by the value scales in a physical system.
of g defined in equation(190), is irrelevant in the RG sense The impact of fluctuations on the values of the critical
(see equation(191)). Hence, at large scales, effective equilib- exponents can be described quantitatively in the framework of
rium is established. This immediately implies that the overall the renormalization group, which provides a systematic way
picture of BoseEinstein condensation in thermal equilibrium to deal with the intricate long-wavelength, low-momentum
in 3D remains valid also in the drivendissipative context: divergences caused by the divergent correlation length. The
in particular, the mean-field analysis of section2.2.2, which critical behavior of drivendissipative systems in 3D was
predicts a continuous phase transition beyond which the clas- investigated in [26, 181] based on the functional renormali-
sical phase rotation symmetry (see section2.4.4) is spontane- zation group approach discussed in section2.5, and in [274]
ously broken and off-diagonal long-range order is established, using the field theoretic perturbative renormalization group
is modified quantitatively but not qualitatively once fluctua- [5]. These studies gave rise to the following picture of driven
tions around the mean-field condensate are taken into account. criticality:
(Note, however, that in equilibrium the condensation trans- The RG fixed point is purely dissipative (see figure6(c)),
ition is induced by lowering the temperature to a critical value, i.e. all complex couplings rotate to the imaginary axis.
whereas drivendissipative condensation is established by However, it is the approach to this fixed point that contains
tuning the single-particle pump rate.) Yet, the non-equilibrium universal information, and this makes it possible to distin-
nature of the drivendissipative system leaves its mark on the guish equilibrium from non-equilibrium critical behavior. The
approach to the long-wavelength thermalized regime in a fully following key properties are identified:
universal way.
An equilibrium system, fine-tuned to a critical point, (i) Asymptotic thermalization of correlation functions.Both
beyond which spontaneous symmetry breaking occurs via the static and dynamical critical exponents, governing,
a second order phase transition, exhibits universal behav- e.g. the asymptotic decay of the spatial and temporal first
ior. This is witnessed in non-analyticities in the free energy, order coherence functions, are found to take the same
and in the long-range decay properties of correlation and values as in the corresponding equilibrium problem. This
response functions. The decay properties are then governed can be made plausible by the fact that the fixed point cou-
by power laws, freed from the generic exponential cutoff plings indeed lie on a single ray in the complex planethe
er / at large distances r due to the divergent correlation imaginary axis. More formally, it is understood in terms of
length , defining the critical point. While the concept an emergent symmetry implying asymptotic emergence
43
Rep. Prog. Phys. 79 (2016) 096001 Review
of detailed balance, or thermalization, at large spatial or to a direct experimental observable, though its small
temporal distances. value poses a challenge for experimental observation.
(ii) Universal decoherence.The approach to the purely
In summary, it is found that the correlation functions (both
dissipative fixed point still hosts information on the
static and dynamic) thermalize, and show identical universal
underlying quantum dynamics. The fadeout of this
behavior to an equilibrium critical system with the same sym-
coherent dynamics is described by a new critical expo-
metries. However, the dynamical response functions contain
nent, which can be shown to be independent of the
universal information distinguishing equilibrium from non-
static and dynamical exponents, i.e. it does not relate to
equilibrium systems. The microscopic drive conditions are
the latter by scaling relations. Moreover, it can be seen
thus witnessed even at the largest macroscopic distances in a
that no more independent exponents could possibly be
fully universal way.
found [26, 274]. This is because the maximum number
In the following, we review how the results summarized
of independent exponents is determined by the maximum
above can be obtained from an open-system functional RG
number of independent microscopic mass scales [84],
approach [26, 181]. The basic ingredients of this method are
which are, in the quadratic part of the action (79), given
discussed in section2.5.
by the real parameters rc, rd , and . In addition, a coupling
f to an external source field, corresponding to a term
4.2.1. Effective action for drivendissipative condensation.
t, x f ( jc q + jq c + c.c.) in the action, has to be taken into
The Wetterich equation(132) describes how the scale-dependent
account, leading in total to four independent parameters.
effective action evolves from the microscopic action S at
No more independent scales can be added to the quad-
= 0 to the full effective action at 0, as fluctuations
ratic part of the action without violating the conditions
with momenta larger than the cutoff are integrated out, and
of (Anti-)Hermiticity and conservation of probability
the latter is gradually lowered. It is an exact non-linear differ-
as explained in section 2.2.1. Correspondingly, the set
ential equationfor the functional . In the absence of an exact
of four independent critical exponents (the correlation
solution, suitable schemes to approximate the functional differ-
length exponent , the anomalous dimension , the
ential equation have to be found. For the analysis of critical
dynamical critical exponent z, and the new exponent) is
behavior, progress can be made by choosing an ansatz for
maximal. Finally, the value of this exponent distinguishes
that has a similar structure to the microscopic action S in equa-
equilibrium from non-equilibrium conditions. This can
tion(73). In this way, the effective action is parameterized
be understood from figure6: the exponent describes the
in terms of the coupling constants appearing in the ansatz, and
fadeout of coherent couplings, i.e. the approach to the
consequently the functional differential equationfor can be
imaginary axis, governed by a power law r, where
cast in the form of a set of ordinary differential equationsfor
r is the new exponent. In the equilibrium case, all cou-
the couplings, as we describe in detail below.
plings rotate uniformly, giving rise to r 0.143 within
Any such ansatz will contain only a finite number of cou-
the FRG approach of [26, 181]. In contrast, in the non-
plings, and will hence truncate the most general structure
equilibrium case, the slowest approach to the real axis is
of . In other words, by choosing an ansatz of a particular
given by r 0.101. The value of the non-equilibrium
form, one makes an approximation, and the question arises,
exponent can also be determined analytically from the
which couplings should be included in the ansatz or trunca-
field theoretic perturbative renormalization group in a
tion in order to obtain meaningful results. For the study of
dimensional expansion, yielding the result (to two-loop
critical phenomena at a second order phase transition, the
order) [274]23
power counting scheme of section 2.3 provides a guideline:
2(4 d )2 4 following the arguments given there, we choose a trunca-
r = ln . (203)
25 3 tion which includes all couplings that are not irrelevant, and
which therefore takes the form of the semiclassical action in
Specifying to d = 3 dimensions, we obtain r 0.023.
equation(79):
The discrepancy between this value and the one obtained
U
from the FRG stated above (r 0.101) indicates that
q (iZ t+K 2)c
the two-loop computation underestimates fluctuation =
t, x c
+ c.c. + i
q q .
corrections in three dimensions24. Moreover, (possibly
significant) quantitative corrections to the values of critical (204)
exponents should also be expected from FRG calculations (Recall that the variables of the effective action are the field
that go beyond the truncation used in [26, 181]. expectation values , = c, q defined in equation (37).)
(iii) Observability.The drive exponent manifests itself, Here, in addition to the complex prefactor K = A + iD of the
for example, in the frequency and momentum resolved Laplacian, we are including a complex wave-function renor-
dynamical single particle response as probed in homo- malization Z = ZR + iZI . In the semiclassical limit, the homo-
dyne detection, see [181] for details. It thus corresponds geneous (i.e. not containing derivatives with respect to time or
23
The exponent c calculated in [274] is related r via r = c . space) part of the action can be written in terms of an effective
24
potential U , which is a function of c = c c. Due to the invari-
We note that in [181] erroneously the value of c from the field theoretic
RG was directly compared to r obtained from the FRG. Similarly, in [274]
the value of c was by mistake compared to A of [181]. ance of the microscopic action under classical phase rotations
44
Rep. Prog. Phys. 79 (2016) 096001 Review
(see section2.4.4), which is inherited by the effective action, Here, the subscript cq in , cq indicates that both the classical
only this combination of fields is allowed in the potential. The and the quantum fields are set to constant but non-zero values;
latter is given by denotes the quantization volume, and we introduced the fol-
1 1 lowing products of fields, which are invariant under classi-
U (c ) = u2(c 0 )2 + u3(c 0 )3. (205) cal phase rotations (see section2.4.4): cq = c q = qc and
2 6
q = q q. From the representation equation(207) it becomes
Here, both the two-body and three-body couplings, u2 = + i
and u3 = 3 + i 3 respectively, are complex. The three-body immediately clear how to project the flow equation(132) for
term is marginal according to power counting, and therefore onto a flow equationfor the derivative of the potential U in
included in the truncation. In the FRG, it is advantageous equation(205) with respect to c: one has to (i) evaluate equa-
to approach the transition from the ordered phase. Then, tion(132) for homogeneous fields, (ii) take the derivative with
the form of the effective potential in equation (205) cor- respect to cq, and finally (iii) set the quantum background
responds to an expansion around the stationary condensate fields to their stationary state value,
density 0. Indeed, this choice implies that the field equa-
1
tions /c = 0, /q = 0 (equations (40) in the absence
U = [ , cq] = = 0 . (208)
cq q q
of sources and evaluated with the scale-dependent effective
action ) are solved by c = 0 and q = 0 on all scales . Here and in the following, we specify flow equationsin terms
In the truncation equation(204), all couplingsincluding of the logarithmic cutoff scale = ln(/ 0). From the poten-
the condensate density 0are scale dependent. As indicated tial U in equation (205), the couplings u2,3 can be obtained
above, by means of such an ansatz for the effective action , by taking further derivatives with respect to c. However,
the functional differential equation(132) can be rewritten as instead of projecting the flow equation (208) in this way
a set of ordinary differential equationsfor these running cou- onto equationsfor u2,3, it is more convenient to introduce res-
plings, with initial conditions given by the microscopic action caled quantities as outlined above. Inserting the representa-
equation (79). This is achieved by applying projection pre- tion equation(206) of the bare quantum field in the effective
scriptions. In the following, we summarize this method for action (207) leads to appearance of a factor 1/Z in front of
the problem at hand. For details we refer the reader to [181]. the term involving the effective potential. This factor can be
absorbed by introducing a rescaled potential via U = ZU . The
flow equationsof the bare and rescaled effective potential are
4.2.2. Non-equilibrium FRG flow equations. The main
related via
idea of a projection prescription on a specific coupling is to
extract this coupling from the effective action by taking U = Z (Z U + U), (209)
appropriate derivatives of the latter with respect to the fields
and coordinates, and subsequently setting the fields to their where Z denotes the anomalous dimension associated with
stationary values c = 0 = 0 and q = 0 (as noted in sec- the wave-function renormalization (Z is specified below in
equation(222)),
tion2.4.6, choosing c to be real does not lead to a loss of gen-
erality). Then, applying the very same projection description Z = Z /Z. (210)
to the Wetterich equation (132) yields the flow equation for
Then, with cq = Z cq and inserting equation (208) on the
the corresponding coupling. For the actual evaluation of the
RHS of equation(211), the flow equationfor the renormalized
resulting flow equations, it is convenient to introduce rescaled
potential becomes (here, primes denote derivatives with
fields, which are related to the bare ones by absorbing the
wave-function renormalization Z in the quantum field: respect to c = cc)
c = c, q = Zq. (206) tU = Z U + , =
1
[ tk, cq] = = 0 . (211)
cq q q
As a consequence of this transformation, and by rescaling all
couplings appropriately, it is possible to obtain a reduced set In analogy to equation (205), the renormalized effective
of flow equations, from which Z is eliminated. In a second potential can be written as
step, the number of flow equations can be diminished fur- 1 1
ther by introducing dimensionless renormalized variables, as U (c ) = u2(c 0 )2 + u3(c 0 )3, (212)
2 6
detailed in section4.2.3 below.
We start by deriving flow equationsfor the non-linear cou- with renormalized couplings defined as u2 = u2 /Z = + i
plings in the effective potential defined in equation (205). and u3 = u/Z = 3 + i 3. To obtain flow equations for un,
To see how one can project the Wetterich equationonto flow n = 2, 3, one simply has to take derivatives of the relation
equations for these couplings, consider the effective action, un = U (n)(0 ) with respect to the cutoff scale , taking into
evaluated for homogeneous, i.e. space- and time-independent account that also 0 is a running coupling:
background fields:
un = (U (n))(0 ) + U (n + 1)(0 )0. (213)
, cq = (U cq + U qc i
q ). (207) Inserting equation(211) on the RHS of this relation leads us to
45
Rep. Prog. Phys. 79 (2016) 096001 Review
u2 = u2 = Z u2 + u30 + c ss , Note that the Goldstone theorem (i.e. the existence of a zero
(214)
eigenvalue of P R( = 0, q = 0), see section 2.4.6) is pre-
served during the flow. At the transition, where 0 0, both
u3 = u3 = Z u3 + 2c , (215) branches of the excitation spectrum that is encoded in the
ss
zeros of det P R(, q) (see equation(77)) become gapless. As
where we evaluate with c set to its stationary value c|ss = 0. pointed out in section 2.5, in the FRG, the resulting infra-
Finally, flow equationsfor the real and imaginary parts of u2 red divergences are regularized by introducing an additional
and u3 can be obtained by taking the real and imaginary parts contribution S, given in equation (127), in the functional
of equation(214) and (215) respectively. integral. In fact, by choosing the following optimized form of
The flow equationfor the stationary density 0 cannot be the cutoff function [278, 279] (which obviously satisfies the
specified without formal ambiguity: as we have already seen requirements stated in equations(130) and (131)),
in section 2.2.2, both real and imaginary parts of the field
equation(75) yield conditions on 0. However, we have seen R, K (q 2 ) = K (2 q 2 )(2 q 2 ), (220)
as well that the condition stemming from the real part can in the regularized inverse propagator (2) + R appearing in
always be satisfied by choosing the proper rotating frame.
the Wetterich equation(132), the terms Aq 2 in equation(219)
Therefore, the physically correct choice is to assume that 0
are replaced by
is actually determined by the imaginary part of the field equa-
tion, i.e. by the condition ImU(0 ) = 0. Taking the derivative A 2 for q 2 < 2,
A [q 2 + (2 q 2 )(2 q 2 )] = 2 (221)
Aq
of this condition with respect to the cutoff , we find for q 2 2,
0 = (Im U)(0 ) / ImU (0 ) = Im ss /, (216) (and there is an analogous replacement for the terms Dq 2).
where is the same as in equation(211). Hence, fluctuations with momenta below the cutoff scale
Having illustrated the main idea, the flow equation for acquire a mass 2, and the infrared divergences are lifted.
