Keldysh Field Theory for Driven Open Quantum Systems
L. M. Sieberer,1 M. Buchhold,2 and S. Diehl2, 3, 4 1Department of Condensed Matter Physics, Weizmann Institute of Science, Rehovot 7610001, Israel 2Institute of Theoretical Physics, TU Dresden, D-01062 Dresden, Germany 3Kavli Institute for Theoretical Physics, University of California, Santa Barbara, CA 93106, USA 4Institut f¨urTheoretische Physik, Universit¨atzu K¨oln,D-50937 Cologne, Germany Recent experimental developments in diverse areas — ranging from cold atomic gases to light-driven semi- conductors to microcavity arrays — move 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 driven-dissipative 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, as well as 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 sys- tematic 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.
CONTENTS 2 Applications 33
I. Introduction2 III. Non-equilibrium stationary states: spin models 33 A. New phenomena3 A. Ising spins in a single-mode cavity 33 B. Theoretical concepts and techniques4 1. Large-spin Holstein-Primakoff representation 34 C. Experimental platforms5 2. Effective Ising spin action for the single-mode cavity 36 1. Cold atoms in an optical cavity, and microcavity arrays: driven-dissipative spin-boson models5 B. Ising spins in a random multi-mode cavity 37 1. Keldysh action and saddle point equations 38 2. Exciton-polariton systems: driven open interacting bosons7 2. Non-equilibrium glass transition 39 3. Thermalization of photons and atoms in the glass 3. Cold atoms in optical lattices: heating phase 40 dynamics9 4. Quenched and Markovian bath coupling in the D. Outline and scope of this review 10 Keldysh formalism 41
IV. Non-equilibrium stationary states: bosonic models 42 1 Theoretical background 11 A. Density-phase representation of the Keldysh action 44 II. Keldysh functional integral for driven open systems 11 B. Critical dynamics in 3D 46 A. From the quantum master equation to the Keldysh 1. Effective action for driven-dissipative functional integral 12 condensation 47 B. Examples 16 2. Non-equilibrium FRG flow equations 48 1. Single-mode cavity, and some exact properties 16 3. Scaling solutions and critical behavior 49 2. Driven-dissipative condensate 20 C. Absence of algebraic order in 2D 51 C. Semiclassical limit of the Keldysh action 22 D. KPZ scaling in 1D 53 D. Symmetries of the Keldysh action 24 1. Thermodynamic equilibrium as a symmetry of V. Universal heating dynamics in 1D 54 arXiv:1512.00637v2 [cond-mat.quant-gas] 4 Aug 2016 the Keldysh action 24 A. Heating an interacting Luttinger liquid 54 2. Semiclassical limit of the thermal symmetry 27 B. Kinetic equation 56 3. The Noether theorem in the Keldysh formalism 28 C. Self-consistent Born approximation 58 4. Closed systems: energy, momentum and particle 1. Kinetics for small momenta 59 number conservation 29 2. Scaling of the self-energy 60 5. Extended continuity equation in open systems 30 D. Heating and universality 60 6. Spontaneous symmetry breaking and the 1. Phonon densities 60 Goldstone theorem 31 2. Non-equilibrium scaling 61 E. Open system functional renormalization group 32 3. Observability 61 2
VI. Conclusions and outlook 61 Systems located at this new interface are not guaranteed to ther- malize, due to the absence of energy conservation and the re- VII. Acknowledgements 62 sulting breaking of detailed balance by the external drive at the microscale. They rather converge to non-equilibrium stationary A. Functional differentiation 63 states of matter, creating scenarios without counterpart in con- densed, equilibrium matter. This rules out conventional theoreti- B. Gaussian functional integration 63 cal equilibrium concepts and techniques to be used, and calls for the development of new theoretical tools. The physical frame- References 64 work sparks broader theoretical questions on the existence of new phases of bosonic [18, 19] and fermionic [20–22] matter, the nature of phase transitions in such driven systems [10, 23– I. INTRODUCTION 26], and the observable consequences of quantum mechanics at the largest scales [27, 28]. Beyond stationary states [29], a fun- damental challenge is set by the time evolution of interacting Understanding the quantum many-particle problem is one of quantum systems, which is currently explored theoretically [30– the grand challenges of modern physics. In thermodynamic 36] and experimentally in cold atomic [37–41] and photonic sys- equilibrium, the combined effort of experimental and theoretical tems [42]. A key goal is to identify universal dynamical regimes research has made tremendous progress over the last decades, that hold beyond specific realizations or precise initial condi- revealing the key concepts of emergent phenomena and uni- tions. Combining idealized closed system evolution with the in- versality. This refers to the observation, that the relevant de- trinsic open system character of any real world experiment takes grees of freedom governing the macrophysics may be vastly dif- this setting to the next stage, and exhibits emergent dynamics ferent from those of the microscopic physics, but on the other markedly different from closed systems both for short [43, 44] hand are constrained by basic symmetries on the short distance and long evolution times [45–51]. scale, restoring predictive power. Regarding the role and power of these concepts in out-of-equilibrium situations, there is a The interplay of coherent and driven-dissipative dynamics can large body of work in the context of classical near equilibrium be a natural consequence of the driving necessary to maintain a and non-equilibrium many-body physics and statistical mechan- certain many-body state. Going one step further, it is possible ics [1–5]. However, analogous scenarios and theoretical tools to exploit and further develop the toolbox of quantum optics for for non-equilibrium quantum systems are much less developed. the driven-dissipative manipulation