It remains to specify the flow equationsfor Z and K . They
the rescaled noise strength = / Z 2 can easily be obtained
can be obtained by choosing specific values of the indices i
along the lines of the derivation of equations(214) and (215),
and j in the flow equation(218) for the inverse propagator, and
with the result (for details of the derivation see [181]; ZR is
by taking derivatives with respect to the frequency and the
the real part of Z )
squared momentum q2, respectively:
i
= = 2ZR [ , cq]ss . (217) 1
q Z = tr [(1 + y )P R(, q)] , (222)
2 = 0, q = 0
Thus far we have specified how to project the Wetterich
equation onto flow equations for the couplings that para-
meterize the homogeneous part of the effective action given in K = q 2(P 22
R
(, q) + iP12
R
(, q)) . (223)
= 0, q = 0
equation(207), where the classical and quantum fields are set
to constant values. In the following we review the derivation of Comparison with equation(219) shows, that these are indeed
flow equationsfor the frequency- and momentum-dependent correct projection prescriptions. Note, however, that there
couplings, i.e. the wave-function renormalization Z and the is some ambiguity in choosing these projection prescrip-
R R
coefficient K of the Laplacian in equation(204). This requires tions: for example, A appears both in P11 and P22. Our choice
us to consider non-constant values of the fields. Moreover, we extracts K corresponding to the Goldstone direction, and
work in a basis of real fields which we introduced already in mixes Goldstone and gapped directions symmetrically in the
section 2.4.6, =
1
( ,1 + i ,2 ) for = c, q. Hence, the projection on Z (see [181] for details). Finally, the flow equa-
2 tionfor the renormalized coefficient K = K /Z is given by
inverse propagator in this basis is given by the second vari-
ational derivative of the effective action with respect to the K = K = Z K + K /Z. (224)
fields i (see equation (41); to ease the notation, we collect
While equations(214)(217) and (224) define a closed sys-
these fields in a vector = ( c,1, c,2, q,1, q,2 ); the comp-
tem of flow equationsfor the couplings u2, u3, , 0, and K (as
onents of this vector are labeled by i = 1, , 4), and the flow
indicated above, the wave-function renormalization Z drops
equationof the inverse propagator reads accordingly:
out of these equations), the explicit evaluation of the various
projections is rather tedious. For details of this calculation we
Pij (, q) ( ) (q q) = .
2
refer the reader to [181].
i(, q)j ( , q)
ss
46
Rep. Prog. Phys. 79 (2016) 096001 Review
implications of this fixed point? First, the value r indicates, where ui are the eigenvectors of N associated with the eigenval-
that the effective action at the fixed point is purely dissipative. ues ni in equation(230). The coefficients ci, which are referred
As we have already mentioned at the beginning of section4, to as scaling fields [178], take the scaling form ci e ni ni.
for vanishing coherent dynamics (or, in the terminology of Hence, for 0, the dominant contribution to r is given by
equilibrium dynamical models [1]: in the absence of revers- r u1n1 = u1r, where we identified the drive exponent
ible mode couplings), the drivendissipative model reduces
r = n1 0.101. (232)
to the equilibrium model A. Thus, the values s are the same
as in model A, and therefore the fixed point itself does not As anticipated in section 4.2 (ii), this exponent governs the
allow to distinguish whether the microscopic starting point of universal fade-out of coherent dynamics r in the driven
the RG flow was in or out of equilibrium. However, the non- dissipative system. Note that the existence of three distinct
equilibrium nature of the drivendissipative condensate is wit- eigenvalues (230) is due to the non-equilibrium character
nessed in the RG flow towards this effective equilibrium fixed of the microscopic model. Indeed, in model A with revers-
point. In the following we consider the universal regime of the ible mode couplings, the equilibrium symmetry discussed in
RG flow, which is reached in the deep IR (i.e. for / 0 1). section2.4.1 allows of only one ratio r = rK = ru2 = ru3 (see
In this regime, when the couplings are close to their values equation(101) and figure6). Then, the block N of the stabil-
at the fixed point, the RG flow can be obtained from a lin- ity matrix in equation(228) has only one single entry, which
earization of the flow equations in s = s s, r = r. The is given by the middle eigenvalue n2 in equation (230).
stability matrix governing the linearized flow takes block This shows that also in the equilibrium setting the dynamics
diagonal form, becomes purely dissipative at the largest scales [268], however,
47
Rep. Prog. Phys. 79 (2016) 096001 Review
the value of the critical exponent that governs universal deco- In a weakly interacting Bose gas in thermal equilibrium,
herence is different. As pointed out above, this is due to the the absence of true long-range order is caused by the vanishing
absence of the equilibrium symmetry in the drivendissipative energy cost of long-wavelength phase fluctuations. These are
case. The fact that the different values can be traced back to a governed by the quadratic effective low-energy action (201),
difference in symmetry supports the strength of the result, as from which the behavior of spatial correlations at long distances
different symmetries are known to give rise to quantitatively can be obtained straightforwardly, e.g. by introducing sources
different critical behavior [86]. as described in section2 and using the formulas for Gaussian
Decoherence at large scales has clear physical signatures functional integration collected in appendix B. The result is
which facilitate probing the drive exponent in experiments;
to wit, it implies that low-momentum excitations are diffus- (x)(x) 0ei((x) (x))
1
ing rather than propagating (note that this is also predicted by = 0 e 2 ((x) (x))
2
48
Rep. Prog. Phys. 79 (2016) 096001 Review
A notable difference between the KPZ equation for ran- The pump strengths at which the KT and KPZ crossovers
domly growing interfaces, and the present context of the occur can be estimated based on the parameters given in [303].
effective long-wavelength description of a drivendissipative It is convenient to introduce dimensionless pumping and loss
condensate is that the analogue of the interface height in the rates as follows [19]:
latter case is a phase, , and as such it is compact, i.e. defined
PR Rl
up to multiples of 2. This means that topological defects x= 1, = . (238)
vorticesin this field are possible27. Proliferation of vortices Rl Ruc
would lead to an even faster, simple exponential decay of spa- Here, P is the rate at which the excitonic reservoir is replen-
tial correlations. The present analysis does not take the pos- ished, while R is the amplification rate of the condensate
sible presence of vortices into account. due to stimulated scattering of polaritons from the reser-
How do these findings compare to experimental results? voir; l and R are, respectively, the decay rates of lower
Both in experiments on incoherently pumped polariton con- polaritons and reservoir excitons, and uc is the coherent
densates [300, 301] and simulations of parametrically pumped polaritonpolariton interaction [159]. For the parameters
systems [302], spatial correlations have been found to decay in [303], the KT transition should be expected at [19]
algebraically within the confines of the system. This, how- xKT 0.02. Denoting by x the pumping strength at which
ever, is not in contradiction to the present analysis based on the KPZ scale L in equation(237) drops below the system
the KPZ equation: indeed, if the microscopic value g0 of the size, we have [19]
rescaled non-linearity (190) is small, which is actually the case
in current experiments [19], a renormalized value of g = 1 is x /xKT = 2 ln(L /0 ) 0.04, (239)
reached in the RG flow only at the exponentially large scale where we took 0.1, 0 2 m, and assumed a pump spot
L = 0 e 8 / g0
, size of L 100 m. Thus, approaching the transition from
(237) above by lowering the pump power, the critical value xKT is
where 0 is a microscopic scale where the RG flow is initial- reached first, and the system loses algebraic order through
ized, e.g. the healing length of the condensate. Indeed, to unbinding of vortices [300302]. On the other hand, the cross-
obtain this result, we have solved equation(191) with initial over to the disordered regime will be controlled by KPZ phys-
condition g0 at the scale 0. Then, in systems of a size L well ics once x xKT, which can be achieved by increasing the
below L, an effective equilibrium description is applicable, loss rate (i.e. reducing the cavity Q) to 0.5.
leading to the observed algebraic decay of correlations (below While the above analysis shows that algebraic order in
the equilibrium KT transition). In other words, even in an infi- 2D drivendissipative condensates prevails only on interme-
nite system we should expect a smooth crossover from alge- diate scales, remarkably it can be restored on all scales in
braic to exponential decay at the scale L. strongly anisotropic systems [19]: consider a generalization
So far, our analysis has been based on the semiclassical of equation (80), where the gradient terms are replaced by
Langevin equation(80) for the condensate dynamics, which i 2
i = x, y K i c for = c, d . Correspondingly, equations(188)
according to the arguments given at the beginning of sec-
and (189) are replaced by
tion 4 correctly captures the universal scaling properties of
drivendissipative condensates. However, in order to obtain Ki u K i uc
an estimate of L for specific experimental parameters, a Di = K ic di + c , i = 2K ic 1 id , (240)
more microscopic model of exciton-polariton condensates is Kc ud K cud
required. Starting from a widely used model, which has been
which for i = x, y are the diffusion constants and non-lineari-
introduced in [159] and consists of a coupled system of equa-
ties appearing in the anisotropic KPZ equation. The RG flow
tionsfor the lower polariton field and the excitonic reservoir,
of this equation has been analyzed in [299, 304]. It can be
the bare value g0 and hence the scale L can be seen to depend
described in terms of the anisotropy parameter = yDx /xDy
on the rate at which the reservoir is replenished [19]. For high
(the system is anisotropic for 1), and the non-linearity
pump rates the KPZ scale L grows rapidly, so that by pump-
ing the system strongly enough, algebraic correlations can be g = 2x/(D2x Dx Dy ). The flow equationsare given by
made to extend over the entire system for any finite system
size L. When the pump rate is reduced, the system can lose dg g2 2
= ( + 4 1),
its algebraic order in two ways: either through the effect of d 32
the KPZ non-linearity if L drops below the system size, or d g
if L is still much larger than the system size at the critical = (1 2 ). (241)
d 32
value of the pump strength for the equilibrium KT transition
to occurthrough the proliferation of vortices. Note that as (Note that these equations differ from the RG equations in
pointed out above, even when KPZ physics becomes relevant, [19] by the sign on the RHS, which is due to the fact that
vortices might still modify equation(236). These considera- here we define the logarithmic scale as = ln(/ 0) instead
tions lead to the finite-size phase diagram reported in [19]. of = ln(L /0 ).) The line = 0 divides the flow into two
regions with distinct fixed-point structure: for > 0, which
27
This difference with the conventional KPZ equationalso arises in active we denote as the regime of weak anisotropy, at large scales
smectics [299]. isotropy is restored, i.e. 1 for , and all the results
49
Rep. Prog. Phys. 79 (2016) 096001 Review
28
discussed above apply . In the regime of strong anisotropy decay of temporal correlations takes the value = 1/4. The
corresponding to < 0, on the other hand, the flow is attracted distinction between one-dimensional condensates in equilib-
to an effective equilibrium fixed point with g = 0 and = 1. rium and drivendissipative condensates thus becomes mani-
Then, algebraic correlations of the condensate field can sur- fest only in the dynamical properties29. Moreover, in order to
vive if the effective temperature at the fixed point, which is actually observe KPZ scaling in the autocorrelation function,
given by the renormalized value of the dimensionless noise a large value of g, corresponding to a system far from equi-
strength = / Dx Dy , is below the critical value c = for librium, is favorable. This can be achieved by making drive
the equilibrium KT transition. Generalizing the microscopic and dissipation the dominant contributions to the dynamics,
model for exciton-polariton condensates mentioned above to as is the case in cavities with a reduced Q factor [265]. To
account for spatial anisotropy, the dependence of the effective be specific, for the parameters reported in [310], the Q factor
temperature on the strength of laser pumping can be obtained would have to be reduced by a factor of 30 (corresponding
[19]. Remarkably, the transition to the algebraically ordered to a polariton lifetime 1 ps instead of 30 ps achieved in
phase is found to be reentrant: upon increasing the pump rate the experiment) in order to make KPZ scaling observable in a
the ordered phase is first entered and then left again at even system of size 100 m.
higher values of the pump rate. Another observable, which conveniently encodes the scal-
ing properties of the phase of the condensate, is defined as
1 2
dx (t , x )2 dx (t , x ) . (243)
4.4. KPZ scaling in 1D 1 L L
50
Rep. Prog. Phys. 79 (2016) 096001 Review
51
Rep. Prog. Phys. 79 (2016) 096001 Review
52
Rep. Prog. Phys. 79 (2016) 096001 Review
state representation, and an evolution of the density matrix of momentum modes. In terms of the Hermitian distribution
towards its diagonal ensemble. In the Luttinger representation, function, the anti-Hermitian Keldysh Greens function can be
this decoherence expresses itself in a permanent production of parameterized according to
photons (254), i.e. a permanent heating of the Luttinger liq-
uid, which features no compensation mechanism in the quad- G qK, t , t = (G qR ! Fq Fq ! G qA)t , t . (258)
ratic sector and the consequent lack of a well defined steady
state. The energy increases constantly, which will lead to an Here, G qR, G qA = (G qR) and Fq are two-time functions evalu-
obvious breakdown of the Luttinger description as soon as the ated at equal momentum q and ! represents the convolution
energy stored in the long-wavelength modes exceeds a critical with respect to time and matrix multiplication according to
value. At this point, the dynamics is no longer dominated by the Nambu structure of the Greens functions. For a system
the coherences of but by its diagonal elements, including the which is diagonal in Nambu space, ! = ! reduces to a simple
breakdown of superfluidity and quasi-long range order. multiplication. In the presence of off-diagonal occupations,
The way in which energy is distributed amongst the long-
i.e. for a qaq 0, the operation ! has to respect the sym-
wavelength modes by the heating (252) is not typical for an
plectic structure of bosonic Nambu space and is promoted to
interacting system at low energies, since it deviates strongly
! = z !, as described in [44]. For a bosonic system in steady
from a BoseEinstein distribution and the associated detailed
state, all terms in (258) are time-translational invariant and
balance of energy. The phonon scattering terms in H (ph3) how-
Fourier transformation yields
ever favor detailed balance and strongly modify the actual
distribution function compared to (254), which makes them G qK, = G qR, zFq, Fq, zG qA, . (259)
non-negligible in the present non-equilibrium setting.
In contrast, for a system out of equilibrium undergoing a non-
5.2. Kinetic equation trivial time evolution, the mode occupations nq, t remain well
defined, but equation(258) is not time-translational invariant
In order to determine the time evolution of the excitation den- anymore. One then has to find a representation for the Greens
sities in interacting systems out of equilibrium, a common functions and F, which reveals the time-dependent occupa-
and often successful strategy is the so-called kinetic equa- tions. This is done in the following, leading to the Wigner rep-
tionapproach [102] (for an application to periodically driven resentation of the bosonic distribution function.
Floquet systems, see [316]), which determines the time evol- A convenient parameterization of non-equilibrium correla-
ution of the distribution function of the excitations in terms of tion and response functions is the so-called Wigner represen-
the systems self-energies. For the present setup, this approach tation in time, which introduces a forward and a relative time
has to be modified in order to take into account the driven coordinate (t, ) for the phonon Greens functions, according to
dissipative nature of the system and the resonant character of
the interactions. The latter lead to a breakdown of perturbation G qR, t , = i( )[aq, t + / 2, aq, t / 2], (260)
theory and require non-perturbative techniques beyond one-
loop corrections. A detailed derivation and discussion of the G qK, t , = i{aq, t + / 2, aq, t / 2}. (261)
applicability and limitations of such an approach can be found
in [43, 44]. Other forms of kinetic equations for interacting A two-time function C (t1, t2 ) can always be transformed to
Luttinger liquids, which focus on a different set of non-equi- t +t
Wigner coordinates, C (t1, t2 ) = C (t , ), by defining t = 1 2 2
librium conditions, include a perturbative treatment of phonon
and = t1 t2. Here, the explicit dependence on t expresses
backscattering terms, resulting from additional disordered or
the forward time evolution of non-equilibrium systems, while
lattice potentials [317, 318], as well as a treatment of cubic
for equilibrium systems in the presence of time-translational
phonon interactions in the presence of a smooth background
invariance, the forward time dependence of generic two-time
potential and a curved phonon dispersion [319].
functions just drops out, C (t , ) C (0, ), for all t. In Wigner
The quantity of interest in this sectionis the time-depend-
coordinates, the parametrization of the Keldysh correlation
ent occupation of phonon modes nq, t = aqaqt in the presence function is
of heating and phonon scattering. For bosonic modes and in
the steady state, the occupation of the modes is related to the G qK, t , = (G qR ! Fq Fq ! G qA)t , . (262)
Keldysh Greens function via
Here, we choose to neglect the subleading off-diagonal contrib-
i G qK, = 2nq + 1. (257) utions according to the above discussion, and consider only
diagonal modes in Nambu space. Equation(262) contains the
For a system in thermal equilibrium, nq = nB( q, T ) is the full Greens functions of the system, which can be expressed
Bose distribution function, see section2.4.1, while for a gen- via the Dyson relation in terms of the self-energies R / A / K ,
GK GR 1 G 0A A
eral non-equilibrium steady state, nq is a positive function,
= R .
0
A
which has to be determined from the specific context. One can
G 0
(263)
now introduce the Hermitian distribution function Fq, as in G 0 R K
equation(72), but generalized to a system with a continuum
53
Rep. Prog. Phys. 79 (2016) 096001 Review
Here, the notation (...)t, expresses the fact that the whole with nq, t = aq, t aq, t being the phonon density. Applying this
expression in brackets should be transformed to Wigner coordi- to equation(266) in the Wigner approximation yields
nates after performing the convolution qR ! Fq qR, t1, t Fq, t , t2
t
i
in ordinary time representation. The functionals R / A / K are the (q,t, q,t, Fq,t, + q,t, Fq,t, )A q,t,.
K R A
tnq, t = (271)
self-energies in the retarded, advanced, and Keldysh sectors, 2
which incorporate the effect of interactions and the heating on For well defined quasi-particles, i.e. excitations with a
the quadratic sector. The temporal derivative on the LHS of well defined energy-momentum relation, the spectral func-
(264) is the Wigner representation of the bare, non-interacting tion reflects the well-defined structure of the excitations and is
Greens functions without heating. Taking the Fourier transform sharply peaked at the quasi-particle dispersion = u| q|, with
of equation(264) with respect to the relative time coordinate
a typical width qR, t u| q|, which is the imaginary part of the
yields the Wigner representation of the distribution function
self-energy evaluated on the mass shell qR, t = Im(qR, t , u | q|).
Fq, t , = eiFq,t,, (265) If this is the case, the full self-energies in equation(271) mul-
tiplied with the spectral function can be approximated by their
for which
on-shell values, as they are expected to vary only smoothly
t Fq, t , = iqK, t, i(R ! F F ! A)q, t , . (266) in the region where A is non-zero. The approximation
qR,/tA, / K A q, t , qR,/tA, u/|Kq|A q, t , is called the quasi-particle
The corresponding transformation for the convolution inside
approximation and in the present case, similar to the Wigner
the parenthesis is
approximation, it is applicable in the entire Luttinger regime
i (
t t) [43, 44]. The latter is a consequence of the subleading,
(R ! F )q, t , = qR, t , e 2 Fq, t , . (267)
RG-irrelevant nature of the interactions, which lead to self-
Its explicit evaluation is nontrivial and in most cases simply energies qR, t u| q|. Performing the quasi-particle approx-
impossible. However, it is possible to approximate the com- imation, equation(271) obtains the simple form
plex exponential by the leading order expansion for many
typical relaxation dynamics [102]. In order to understand t nq, t = qK, t qR, t(2nq, t + 1). (272)
equation(267), one should take a closer look at its expansion
up to first order in derivatives: In this equation, the anti-Hermitian Keldysh self-energy qK, t ,
R has been replaced by its on-shell value qK, t , u | q| = 2iqK, t,
i tFq, t , q, t ,
(R ! F )q, t , = qR, t , Fq, t , 1 + ( t ) + O( 2). with the real function qK, t.
2 Fq, t , qR, t , The kinetic equation(272) describes the time evolution of
(268)
t Fq, t , the phonon density nq, t in terms of the on-shell self-energies
The ratio qf is the rate with which the distribution
Fq, t , R, K , which in turn are determined by both the interactions
F is changing in forward time, i.e. the forward time evolution and the heating term. In order to identify the contribution from
rate, which is determined by the interplay between heating the heating, one has to identify the impact of the dissipative
and collective quasi-particle scattering. On the other hand, contribution in equation(250) on the action S. Following the
qR, t , steps in section2.1 carefully and setting the off-diagonal den-
(rq)1 is identified with the inverse rate of the rela-
qR, t , sity contributions to zero, according to equation(254), the dis-
tive time dynamics, which is dominated by fast single phonon sipative contribution to the microscopic action S is
propagation. As a consequence rq qf and the correction
| p|
terms in (268) can be safely neglected. This is a typical situ- SD = i p,t K
aq, p, t aq, p, t . (273)
ation for many kinetic equationapproaches and is termed the
Wigner approximation. For the present setup, a more careful Here, p is the momentum variable and q labels the quantum
analysis has shown that the Wigner approximation is indeed component of the Keldysh field variable. The dissipation thus
satisfied as long as the Luttinger representation of the problem enters the action only in the quantumquantum sector and
is valid [43]. does not modify the spectrum in the quantumclassical sec-
In order to project equation (266) onto the quasi-par- tor. This is again a consequence of the Hermitian nature of the
ticle densities, it is multiplied by the spectral function Lindblad operators, which lead to a continuously increasing
A q, t , = iG qR, t , iG qA, t , , followed by a subsequent integration occupation of phonons but do not introduce a compensating
over frequencies . The spectral function fulfills the sum rules dissipative mechanism in the retarded and advanced sector of
54
Rep. Prog. Phys. 79 (2016) 096001 Review
30
the action . This is drastically different from the situation in self-consistent Born approximation, which we discuss in the
the models of sections3 (see equations(141) and (140)) and following.
4 (see equations(204) and (205)), where the interplay of dis- The Keldysh action for interacting Luttinger liquids with
sipation in the retarded/advanced sectors and fluctuation or heating is composed of a dissipative part SD, which has been
noise in the Keldysh sector of the action lead to non-equilib- discussed in equation(273), and a Hamiltonian part SH, which
rium fluctuation-dissipation relations describing well-defined results from the Hamiltonian dynamics in equations (250),
stationary states different from the trivial state 1. (251). In the Keldysh representation, the action is
itu| p|i0+
With the form of (273), the Keldysh self-energy is
ac, p, t
0
| p|
qK, t = 2K + qK, t, where the bare qK, t in this form is deter-
S=
p, t
(ac, p, t , aq, p, t )
itu| p|+i0 + | p| aq, p, t
mined by the interactions alone. The resulting kinetic equa- i
K
tionconsists of three contributions
| q| + p,k,t
v( p, k )[2ac, k + p, t ac, k, t aq, p, t
tnq, t = + qK, t qR, t(2nq, t + 1) .
2K !#####"#####$ (274) (275)
in term scattering out term scattering + aq, k + p, t (ac, k, t ac, p, t + aq, k, t aq, p, t ) + h.c.]
in term heating
mean-field Mott initial states, which are the exact ground states of the
BoseHubbard model for fine-tuned J = 0, are pure states which are not (280)
touched by the dissipator considered in this section. One point of view on
+ 2v 2( p, q ) G KP + QG KP + G PA+ QG PR + G PR + QG PR , (281)
( )
symmetries to be physically relevant (such as a spatially local gauge sym-
metry in the J = 0 example).
55
Rep. Prog. Phys. 79 (2016) 096001 Review
Inserting (282) in the definition for the retarded self-energy where the time-dependent integral Jt is q-independent and
and evaluating the self-energy on-shell, one obtains reads
n p 2 np(np + 1)
t p, t + 2np, t + 1 qp(q p) + pq( p + q ) Jt = 2v20 0<p . (286)
qR, t = v20
0 < p q, t
R
p, t + q p, t
R R
p, t + p + q , t
R R Rp
(283) As a consequence, the resulting change of the phonon density
with the vertex prefactor v0 = v(1, 1), see the definition below for small momenta |q| is linearly proportional to |q|, i.e.
equation(275). A similar equationis obtained for the Keldysh
t
on-shell self-energy qK, t. Inserting equation (283) and the nq, t = nq,0 + | q| 0<t <t Jt .
2K
+ (287)
result for the Keldysh self-energy, which we do not discuss
here but can be found in [44], into the kinetic equation(274), The momentum regime 0 < | q| < q2, defined as the regime
one finds where the non-linear contributions in |q| are negligible,
depends on the actual value of the phonon density nq, t, and
| q|
tnq = has to be determined for each specific scenario individually.
2K However, the existence of q2 is guaranteed by the above gen-
pq(q p)(npnq p nq(1 + np + nq p ))
+ 2v20 0<p<q qR + Rp + qR p
eral arguments, and the fact that t nq = 0, t = 0 for all times t.
The latter is a consequence of the U(1) symmetry of the pres-
pq(q + p)(nq + p(nq + n p + 1) nqn p ) ent setup and the global particle number conservation, as dis-
+ 4v20 0<p qR + Rp + qR+ p
. cussed in [44].
(284)
5.3.2. Scaling of the self-energy. The fact that the present
The kinetic equation(284) and the equationfor the on-shell system is described by a U(1)-invariant, massless field the-
self-energy (283) determine the forward time evolution of the ory is reflected by the absence of a scale in the self-energy
systems phonon density nq, t and self-energy qR, t in a self- q0
equation(283). One important consequence is that qR, t 0
consistent manner. For a general phonon density, both equa-
generically, i.e. the generation of a mass gap is forbidden by
tions have to be solved iteratively according to the scheme
symmetry. A further consequence of equation (283) is that,
depicted in figure12. Before the numerical results for dynam-
whenever the term in brackets obeys a scaling law,
ics in the presence of heating, i.e. the numerical solution of
equations(283) and (284), are discussed, it is useful to study t np
certain limiting cases. This facilitates the understanding of the + 2np + 1 n| p|n , (288)
Rp
numerical results in the subsequent section.
the solution for the self-energy will be a scaling function
5.3.1. Kinetics for small momenta. For sufficiently small qR = R| p|R as well. Inserting this scaling ansatz in equa-
momenta q 1, the kinetic equationsimplifies considerably. tion(283) yields
In this case, the second line of equation(284) can be discarded
completely, since the integral is performed over a very small v20n 4 +
momentum interval 0 < p < q. On the other hand, for the int- R| p|R = | p| n R I n, R (289)
R
egral in the third line of equation(284), all terms (q + p) p
for the dominant part of the integral. Therefore and identifies
56
Rep. Prog. Phys. 79 (2016) 096001 Review
With the preparations from the previous sections, one can 5.4.2. Non-equilibrium scaling. Away from the crossover
simulate the heating dynamics of an interacting Luttinger scale q xth1, the phonon density exhibits scaling behavior
liquid in terms of the time dependent phonon population nq, t and can be written as
and self-energy qR, t, as has been done in [43, 44]. Considering
c(t )| q| for | q| x1
the system initially to be in the ground state, the simula- th
tion is initialized with a phonon density nq, t=0 = 0, which nq , t = T (t ) 1 , (292)
for | q| x th
for t > 0 is continuously increased due to heating. The cen- u| q|
tral result of the analysis is a scaling solution for the self-
energies qR, t | q|R with a new non-equilibrium exponent where the functions c(t), T(t) have to be determined numer-
ically. In these regimes, according to section5.3.2, one finds
R = 5/3, which is observable in the long-wavelength regime,
scaling behavior of the phonon self-energy as well. The
i.e. on distances x > x th(t ). Here, x th(t ) marks a thermal dis-
corresponding scaling exponent is determined by
tance, below which the dynamics is dominated by thermalized
short distance modes and above which the phonon density t nq, t
fq, t + 2nq, t + 1 | q|n . (293)
increases linearly in momentum nq, t | q|, as it was the case qR, t
for the bare heating (252). The thermal distance increases sub-
ballistically x th(t ) t 4 / 5 in time, with a characteristic heat- In the heating dominated regime, nq, t = c(t )| q| and therefore
ing exponent h = 4/5, while at the same time, the effective c (t ) 1 R R > 1 c (t )
temperature describing the distribution of the short distance fq, t = |q| + c(t ) | q| + 1 | q|1 R (294)
R R
modes increases linearly in time T (t ) t.
for small momenta, since R > 1 is guaranteed by the sublead-
5.4.1. Phonon densities. The results of a numerical simula- ing nature of the interactions. This directly implies n = 1 R
tion of the phonon densities are shown in figure13. One can 5
and a super-diffusive exponent R = 3 in this regime, which
clearly identify the crossover from an interaction dominated
T (t ) lacks any equilibrium counterpart.
thermal regime with nq, t = u | q| at large momenta to a heating
On the other hand, in the large momentum regime fq, t is
dominated regime with nq, t | q|. The crossover momentum obviously dominated by the term proportional to nq, t and
between the two regimes represents the inverse thermal length n = 1, which leads to the known thermal equilibrium expo-
(x th(t ))1 t 4 / 5. 3
nent R = 2 [320] and a thermal scaling behavior for large
In the large momentum regime, the dominant contribution
to the kinetic equationis the collision term, which establishes momenta not only in the distribution function but also in the
an approximate detailed balance between phonon absorption self-energy. The crossover between the two scaling regimes
and emission in the presence of the heating. This is expressed can be estimated by equating the two relevant terms, which
by the evolution of the phonon density towards a Bose tend to dominate fq, t in the corresponding regimes. This yields
Einstein distribution, which is the fixed point of the collision t n qth
T (t ) = qRth, (295)
term alone. In this regime nq, t = nB(u| q| , T (t )) u | q|
with 2n qth + 1
very good agreement and the only indicator of the permanent
heating is the continuously increasing temperature T (t ) t. for qth = xth1 and can be used to analytically estimate the
This is contrasted by the evolution of the phonon density in the scaling of x th(t ) in time [43], resulting in x th(t ) t 4 / 5, which
low momentum regime, which is dominated by strong phonon agrees well with the numerical findings.
production. In this regime, the scattering of high-momentum
phonons into low-momentum modes enhances the effect of 5.4.3. Observability. The universal scaling of the phonon
the heating and leads, due to the structure of the vertex, to self-energies as well as the scaling regimes of the phonon
57
Rep. Prog. Phys. 79 (2016) 096001 Review
58
Rep. Prog. Phys. 79 (2016) 096001 Review
Hamilton dynamics gives rise to an additional relevant direc- Lukin, J Marino, S Sachdev, P Strack, U Tuber, and J Toner
tion in parameter space, leading to a new kind of absorbing for collaboration on projects discussed in this review. We
state phase transition without immediate counterpart in mod- are also grateful for inspiring and useful discussions with
els of classical origin [327]. many people at the KITP workshop Many-Body Physics
Certainly, progress in this respect will necessitate a more with Light. In particular, we thank A Altland, B Altshuler,
comprehensive understanding of the structure of quantum I Carusotto, A Daley, A Gorshkov, M Hafezi, M Hartmann,
dynamical field theories. One relevant issue is to reveal uni- R Fazio, M Fleischhauer, A Imamoglu, J Koch, I Lerner, I
versal aspects of the low frequency dynamics tied to the pres- Lesanovsky, P Rabl, A Rosch, G Shlyapnikov, M Szymanska,
ence of conservation laws. A concrete goal is the systematic J Taylor and H Treci for critical remarks and feedback on
construction of dynamical slow modes on the basis of the the manuscript. We are indebted to J Koch for proofreading
symmetries (and their breaking patterns) of the Keldysh func- the manuscript. We acknowledge support by the Austrian
tional integral. Science Fund (FWF) through the START grant Y 581-N16
Topology in open quantum systemsRecently, experiments and the German Research Foundation (DFG) via ZUK 64 and
with photonic lattices have started to address quantum Hall through the Institutional Strategy of the University of Cologne
physics in driven open quantum systems [328, 329]. Although within the German Excellence Initiative (ZUK 81). SD also
these systems can be idealized to some extent as closed sys- acknowledges support by the European Research Council
tems, it is a fundamental challenge to explore the fate of the (ERC) under the European Unions Horizon 2020 research
quantum Hall effector more generally, physical phenomena and innovation programme (grant agreement No 647434) and
related to topologyunder general non-equilibrium condi- by the Kavli Institute for Theoretical Physics, via the National
tions. Another angle is provided by theoretical proposals, Science Foundation under Grant No. NSF PHY11-25915, and
where drive and/or dissipation do not occur as a small per- LMS support from the European Research Council through
turbation, but rather as the dominant resource of many-body Synergy Grant UQUAM.
dynamics, guiding the system density matrix into topologi-
cally nontrivial states, which sometimes even do not have
Appendix A. Functional differentiation
a direct equilibrium counterpart. This concerns topologically
non-trivial dark states in driven open atomic fermion sys-
In this appendix, we give a brief account of functional or
tems [70, 330] and periodically driven (Floquet) dynamics
variational differentiation, following the presentation in
[331, 332] alike.
[233]. The most transparent way to introduce the basic rela-
The density matrices describing such systems typically do
tions of functional differentiation is by drawing an analogy
not correspond to pure, but rather to mixed states. While such
to the familiar formulas of partial differentiation. Indeed,
density matrices can still host non-trivial topological proper-
we can consider a field (x ) with x Rd to be the continuum
ties [216, 333, 334], the extent to which this translates into
limit of a function i defined on lattice points i Zd. With this
physically observable consequences in the correlations and
identification, relations involving partial derivatives can be
responses to (artificial) external gauge fields is at the moment
translated into corresponding ones for functional derivatives
a wide open issue. This calls for the development of non-
simply by replacing discrete indices i by continuous ones x,
equilibrium topological field theories in the framework of
the Keldysh functional integral: the established equilibrium sums i by integrals = d dx, and partial derivatives /i
x
counterpart has proven to be able to efficiently describe both by functional derivatives /(x ). Following this prescription,
bulk correlations and responses to external gauge potentials, the basic formula i /j = ij leads to
as well as to provide a proper notion of the bulkboundary
correspondence (see e.g. [69]), giving access to the edge phys- (x )
= ( x x ). (A.1)
ics in both non-interacting and interacting systems. (x )
DynamicsAddressing the time evolution of open sys- When we are working with complex fields, and are usu-
tems adds a new twist to the question of thermalization ally treated as independent variables. Then, the generalization
[32, 37, 335, 336] or, more generally, equilibration of of
quantum systems. This is also a necessary step to achieve a
ij j = Akj j
realistic description of broad classes of experiments, in part-
k i, j i
A (A.2)
icular, with ultracold atoms, as well as certain solid states sys-
j
tems in pump-probe setups [337, 338]. While first instances
of universal behavior have been identified in low dimension reads
in the short [43] and long [45, 47, 51] dynamics, it is certainly
fair to say that general principles so far remain elusive.
(x ) y,y ( y)A( y, y)( y) = y A(x, y)( y). (A.3)
59
Rep. Prog. Phys. 79 (2016) 096001 Review
given by
= (det D )1e q
1
i (q )D ( q ) ( q )
. (B.8)
i
x,x (x ) D (x , x ) (x ) + i
T
x j (x ) (x )
T
D [ ] e 2
i x,x j(x )T D1(x, x )j(x ), References
= (det D )1 / 2e 2 (B.1)
where = dd x, j(x ) = ( j1 (x ), , jn (x )), and the inverse of [1] HohenbergP and HalperinB 1977 Theory of dynamic critical
x
the integral kernel D(x, x ) is defined by means of the relation phenomena Rev. Mod. Phys. 4943579
[2] Halpin-HealyT and ZhangY-C 1995 Kinetic roughening
phenomena, stochastic growth, directed polymers and all
D(x, )D1( , x) = (x x)1. (B.2) that. Aspects of multidisciplinary statistical mechanics
Phys. Rep. 254215414
The above formula is valid for invertible symmetric kernels, [3] JensenHJ 1998 Self Organized Criticality (Cambridge:
Cambridge University Press)
D(x, x ) = D(x , x )T , with positive semi-definite imaginary [4] HinrichsenH 2000 Non-equilibrium critical phenomena
part. If D(x, x ) is not symmetric, on the LHS of equation(B.1) and phase transitions into absorbing states Adv. Phys.
it can be replaced by the symmetrized kernel 49815958
[5] TuberUC 2014 Critical Dynamics: a Field Theory Approach
1
D (x, x ) = [D(x, x ) + D(x , x )T ] , (B.3) to Equilibrium and Non-Equilibrium Scaling Behavior
2 (Cambridge: Cambridge University Press)
[6] BlochI, DalibardJ and ZwergerW 2008 Many-body physics
leaving the value of the exponent invariant. Then, equa- with ultracold gases Rev. Mod. Phys. 80885964
tion(B.1) can again be applied. [7] LewensteinM, SanperaA, AhufingerV, DamskiB, Sen(De)A
In case that the integral kernel is translationally invariant, and SenU 2007 Ultracold atomic gases in optical lattices:
i.e. it satisfies D(x, x ) = D(x x ), it is advantageous to work mimicking condensed matter physics and beyond
in Fourier space. Then, equation(B.1) can be written as Adv. Phys. 56243379
[8] Ignacio CiracJ and ZollerP 2012 Goals and opportunities in
i quantum simulation Nat. Phys. 82646
D [ ] e 2 q j(q)T (q)
(q )T D(q )(q ) + i
q [9] BaumannK, GuerlinC, BrenneckeF and EsslingerT 2010
Dicke quantum phase transition with a superfluid gas in an
i q j(q)T D1(q)j(q), optical cavity Nature 46413016
= (det D )1 / 2e 2 (B.4)
[10] RitschH, DomokosP, BrenneckeF and EsslingerT 2013 Cold
dq atoms in cavity-generated dynamical optical potentials
where we are using the shorthand notation (2 )d . The Rev. Mod. Phys. 85553601
q
Fourier transformation turns the convolution in equation(B.3) [11] SchauszP, CheneauM, EndresM, FukuharaT, HildS,
into a multiplication, showing that D1(q) can be obtained by OmranA, PohlT, GrossC, KuhrS and BlochI 2012
Observation of spatially ordered structures in a
inversion of the matrix D(q), two-dimensional Rydberg gas Nature 4918791
[12] KasprzakJ etal 2006 BoseEinstein condensation of exciton
D(q )D1(q ) = 1. (B.5) polaritons Nature 44340914
60
Rep. Prog. Phys. 79 (2016) 096001 Review
[13] CarusottoI and CiutiC 2013 Quantum fluids of light general theory and response to quantum quenches Phys.
Rev. Mod. Phys. 85299366 Rev. A 92043802
[14] HouckAA, TreciHE and KochJ 2012 On-chip quantum [37] GringM, KuhnertM, LangenT, KitagawaT, RauerB,
simulation with superconducting circuits Nat. Phys. SchreitlM, MazetsI, Adu SmithD, DemlerE and
82929 SchmiedmayerJ 2012 Relaxation and prethermalization in
[15] UnderwoodDL, ShanksWE, KochJ and HouckAA 2012 an isolated quantum system Science 337131822
Low-disorder microwave cavity lattices for quantum [38] TrotzkyS, ChenY-A, FleschA, McCullochIP, SchollwckU,
simulation with photons Phys. Rev. A 86023837 EisertJ and BlochI 2012 Probing the relaxation towards
[16] BrittonJW, SawyerBC, KeithAC, Joseph WangC-C, equilibrium in an isolated strongly correlated one-
FreericksJK, UysH, BiercukMJ and BollingerJJ 2012 dimensional bose gas Nat. Phys. 832530
Engineered two-dimensional Ising interactions in a trapped- [39] MeinertF, MarkMJ, KirilovE, LauberK, WeinmannP,
ion quantum simulator with hundreds of spins Nature DaleyAJ and NgerlH-C 2013 Quantum quench in
48448992 an atomic one-dimensional ising chain Phys. Rev. Lett.
[17] BlattR and RoosCF 2012 Quantum simulations with trapped 111053003
ions Nat. Phys. 827784 [40] HungC-L, GurarieV and ChinC 2013 From cosmology
[18] LechnerW and ZollerP 2013 From classical to quantum glasses to cold atoms: observation of sakharov oscillations in a
with ultracold polar molecules Phys. Rev. Lett. 111185306 quenched atomic superfluid Science 34112135
[19] AltmanE, SiebererLM, ChenL, DiehlS and TonerJ 2015 [41] JaskulaJ-C, PartridgeGB, BonneauM, LopesR, RuaudelJ,
Two-dimensional superfluidity of exciton polaritons BoironD and WestbrookCI 2012 Acoustic analog to the
requires strong anisotropy Phys. Rev. X 5011017 dynamical Casimir effect in a BoseEinstein condensate
[20] PiazzaF and StrackP 2014 Umklapp superradiance with a Phys. Rev. Lett. 109220401
collisionless quantum degenerate fermi gas Phys. Rev. Lett. [42] NardinG, LagoudakisKG, WoutersM, RichardM, BaasA,
112143003 AndrR, DangLS, PietkaB and Deveaud-PldranB 2009
[21] KeelingJ, BhaseenMJ and SimonsBD 2014 Fermionic Dynamics of long-range ordering in an exciton-polariton
superradiance in a transversely pumped optical cavity Phys. condensate Phys. Rev. Lett. 103256402
Rev. Lett. 112143002 [43] BuchholdM and DiehlS 2015 Nonequilibrium universality in
[22] KollathC, SheikhanA, WolffS and BrenneckeF 2016 the heating dynamics of interacting Luttinger liquids
Ultracold fermions in a cavity-induced artificial magnetic Phys. Rev. A 92013603
field Phys. Rev. Lett. 116060401 [44] BuchholdM and DiehlS 2015 Kinetic theory for interacting
[23] RafteryJ, SadriD, SchmidtS, TreciHE and HouckAA Luttinger liquids Eur. Phys. J. D 69120
2014 Observation of a dissipation-induced classical to [45] PolettiD, BarmettlerP, GeorgesA and KollathC 2013
quantum transition Phys. Rev. X 4031043 Emergence of glasslike dynamics for dissipative and
[24] HartmannMJ, BrandoFGSL and PlenioMB 2008 strongly interacting bosons Phys. Rev. Lett. 111195301
Quantum many-body phenomena in coupled cavity arrays [46] CaiZ and BarthelT 2013 Algebraic versus exponential
Laser Photonics Rev. 252756 decoherence in dissipative many-particle systems Phys. Rev.
[25] Dalla TorreEG, DemlerE, GiamarchiT and AltmanE 2010 Lett. 111150403
Quantum critical states and phase transitions in the presence [47] SchachenmayerJ, PolletL, TroyerM and DaleyAJ 2014
of non-equilibrium noise Nat. Phys. 680610 Spontaneous emission and thermalization of cold bosons in
[26] SiebererLM, HuberSD, AltmanE and DiehlS 2013 optical lattices Phys. Rev. A 89011601
Dynamical critical phenomena in drivendissipative [48] DaleyAJ 2014 Quantum trajectories and open many-body
systems Phys. Rev. Lett. 110195301 quantum systems Adv. Phys. 6377149
[27] van HorssenM and GarrahanJP 2015 Open quantum [49] OlmosB, LesanovskyI and GarrahanJP 2012 Facilitated
reaction-diffusion dynamics: absorbing states and relaxation spin models of dissipative quantum glasses Phys. Rev. Lett.
Phys. Rev. E 91032132 109020403
[28] MarinoJ and DiehlS 2016 Driven Markovian quantum [50] MarcuzziM, LeviE, LiW, GarrahanJP, OlmosB and
criticality Phys. Rev. Lett. 116070407 LesanovskyI 2014 Emergence of directed percolation in
[29] HsiangJ-T and HuBL 2015 Nonequilibrium steady state the dynamics of dissipative Rydberg gases arXiv:1411.7984
in open quantum systems: influence action, stochastic [51] LangJ and PiazzaF 2016 Critical relaxation with
equationand power balance Ann. Phys. 36213969 overdamped quasi-particles in drivendissipative systems
[30] ReyAM, HuBL, CalzettaE, RouraA and ClarkCW arXiv:1602.05102
2004 Nonequilibrium dynamics of optical-lattice-loaded [52] PoyatosJF, CiracJI and ZollerP 1996 Quantum reservoir
BoseEinstein-condensate atoms: beyond the Hartree engineering with laser cooled trapped ions Phys. Rev. Lett.
FockBogoliubov approximation Phys. Rev. A 69033610 77472831
[31] BergesJ, BorsnyiSz and WetterichC 2004 Prethermalization [53] BoseS, KnightPL, PlenioMB and VedralV 1999 Proposal
Phys. Rev. Lett. 93142002 for teleportation of an atomic state via cavity decay Phys.
[32] CalabreseP and CardyJ 2006 Time dependence of correlation Rev. Lett. 83515861
functions following a quantum quench Phys. Rev. Lett. [54] DiehlS, MicheliA, KantianA, KrausB, BchlerHP and
96136801 ZollerP 2008 Quantum states and phases in driven open
[33] MitraA and GiamarchiT 2011 Mode-coupling-induced quantum systems with cold atoms Nat. Phys. 487883
dissipative and thermal effects at long times after a quantum [55] VerstraeteF, WolfMM and IgnacioJ 2009 Quantum
quench Phys. Rev. Lett. 107150602 computation and quantum-state engineering driven by
[34] KogheeS and WoutersM 2014 Dynamical Casimir emission dissipation Nat. Phys. 56336
from polariton condensates Phys. Rev. Lett. 112036406 [56] WeimerH, MllerM, LesanovskyI, ZollerP and BchlerHP
[35] CarusottoI, BalbinotR, FabbriA and RecatiA 2010 Density 2010 A Rydberg quantum simulator Nat. Phys. 63828
correlations and analog dynamical Casimir emission of [57] MllerM, DiehlS, PupilloG and ZollerP 2012 Engineered
Bogoliubov phonons in modulated atomic BoseEinstein open systems and quantum simulations with atoms and ions
condensates Eur. Phys. J. D 56391404 Advances in Atomic, Molecular and Optical Physics vol 61,
[36] LarrP- and CarusottoI 2015 Propagation of a quantum ed Paul R Berman, Ennio Arimondo and Chun C Lin
fluid of light in a cavityless nonlinear optical medium: (New York: Academic) pp 180
61
Rep. Prog. Phys. 79 (2016) 096001 Review
[58] KrauterH, MuschikCA, JensenK, WasilewskiW, [81] GardinerCW and ZollerP 2000 Quantum Noise (Springer
PetersenJM, Ignacio CiracJ and PolzikES 2011 Series in Synergetics vol 56) 2nd edn (Berlin: Springer)
Entanglement generated by dissipation and steady state [82] BreuerH-P and PetruccioneF 2002 The Theory of Open
entanglement of two macroscopic objects Phys. Rev. Lett. Quantum Systems (Oxford: Oxford University Press)
107080503 [83] CarusottoI and CiutiC 2005 Spontaneous microcavity-
[59] BarreiroJT, MllerM, SchindlerP, NiggD, MonzT, polariton coherence across the parametric threshold:
ChwallaM, HennrichM, RoosCF, ZollerP and BlattR quantum Monte Carlo studies Phys. Rev.
2011 An open-system quantum simulator with trapped ions B 72125335
Nature 47048691 [84] GoldenfeldN 1992 Lectures on Phase Transitions and the
[60] SchindlerP, MullerM, NiggD, BarreiroJT, MartinezEA, Renormalization Group (Frontiers in Physics Perseus
HennrichM, MonzT, DiehlS, ZollerP and BlattR 2013 Books Reading) (Reading, MA: Addison-Wesley)
Quantum simulation of dynamical maps with trapped ions [85] CardyJ 1996 Scaling and Renormalization in Statistical
Nat. Phys. 93617 Physics (Cambridge Lecture Notes in Physics)
[61] EisertJ and ProsenT 2010 Noise-driven quantum criticality (Cambridge: Cambridge University Press)
arXiv:1012.5013 [86] Zinn-JustinJ 2002 Quantum Field Theory and Critical
[62] KastoryanoMJ, ReiterF and SrensenAS 2011 Dissipative Phenomena (International Series of Monographs on
preparation of entanglement in optical cavities Phys. Rev. Physics vol 113) 4th edn (Oxford: Oxford University
Lett. 106090502 Press)
[63] HorstmannB, Ignacio CiracJ and GiedkeG 2013 Noise- [87] DaleyAJ, TaylorJM, DiehlS, BaraandM 2009 Atomic
driven dynamics and phase transitions in fermionic systems three-body loss as a dynamical three-body interaction
Phys. Rev. A 87012108 Phys. Rev. Lett. 102040402
[64] HningM, MoosM and FleischhauerM 2012 Critical [88] BernierJ-S, BarmettlerP, PolettiD and KollathC 2013
exponents of steady-state phase transitions in fermionic Emergence of spatially extended pair coherence through
lattice models Phys. Rev. A 86013606 incoherent local environmental coupling Phys. Rev. A
[65] KapitE, HafeziM and SimonSH 2014 Induced self- 87063608
stabilization in fractional quantum hall states of light Phys. [89] BonnesL, CharrierD and LuchliAM 2014 Dynamical
Rev. X 4031039 and steady-state properties of a BoseHubbard chain
[66] LangN and BchlerHP 2015 Exploring quantum phases by with bond dissipation: a study based on matrix product
driven dissipation Phys. Rev. A 92012128 operators Phys. Rev. A 90033612
[67] KitaevA 2006 Anyons in an exactly solved model and beyond [90] BonnesL and LuchliAM 2014 Superoperators versus
Ann. Phys. 3212111 trajectories for matrix product state simulations of open
[68] HasanMZ and KaneCL 2010 Colloquium: Topological quantum system: a case study (arXiv:1411.4831)
insulators Rev. Mod. Phys. 82304567 [91] SchwingerJ 1961 Brownian motion of a quantum oscillator
[69] QiX-L and ZhangS-C 2011 Topological insulators and J. Math. Phys. 2407
superconductors Rev. Mod. Phys. 831057110 [92] SchwingerJ 1960 Field theory of unstable particles Ann.
[70] DiehlS, RicoE, BaraandMA and ZollerP 2011 Topology Phys. 916993
by dissipation in atomic quantum wires Nat. Phys. [93] KeldyshLV 1965 Diagram technique for nonequilibrium
79717 processes Sov. Phys.JETP 20101826
[71] BardynC-E, BaranovMA, RicoE, ImamogluA, ZollerP and [94] MahanthappaK 1962 Multiple production of photons in
DiehlS 2012 Majorana modes in drivendissipative atomic quantum electrodynamics Phys. Rev. 12632940
superfluids with a zero Chern number Phys. Rev. Lett. [95] BakshiPM and MahanthappaKT 1963 Expectation value
109130402 formalism in quantum field theory. I J. Math. Phys. 41
[72] BudichJC, ZollerP and DiehlS 2015 Dissipative preparation [96] BakshiPM and MahanthappaKT 1963 Expectation value
of Chern insulators Phys. Rev. A 91042117 formalism in quantum field theory. II J. Math. Phys.
[73] KardarM, ParisiG and ZhangY-C 1986 Dynamic scaling of 412
growing interfaces Phys. Rev. Lett. 5688992 [97] LiACY, PetruccioneF and KochJ 2014 Perturbative
[74] PolettiD, BernierJ-S, GeorgesA and KollathC 2012 approach to Markovian open quantum systems Sci. Rep.
Interaction-induced impeding of decoherence and 44887
anomalous diffusion Phys. Rev. Lett. 109045302 [98] LiACY, PetruccioneF and KochJ 2015 Resummation for
[75] LesanovskyI and GarrahanJP 2013 Kinetic constraints, nonequilibrium perturbation theory and application to
hierarchical relaxation and onset of glassiness in strongly open quantum lattices arXiv:1511.01551
interacting and dissipative Rydberg gases Phys. Rev. Lett. [99] Degenfeld-SchonburgP, Navarrete-BenllochC and
111215305 HartmannMJ 2015 Self-consistent projection operator
[76] LesanovskyI and GarrahanJP 2014 Out-of-equilibrium theory in nonlinear quantum optical systems: a case study
structures in strongly interacting Rydberg gases with on degenerate optical parametric oscillators Phys. Rev. A
dissipation Phys. Rev. A 90011603 91053850
[77] OlmosB, LesanovskyI and GarrahanJP 2014 Out-of- [100] WeimerH 2015 Variational principle for steady states of
equilibrium evolution of kinetically constrained many-body dissipative quantum many-body systems Phys. Rev. Lett.
quantum systems under purely dissipative dynamics Phys. 114040402
Rev. E 90042147 [101] FinazziS, Le BoitA, StormeF, BaksicA and CiutiC
[78] DiehlS, YiW, DaleyAJ and ZollerP 2010 Dissipation- 2015 Corner-space renormalization method for driven
induced d-wave pairing of fermionic atoms in an optical dissipative two-dimensional correlated systems Phys. Rev.
lattice Phys. Rev. Lett. 105227001 Lett. 115080604
[79] DalibardJ, CastinY and MlmerK 1992 Wave-function [102] KamenevA 2011 Field Theory of Non-Equilibrium Systems
approach to dissipative processes in quantum optics Phys. (Cambridge: Cambridge University Press)
Rev. Lett. 685803 [103] CalzettaEA and HuB-LB 2008 Nonequilibrium Quantum
[80] PlenioMB and KnightPL 1998 The quantum-jump approach Field Theory (Cambridge: Cambridge University Press)
to dissipative dynamics in quantum optics Rev. Mod. Phys. [104] JarzynskiC 1997 Nonequilibrium equality for free energy
7010144 differences Phys. Rev. Lett. 7826903
62
Rep. Prog. Phys. 79 (2016) 096001 Review
[105] JarzynskiC 1997 Equilibrium free-energy differences [127] WangYK and HioeFT 1973 Phase transition in the Dicke
from nonequilibrium measurements: a master-equation model of superradiance Phys. Rev. A 78316
approach Phys. Rev. E 56501835 [128] EmaryC and BrandesT 2003 Chaos and the quantum phase
[106] CrooksGE 1999 Entropy production fluctuation theorem transition in the Dicke model Phys. Rev. E 67066203
and the nonequilibrium work relation for free energy [129] DimerF, EstienneB, ParkinsAS and CarmichaelHJ 2007
differences Phys. Rev. E 6027216 Proposed realization of the Dicke-model quantum phase
[107] CardyJ 1996 Renormalisation group approach to reaction- transition in an optical cavity QED system Phys. Rev. A
diffusion problems arXiv:cond-mat/9607163 75013804
[108] DoiM 1976 Second quantization representation for classical [130] MurchKW, MooreKL, GuptaS and Stamper-KurnDM
many-particle system J. Phys. A. Math. Gen. 9146577 2008 Observation of quantum-measurement backaction
[109] PelitiL 1985 Path integral approach to birth-death processes with an ultracold atomic gas Nat. Phys. 45614
on a lattice J. Phys. 46146983 [131] StrackP and SachdevS 2011 Dicke quantum spin glass of
[110] DiehlS, TomadinA, MicheliA, FazioR and ZollerP atoms and photons Phys. Rev. Lett. 107277202
2010 Dynamical phase transitions and instabilities [132] AngelakisDG, SantosMF and BoseS 2007 Photon-
in open atomic many-body systems Phys. Rev. Lett. blockade-induced Mott transitions and XY spin models in
105015702 coupled cavity arrays Phys. Rev. A 76031805
[111] SachdevS 2011 Quantum Phase Transitions 2nd edn [133] HartmannMJ, BrandaoFGSL and PlenioMB 2006
(Cambridge: Cambridge University Press) Strongly interacting polaritons in coupled arrays of
[112] Dalla TorreEG, DiehlS, LukinM, SachdevS and StrackP cavities Nat. Phys. 284955
2013 Keldysh approach for nonequilibrium phase [134] SchmidtS and KochJ 2013 Circuit QED lattices: towards
transitions in quantum optics: beyond the Dicke model in quantum simulation with superconducting circuits
optical cavities Phys. Rev. A 87023831 Ann. Phys. 525395412
[113] BuchholdM, StrackP, SachdevS and DiehlS 2013 Dicke- [135] SchirM, BordyuhM, ztopB and TreciHE 2013
model quantum spin and photon glass in optical cavities: Quantum phase transition of light in the RabiHubbard
nonequilibrium theory and experimental signatures Phys. model J. Phys. B: At. Mol. Opt. Phys. 46224021
Rev. A 87063622 [136] SchirM, BordyuhM, OztopB and TreciHE 2012 Phase
[114] MaghrebiMF and GorshkovAV 2016 Nonequilibrium transition of light in cavity QED lattices Phys. Rev. Lett.
many-body steady states via Keldysh formalism Phys. Rev. 109053601
B 93014307 [137] SchirM, JoshiC, BordyuhM, FazioR, KeelingJ and
[115] PiazzaF and StrackP 2014 Quantum kinetics of ultracold TreciHE 2016 Exotic attractors of the non-equilibrium
fermions coupled to an optical resonator Phys. Rev. A RabiHubbard model Phys. Rev. Lett. 116 143603
90043823 [138] ZouLJ, MarcosD, DiehlS, PutzS, SchmiedmayerJ,
[116] BernierJ-S, PolettiD and KollathC 2014 Dissipative MajerJ and RablP 2014 Implementation of the Dicke
quantum dynamics of fermions in optical lattices: a slave- lattice model in hybrid quantum system arrays Phys. Rev.
spin approach Phys. Rev. B 90205125 Lett. 113023603
[117] MabuchiH and DohertyAC 2002 Cavity quantum [139] Ruiz-RivasJ, del ValleE, GiesC, GartnerP and
electrodynamics: coherence in context Science 29813727 HartmannMJ 2014 Spontaneous collective coherence in
[118] HoodCJ, LynnTW, DohertyAC, ParkinsAS and driven dissipative cavity arrays Phys. Rev. A 90033808
KimbleHJ 2000 The atom-cavity microscope: single [140] MarcosD, TomadinA, DiehlS and RablP 2012 Photon
atoms bound in orbit by single photons Science condensation in circuit quantum electrodynamics by
287144753 engineered dissipation New J. Phys. 14055005
[119] NagyD, KnyaG, SzirmaiG and DomokosP 2010 Dicke- [141] TomadinA and FazioR 2010 Many-body phenomena in
model phase transition in the quantum motion of a QED-cavity arrays J. Opt. Soc. Am. B 27A1306
BoseEinstein condensate in an optical cavity Phys. Rev. [142] KepesidisKV and HartmannMJ 2012 BoseHubbard model
Lett. 104130401 with localized particle losses Phys. Rev. A 85063620
[120] BaumannK, MottlR, BrenneckeF and EsslingerT 2011 [143] Le BoitA, OrsoG and CiutiC 2013 Steady-state phases
Exploring symmetry breaking at the dicke quantum phase and tunneling-induced instabilities in the driven dissipative
transition Phys. Rev. Lett. 107140402 BoseHubbard model Phys. Rev. Lett. 110233601
[121] MaschlerC and RitschH 2005 Cold atom dynamics in [144] JinJ, RossiniD, LeibM, HartmannMJ and FazioR 2014
a quantum optical lattice potential Phys. Rev. Lett. Steady-state phase diagram of a driven QED-cavity
95260401 array with cross-Kerr nonlinearities Phys. Rev. A
[122] ztopB, BordyuhM, Mstecaplolu E and TreciHE 90023827
2012 Excitations of optically driven atomic condensate in [145] Le BoitA, OrsoG and CiutiC 2014 Bose-Hubbard model:
a cavity: theory of photodetection measurements New J. relation between drivendissipative steady states and
Phys. 14085011 equilibrium quantum phases Phys. Rev. A 90063821
[123] KulkarniM, ztopB and TreciHE 2013 Cavity-mediated [146] BiellaA, MazzaL, CarusottoI, RossiniD and FazioR
near-critical dissipative dynamics of a driven condensate 2015 Photon transport in a dissipative chain of nonlinear
Phys. Rev. Lett. 111220408 cavities Phys. Rev. A 91053815
[124] LandigR, BrenneckeF, MottlR, DonnerT and EsslingerT [147] DengH, HaugH and YamamotoY 2010 Exciton-polariton
2015 Measuring the dynamic structure factor of a quantum BoseEinstein condensation Rev. Mod. Phys. 821489537
gas undergoing a structural phase transition Nat. Commun. [148] ByrnesT, KimNY and YamamotoY 2014 Exciton-polariton
67046 condensates Nat. Phys. 1080313
[125] BrenneckeF, MottlR, BaumannK, LandigR, DonnerT and [149] KeelingJ, SzymaskaMH and LittlewoodPB 2010
EsslingerT 2013 Real-time observation of fluctuations at Keldysh Greens function approach to coherence in a
the drivendissipative Dicke phase transition Proc. Natl non-equilibrium steady state: connecting Bose-Einstein
Acad. Sci. USA 110117637 condensation and lasing Optical Generation and Control
[126] HeppK and LiebEH 1973 On the superradiant phase of Quantum Coherence in Semiconductor Nanostructures
transition for molecules in a quantized radiation field: the (NanoScience and Technology ed G Slavcheva and
Dicke maser model Ann. Phys. 76360404 P Roussignol) (Berlin: Springer)
63
Rep. Prog. Phys. 79 (2016) 096001 Review
[150] SzymaskaMH, KeelingJ and LittlewoodPB 2012 Non- [171] LeibfriedD, BlattR, MonroeC and WinelandD 2003
equilibrium BoseEinstein condensation in a dissipative Quantum dynamics of single trapped ions Rev. Mod. Phys.
environment (arXiv:1206.1784) 75281324
[151] KlaersJ, SchmittJ, VewingerF and WeitzM 2010 Bose [172] MarcuzziM, LeviE, DiehlS, GarrahanJP and LesanovskyI
Einstein condensation of photons in an optical microcavity 2014 Universal nonequilibrium properties of dissipative
Nature 4685458 Rydberg gases Phys. Rev. Lett. 113210401
[152] DemokritovSO, DemidovVE, DzyapkoO, MelkovGA, [173] MarquardtF and GirvinSM 2009 Optomechanics Physics
SergaAA, HillebrandsB and SlavinAN 2006 Bose 240
Einstein condensation of quasi-equilibrium magnons at [174] LindbladG 1976 On the generators of quantum dynamical
room temperature under pumping Nature 4434303 semigroups Commun. Math. Phys. 4811930
[153] AlloingM, BeianM, LewensteinM, FusterD, GonzlezY, [175] KossakowskiA 1972 On quantum statistical mechanics of
GonzlezL, CombescotR, CombescotM and DubinF non-hamiltonian systems Rep. Math. Phys. 324774
2014 Evidence for a bose-einstein condensate of excitons [176] BaconD, ChildsAM, ChuangIL, KempeJ, LeungDW and
Europhys. Lett. 10710012 ZhouX 2001 Universal simulation of Markovian quantum
[154] FalkenauM, VolchkovVV, RhrigJ, GriesmaierA and dynamics Phys. Rev. A 64062302
PfauT 2011 Continuous loading of a conservative potential [177] DaviesEB 1974 Markovian master equations Commun.
trap from an atomic beam Phys. Rev. Lett. 106163002 Math. Phys. 3991
[155] MewesM-O, AndrewsMR, KurnDM, DurfeeDS, [178] AltlandA and SimonsB 2010 Condensed Matter Field
TownsendCG and KetterleW 1997 Output coupler for Theory 2nd edn (Cambridge: Cambridge University Press)
BoseEinstein condensed atoms Phys. Rev. Lett. 785825 [179] FeynmanR 1948 Space-time approach to non-relativistic
[156] RobinsNP, FiglC, JeppesenM, DennisGR and CloseJD quantum mechanics Rev. Mod. Phys. 2036787
2008 A pumped atom laser Nat. Phys. 47316 [180] NegeleJW and OrlandH 1998 Quantum Many-Particle
[157] RobinsNP, AltinPA, DebsJE and CloseJD 2013 Atom Systems (Boulder, CO: Westview Press)
lasers: production, properties and prospects for precision [181] SiebererLM, HuberSD, AltmanE and DiehlS 2014
inertial measurement Phys. Rep. 52926596 Nonequilibrium functional renormalization for driven
[158] ImamogluA, RamRJ, PauS and YamamotoY 1996 dissipative BoseEinstein condensation Phys. Rev. B
Nonequilibrium condensates and lasers without inversion: 89134310
exciton-polariton lasers Phys. Rev. A 5342503 [182] AmitDJ and Martin-MayorV 2005 Field Theory, the
[159] WoutersM and CarusottoI 2007 Excitations in a Renormalization Group and Critical Phenomena 3rd edn
nonequilibrium BoseEinstein condensate of exciton (Singapore: World Scientific)
polaritons Phys. Rev. Lett. 99140402 [183] CalzettaE and HuBL 1988 Nonequilibrium quantum fields:
[160] WoutersM and CarusottoI 2010 Superfluidity and critical closed-time-path effective action, Wigner function and
velocities in nonequilibrium BoseEinstein condensates Boltzmann equation Phys. Rev. D 372878900
Phys. Rev. Lett. 105020602 [184] PatashinskiiAZ and PokrovskiiVL 1973 Longitudinal
[161] CiutiC and CarusottoI 2006 Inputoutput theory of cavities susceptibility and correlations in degenerate systems Sov.
in the ultrastrong coupling regime: the case of time- Phys.JETP 37733
independent cavity parameters Phys. Rev. A 74033811 [185] ZwergerW 2004 Anomalous fluctuations in phases with a
[162] ChiocchettaA and CarusottoI 2014 Quantum Langevin broken continuous symmetry Phys. Rev. Lett. 92027203
model for nonequilibrium condensation Phys. Rev. A [186] WallsDF and MilburnGJ 2008 Quantum Optics 2nd edn
90023633 (Berlin: Springer)
[163] HackermllerL, SchneiderU, Moreno-CardonerM, [187] TalknerP 1986 The failure of the quantum regression
KitagawaT, BestT, WillS, DemlerE, AltmanE, BlochI hypothesis Ann. Phys. 167390436
and ParedesB 2010 Anomalous expansion of attractively [188] FordG and OConnellR 1996 There is no quantum
interacting fermionic atoms in an optical lattice Science regression theorem Phys. Rev. Lett. 77798801
32716214 [189] SiebererLM, ChiocchettaA, GambassiA, TuberUC
[164] MeinertF, MarkMJ, KirilovE, LauberK, WeinmannP, and DiehlS 2015 Thermodynamic equilibrium as a
GrbnerM, DaleyAJ and NgerlH-C 2014 Observation symmetry of the Schwinger-Keldysh action Phys. Rev. B
of many-body dynamics in long-range tunneling after a 92134307
quantum quench Science 344125962 [190] LifshitzEM and PitaevskiiLP 1980 Statistical Physics, Part
[165] PreissPM, MaR, Eric TaiM, LukinA, RispoliM, 2: Theory of the Condensed State 2nd edn (New York:
ZupancicP, LahiniY, IslamR and GreinerM 2015 Pergamon)
Strongly correlated quantum walks in optical lattices [191] WoutersM and CarusottoI 2006 Absence of long-range
Science 347122933 coherence in the parametric emission of photonic wires
[166] PichlerH, DaleyAJ and ZollerP 2010 Nonequilibrium Phys. Rev. B 74245316
dynamics of bosonic atoms in optical lattices: decoherence [192] SzymaskaM, KeelingJ and LittlewoodP 2006
of many-body states due to spontaneous emission Phys. Nonequilibrium quantum condensation in an incoherently
Rev. A 82063605 pumped dissipative system Phys. Rev. Lett. 96230602
[167] PichlerH, SchachenmayerJ, DaleyAJ and ZollerP 2013 [193] ZhouG-Z, SuZ-B, HaoB-L and YuL 1980 Closed time
Heating dynamics of bosonic atoms in a noisy optical path Greens functions and critical dynamics Phys. Rev. B
lattice Phys. Rev. A 87033606 223385407
[168] FisherMPA, WeichmanPB, GrinsteinG and FisherDS [194] ChouK-C, SuZ-B, HaoB-L and YuL 1985 Equilibrium
1989 Boson localization and the superfluid-insulator and nonequilibrium formalisms made unified Phys. Rep.
transition Phys. Rev. B 4054670 1181131
[169] JakschD, BruderC, CiracJI, GardinerCW and ZollerP [195] MartinP, SiggiaE and RoseH 1973 Statistical dynamics of
1998 Cold bosonic atoms in optical lattices Phys. Rev. classical systems Phys. Rev. A 842337
Lett. 81310811 [196] De DominicisC 1976 Techniques de renormalisation de
[170] BarnettSM and RadmorePM 1997 Methods in Theoretical la theorie des champs et dynamique des phenomenes
Quantum Optics (Oxford: Oxford University Press) critiques J. Phys. 37C24753
64
Rep. Prog. Phys. 79 (2016) 096001 Review
[197] JanssenH-K 1976 On a Lagrangean for classical field [221] WatanabeH and MurayamaH 2012 Unified description of
dynamics and renormalization group calculations of NambuGoldstone bosons without Lorentz invariance
dynamical critical properties Z. Phys. B: Condens. Matter Phys. Rev. Lett. 108251602
Quanta 2337780 [222] HidakaY 2013 Counting rule for NambuGoldstone modes
[198] BauschR, JanssenHK and WagnerH 1976 Renormalized in nonrelativistic systems Phys. Rev. Lett. 110091601
field theory of critical dynamics Z. Phys. B: Condens. [223] WatanabeH and MurayamaH 2014 Effective Lagrangian for
Matter Quanta 2411327 nonrelativistic systems Phys. Rev. X 4031057
[199] ChaikinPM and LubenskyTC 1995 Principles of condensed [224] HayataT and HidakaY 2015 Dispersion relations of
matter physics (Cambridge: Cambridge University Press) NambuGoldstone modes at finite temperature and density
[200] KuboR 1966 The fluctuation-dissipation theorem Rep. Prog. Phys. Rev. D 91056006
Phys. 2925584 [225] SnokeDW 2008 Solid State Physics: Essential Concepts
[201] ChiocchettaA, GambassiA and CarusottoI 2015 Laser (Reading, MA: Addison-Wesley)
operation and BoseEinstein condensation: analogies and [226] GrahamR and HakenH 1970 Laserlightfirst example of
differences arXiv:1503.02816 a second-order phase transition far away from thermal
[202] KuboR 1957 Statistical-mechanical theory of irreversible equilibrium Z. Phys. 2373146
processes. I. General theory and simple applications to [227] DeGiorgioV and MarlanScully 1970 Analogy between the
magnetic and conduction problems J. Phys. Soc. Japan laser threshold region and a second-order phase transition
1257086 Phys. Rev. A 211707
[203] MartinP and SchwingerJ 1959 Theory of many-particle [228] WetterichC 1993 Exact evolution equationfor the effective
systems. I Phys. Rev. 115134273 potential Phys. Lett. B 301904
[204] JakobsSG, PletyukhovM and SchoellerH 2010 [229] BergesJ, TetradisN and WetterichC 2002 Non-perturbative
Properties of multi-particle Greens and vertex renormalization flow in quantum field theory and
functions within Keldysh formalism J. Phys. A: Math. statistical physics Phys. Rep. 363223386
Theor. 43103001 [230] DelamotteB 2012 An introduction to the nonperturbative
[205] PeskinME and SchroederDV 1995 An Introduction to renormalization group Renormalization Group and
Quantum Field Theory (Boulder, CO: Westview Press) Effective Field Theory Approaches to Many-Body Systems
[206] JanssenHK 1979 Field-theoretic method applied to critical SE2 (Lecture Notes in Physics vol 852) ed ASchwenk
dynamics Dynamical Critical Phenomena and Related and JPolonyi (Berlin: Springer) pp 49132
Topics (Lecture Notes in Physics vol 104) ed CPEnz [231] BlaizotJ-P, Mndez-GalainR and WscheborN 2006 A new
(Berlin: Springer) pp 2547 method to solve the non-perturbative renormalization
[207] AronC, BiroliG and CugliandoloLF 2010 Symmetries of group equations Phys. Lett. B 6325718
generating functionals of Langevin processes with colored [232] PawlowskiJM 2007 Aspects of the functional
multiplicative noise J. Stat. Mech P11018 renormalisation group Ann. Phys. 3222831915
[208] AronC, BarciDG, CugliandoloLF, ArenasZG [233] BoettcherI, PawlowskiJM and DiehlS 2012 Ultracold
and LozanoGS 2014 Dynamical symmetries of atoms and the functional renormalization group Nucl.
Markov processes with multiplicative white noise Phys. B: Proc. Suppl. 22863135
arXiv:1412.7564 [234] JakobsSG, MedenV and SchoellerH 2007 Nonequilibrium
[209] AltlandA, De MartinoA, EggerR and NarozhnyB 2010 functional renormalization group for interacting quantum
Transient fluctuation relations for time-dependent particle systems Phys. Rev. Lett. 99150603
transport Phys. Rev. B 82115323 [235] JakobsSG, PletyukhovM and SchoellerH 2010
[210] MessiahA 1965 Quantum Mechanics II 3rd edn Nonequilibrium functional renormalization group with
(Amsterdam: North-Holland) frequency-dependent vertex function: a study of the
[211] HafeziM, AdhikariP and TaylorJM 2014 Engineering single-impurity Anderson model Phys. Rev. B 81195109
three-body interaction and Pfaffian states in circuit QED [236] KarraschC, AndergassenS, PletyukhovM, SchurichtD,
systems Phys. Rev. B 90060503 BordaL, MedenV and SchoellerH 2010 Non-equilibrium
[212] GrahamR and TlT 1990 Steady-state ensemble for the current and relaxation dynamics of a charge-fluctuating
complex GinzburgLandau equationwith weak noise quantum dot Europhys. Lett. 9030003
Phys. Rev. A 42466177 [237] MetznerW, SalmhoferM, HonerkampC, MedenV and
[213] MitraA, TakeiS, KimYB and MillisAJ 2006 SchnhammerK 2012 Functional renormalization group
Nonequilibrium quantum criticality in open electronic approach to correlated fermion systems Rev. Mod. Phys.
systems Phys. Rev. Lett. 97236808 84299352
[214] MitraA and GiamarchiT 2012 Thermalization and [238] CanetL and ChatH 2007 A non-perturbative approach to
dissipation in out-of-equilibrium quantum systems: a critical dynamics J. Phys. A: Math. Theor. 40193749
perturbative renormalization group approach Phys. Rev. B [239] MesterhzyD, StockemerJH, PalharesLF and BergesJ
85075117 2013 Dynamic universality class of Model C from the
[215] GebauerR and CarR 2004 Current in open quantum systems functional renormalization group Phys. Rev. B 88174301
Phys. Rev. Lett. 93160404 [240] RostenOJ 2012 Fundamentals of the exact renormalization
[216] AvronJE, FraasM and GrafGM 2012 Adiabatic response group Phys. Rep. 511177272
for Lindblad dynamics J. Stat. Phys. 14880023 [241] ReuterM 1998 Nonperturbative evolution equationfor
[217] GoldstoneJ 2008 Field theories with Superconductor quantum gravity Phys. Rev. D 5797185
solutions Nuovo Cim. 1915464 [242] GezziR, PruschkeTh and MedenV 2007 Functional
[218] GoldstoneJ, SalamA and WeinbergS 1962 Broken renormalization group for nonequilibrium quantum many-
symmetries Phys. Rev. 12796570 body problems Phys. Rev. B 75045324
[219] NambuY and Jona-LasinioG 1961 Dynamical model [243] BergesJ, RothkopfA and SchmidtJ 2008 Nonthermal fixed
of elementary particles based on an analogy with points: effective weak coupling for strongly correlated
superconductivity. I Phys. Rev. 12234558 systems far from equilibrium Phys. Rev. Lett. 101041603
[220] MinamiY and HidakaY 2015 Spontaneous symmetry [244] BergesJ and HoffmeisterG 2009 Nonthermal fixed points
breaking and NambuGoldstone modes in dissipative and the functional renormalization group Nucl. Phys. B
systems arXiv:1509.05042 813383407
65
Rep. Prog. Phys. 79 (2016) 096001 Review
[245] BergesJ and MesterhzyD 2012 Introduction to the [267] JiK, GladilinVN and WoutersM 2015 Temporal coherence
nonequilibrium functional renormalization group Nucl. of one-dimensional nonequilibrium quantum fluids Phys.
Phys. B: Proc. Suppl. 2283760 Rev. B 91045301
[246] GasenzerT and PawlowskiJM 2008 Towards far-from- [268] De DominicisC, BrzinE and Zinn-JustinJ 1975 Field-
equilibrium quantum field dynamics: a functional theoretic techniques and critical dynamics. I. Ginzburg
renormalisation-group approach Phys. Lett. B 67013540 Landau stochastic models without energy conservation
[247] CanetL, DelamotteB, DeloubrireO and WscheborN 2004 Phys. Rev. B 12494553
Nonperturbative renormalization-group study of reaction- [269] SivashinskyGI 1977 Nonlinear analysis of hydrodynamic
diffusion processes Phys. Rev. Lett. 92195703 instability in laminar flames I. Derivation of basic
[248] CanetL, ChatH, DelamotteB, DornicI and MuozM 2005 equations Acta Astronaut. 41177206
Nonperturbative fixed point in a nonequilibrium phase [270] KuramotoY 1984 Chemical Oscillations, Waves and
transition Phys. Rev. Lett. 95100601 Turbulence (Berlin: Springer)
[249] CanetL 2006 Reaction-diffusion processes and non- [271] GrinsteinG, DavidMukamel, SeidinR and BennettC 1993
perturbative renormalization group J. Phys. A: Math. Gen. Temporally periodic phases and kinetic roughening Phys.
39790112 Rev. Lett. 70360710
[250] CanetL, ChatH and DelamotteB 2004 Quantitative phase [272] GrinsteinG, JayaprakashC and PanditR 1996 Conjectures
diagrams of branching and annihilating random walks about phase turbulence in the complex GinzburgLandau
Phys. Rev. Lett. 92255703 equation Phys. D :Nonlinear Phenom. 9096106
[251] CanetL, ChatH and DelamotteB 2011 General framework [273] FisherMPA and GrinsteinG 1992 Nonlinear transport and
of the non-perturbative renormalization group for 1/ f noise in insulators Phys. Rev. Lett. 6923225
non-equilibrium steady states J. Phys. A: Math. Theor. [274] TuberUC and DiehlS 2014 Perturbative field-theoretical
44495001 renormalization group approach to drivendissipative
[252] CanetL, ChatH, DelamotteB and WscheborN 2010 BoseEinstein criticality Phys. Rev. X 4021010
Nonperturbative renormalization group for the Kardar [275] MerminND and WagnerH 1966 Absence of ferromagnetism
ParisiZhang equation Phys. Rev. Lett. 104150601 or antiferromagnetism in one- or two-dimensional
[253] CanetL, ChatH, DelamotteB and WscheborN 2011 isotropic Heisenberg models Phys. Rev. Lett. 1711336
Nonperturbative renormalization group for the Kardar [276] WoutersM and CarusottoI 2007 Goldstone mode of
ParisiZhang equation: general framework and first optical parametric oscillators in planar semiconductor
applications Phys. Rev. E 84061128 microcavities in the strong-coupling regime Phys. Rev. A
[254] CanetL, ChatH, DelamotteB and WscheborN 2012 76043807
Erratum: Nonperturbative renormalization group for the [277] KlossT, CanetL and WscheborN 2012 Nonperturbative
KardarParisiZhang equation: general framework and renormalization group for the stationary KardarParisi
first applications (2011 Phys. Rev. E 84 061128) Phys. Zhang equation: scaling functions and amplitude ratios in
Rev. E 8619904 1 + 1, 2 + 1 and 3 + 1 dimensions Phys. Rev. E 86051124
[255] KogutJB 1979 An introduction to lattice gauge theory and [278] LitimDF 2000 Optimisation of the exact renormalisation
spin systems Rev. Mod. Phys. 51659713 group Phys. Lett. B 486929
[256] YeJ, SachdevS and ReadN 1993 Solvable spin glass of [279] LitimDF 2001 Optimized renormalization group flows Phys.
quantum rotors Phys. Rev. Lett. 7040114 Rev. D 64105007
[257] ReadN, SachdevS and YeJ 1995 Landau theory of quantum [280] GuidaR and Zinn-JustinJ 1998 Critical exponents of the
spin glasses of rotors and ising spins Phys. Rev. B N-vector model J. Phys. A: Math. Gen. 31810321
52384410 [281] UtsunomiyaS etal 2008 Observation of Bogoliubov
[258] GopalakrishnanS, LevBL and GoldbartPM 2011 excitations in exciton-polariton condensates Nat. Phys.
Frustration and glassiness in spin models with cavity- 47005
mediated interactions Phys. Rev. Lett. 107277201 [282] StewartJT, GaeblerJP and JinDS 2008 Using
[259] MllerM, StrackP and SachdevS 2012 Quantum charge photoemission spectroscopy to probe a strongly interacting
glasses of itinerant fermions with cavity-mediated long- Fermi gas Nature 4547447
range interactions Phys. Rev. A 86023604 [283] KimJ and KosterlitzJ 1989 Growth in a restricted solid-on-
[260] HabibianH, WinterA, PaganelliS, RiegerH and MorigiG solid model Phys. Rev. Lett. 62228992
2013 Bose-glass phases of ultracold atoms due to cavity [284] MirandaVG and Aaro ReisFDA 2008 Numerical study
backaction Phys. Rev. Lett. 110075304 of the KardarParisiZhang equation Phys. Rev. E
[261] CugliandoloLF and LozanoG 1999 Real-time 77031134
nonequilibrium dynamics of quantum glassy systems [285] MarinariE, PagnaniA and ParisiG 2000 Critical exponents
Phys. Rev. B 5991542 of the KPZ equationvia multi-surface coding numerical
[262] CugliandoloLF, GrempelDR, LozanoG, LozzaH and da simulations J. Phys. A: Math. Gen. 33818192
Silva SantosCA 2002 Dissipative effects on quantum [286] GhaisasS 2006 Stochastic model in the KardarParisiZhang
glassy systems Phys. Rev. B 66014444 universality class with minimal finite size effects Phys.
[263] CugliandoloLF, GrempelDR, LozanoG and LozzaH 2004 Rev. E 73022601
Effects of dissipation on disordered quantum spin models [287] ChinC-S and den NijsM 1999 Stationary-state skewness in
Phys. Rev. B 70024422 two-dimensional KardarParisiZhang type growth Phys.
[264] SzymaskaMH, KeelingJ and LittlewoodPB 2007 Rev. E 59263341
Mean-field theory and fluctuation spectrum of a pumped [288] TangL-H, ForrestB and WolfD 1992 Kinetic surface
decaying Bose-Fermi system across the quantum roughening. II. Hypercube-stacking models Phys. Rev. A
condensation transition Phys. Rev. B 75195331 45716279
[265] HeL, SiebererLM, AltmanE and DiehlS 2015 Scaling [289] Ala-NissilaT, HjeltT, KosterlitzJM and VenlinenO
properties of one-dimensional drivendissipative 1993 Scaling exponents for kinetic roughening in higher
condensates Phys. Rev. B 92155307 dimensions J. Stat. Phys. 7220725
[266] GladilinVN, JiK and WoutersM 2014 Spatial coherence [290] CastellanoC, MarsiliM, MuozMA and PietroneroL 1999
of weakly interacting one-dimensional nonequilibrium Scale invariant dynamics of surface growth Phys. Rev. E
bosonic quantum fluids Phys. Rev. A 90023615 59646075
66
Rep. Prog. Phys. 79 (2016) 096001 Review
[291] Aaro ReisFDA 2004 Universality in two-dimensional [310] WertzE etal 2012 Propagation and amplification
KardarParisiZhang growth Phys. Rev. E 69021610 dynamics of 1D polariton condensates Phys. Rev. Lett.
[292] KellingJ and dorG 2011 Extremely large-scale simulation 109216404
of a KardarParisiZhang model using graphics cards [311] GagelP, OrthPP and SchmalianJ 2014 Universal
Phys. Rev. E 84061150 postquench prethermalization at a quantum critical point
[293] Halpin-HealyT 2012 (2 + 1)dimensional directed polymer Phys. Rev. Lett. 113220401
in a random medium: scaling phenomena and universal [312] GagelP, OrthPP and SchmalianJ 2015 Universal
distributions Phys. Rev. Lett. 109170602 postquench coarsening and aging at a quantum critical
[294] PagnaniA and ParisiG 2015 Numerical estimate of the point Phys. Rev. B 92115121
KardarParisiZhang universality class in (2 + 1) [313] ChiocchettaA, TavoraM, GambassiA and MitraA 2015
dimensions Phys. Rev. E: Stat. Nonlin. Soft Matter Phys. Short-time universal scaling in an isolated quantum system
92010101 after a quench Phys. Rev. B 91220302
[295] Halpin-HealyT 2013 Extremal paths, the stochastic heat [314] StrackP 2015 Dynamic criticality far from equilibrium: one-
equation, and the three-dimensional KardarParisiZhang loop flow of BurgersKardarParisiZhang systems with
universality class Phys. Rev. E: Stat. Nonlin. Soft Matter broken Galilean invariance Phys. Rev. E 91032131
Phys. 88042118 [315] MarcuzziM, LeviE, LiW, GarrahanJP, OlmosB and
[296] Halpin-HealyT 2013 Erratum: Extremal paths, the stochastic LesanovskyI 2015 Non-equilibrium universality in the
heat equationand the three-dimensional KardarParisi dynamics of dissipative cold atomic gases New J. Phys.
Zhang universality class (2013 Phys. Rev. E 88 042118) 1772003
Phys. Rev. E 88069903 [316] GenskeM and RoschA 2015 FloquetBoltzmann equationfor
[297] Halpin-HealyT and PalasantzasG 2014 Universal periodically driven Fermi systems Phys. Rev. A 92062108
correlators and distributions as experimental signatures [317] TavoraM, RoschA and MitraA 2014 Quench dynamics
of (2 + 1)-dimensional KardarParisiZhang growth of one-dimensional interacting bosons in a disordered
Europhys. Lett. 10550001 potential: elastic dephasing and critical speeding-up of
[298] ChiocchettaA and CarusottoI 2013 Non-equilibrium thermalization Phys. Rev. Lett. 113010601
quasi-condensates in reduced dimensions Europhys. Lett. [318] TavoraM and MitraA 2013 Quench dynamics of one-
10267007 dimensional bosons in a commensurate periodic potential:
[299] ChenL and TonerJ 2013 Universality for moving stripes: a a quantum kinetic equationapproach Phys. Rev. B
hydrodynamic theory of polar active smectics Phys. Rev. 88115144
Lett. 111088701 [319] ProtopopovIV, GutmanDB, OldenburgM and MirlinAD
[300] RoumposG etal 2012 Power-law decay of the spatial 2014 Dissipationless kinetics of one-dimensional
correlation function in exciton-polariton condensates Proc. interacting fermions Phys. Rev. B 89161104
Natl Acad. Sci. USA 109646772 [320] AndreevAF 1980 The hydrodynamics of two- and one-
[301] NitscheWH, KimNY, RoumposG, SchneiderC, dimensional liquids Sov. Phys.JETP 511038
KampM, HflingS, ForchelA and YamamotoY 2014 [321] Stamper-KurnDM, ChikkaturAP, GrlitzA, InouyeS,
Algebraic order and the BerezinskiiKosterlitzThouless GuptaS, PritchardDE and KetterleW 1999 Excitation of
transition in an exciton-polariton gas Phys. Rev. B phonons in a BoseEinstein condensate by light scattering
90205430 Phys. Rev. Lett. 8328769
[302] DagvadorjG, FellowsJ-M, MatyjakiewiczS, MarchettiF-M, [322] StferleT, MoritzH, SchoriC, KhlM and EsslingerT 2004
CarusottoI and SzymaskaM-H 2015 Nonequilibrium Transition from a strongly interacting 1d superfluid to a
phase transition in a two-dimensional driven open mott insulator Phys. Rev. Lett. 92130403
quantum system Phys. Rev. X 5041028 [323] BissbortU, GtzeS, LiY, HeinzeJ, KrauserJS,
[303] LagoudakisKG, WoutersM, RichardM, BaasA, WeinbergM, BeckerC, SengstockK and HofstetterW
CarusottoI, AndrR, DangLS and Deveaud-PldranB 2011 Detecting the amplitude mode of strongly interacting
2008 Quantized vortices in an exciton-polariton lattice Bosons by Bragg scattering Phys. Rev. Lett.
condensate Nat. Phys. 470610 106205303
[304] WolfD 1991 Kinetic roughening of vicinal surfaces Phys. [324] MeinertF, PanfilM, MarkMJ, LauberK, CauxJ-S and
Rev. Lett. 6717836 NgerlH-C 2015 Probing the excitations of a lieb-
[305] ShelykhIA, KavokinAV, RuboYG, LiewTCH and liniger gas from weak to strong coupling Phys. Rev. Lett.
MalpuechG 2010 Polariton polarization-sensitive 115085301
phenomena in planar semiconductor microcavities [325] SciollaB, PolettiD and KollathC 2015 Two-time
Semicond. Sci. Technol. 25013001 correlations probing the dynamics of dissipative many-
[306] StevensonRM, AstratovVN, SkolnickMS, WhittakerDM, body quantum systems: aging and fast relaxation Phys.
Emam-IsmailM, TartakovskiiAI, SavvidisPG, Rev. Lett. 114170401
BaumbergJJ and RobertsJS 2000 Continuous wave [326] CardyJ and TuberU 1996 Theory of branching and
observation of massive polariton redistribution by annihilating random walks Phys. Rev. Lett. 7747803
stimulated scattering in semiconductor microcavities Phys. [327] MarcuzziM, BuchholdM, DiehlS and LesanovskyI 2016
Rev. Lett. 8536803 Absorbing state phase transition with competing quantum
[307] BaumbergJJ, SavvidisPG, StevensonRM, and classical fluctuations arXiv:1601.07305
TartakovskiiAI, SkolnickMS, WhittakerDM and [328] MittalS, FanJ, FaezS, MigdallA, TaylorJM and HafeziM
RobertsJS 2000 Parametric oscillation in a vertical 2014 Topologically robust transport of photons in a
microcavity: a polariton condensate or micro-optical synthetic gauge field Phys. Rev. Lett. 113087403
parametric oscillation Phys. Rev. B 62R1624750 [329] MittalS, GaneshanS, FanJ, VaeziA and HafeziM 2016
[308] LoveAPD etal 2008 Intrinsic decoherence mechanisms Measurement of topological invariants in a 2d photonic
in the microcavity polariton condensate Phys. Rev. Lett. system Nat. Photon. 10180
101067404 [330] BardynC-E, BaranovMA, KrausCV, RicoE, ImamogluA,
[309] EdwardsSF and WilkinsonDR 1982 The surface statistics ZollerP and DiehlS 2013 Topology by dissipation
of a granular aggregate Proc. R. Soc. A 3811731 New J. Phys. 15085001
67
Rep. Prog. Phys. 79 (2016) 096001 Review
[331] LindnerNH, RefaelG and GalitskiV 2011 Nat. Phys. [336] MoeckelM and KehreinS 2008 Interaction quench in the
7490 Hubbard model Phys. Rev. Lett. 100175702
[332] KarzigT, BardynC-E, LindnerNH and RefaelG 2015 [337] BasovDN, FoglerMM, LanzaraA, WangF and ZhangY
Topological polaritons Phys. Rev. X 5031001 2014 Colloquium: Graphene spectroscopy Rev. Mod. Phys.
[333] ViyuelaO, RivasA and Martin-DelgadoMA 2014 Uhlmann 8695994
phase as a topological measure for one-dimensional [338] UlbrichtR, HendryE, ShanJ, HeinzTF and BonnM
fermion systems Phys. Rev. Lett. 112130401 2011 Carrier dynamics in semiconductors studied with
[334] BudichJC and DiehlS 2015 Topology of density matrices time-resolved terahertz spectroscopy Rev. Mod. Phys.
Phys. Rev. B 91165140 8354386
[335] RigolM, DunjkoV and OlshaniiM 2008 Thermalization [339] FollandGB 2008 Quantum Field Theory: a Tourist
and its mechanism for generic isolated quantum systems Guide for Mathematicians (Providence, RI: American
Nature 4528548 Mathematical Society)
68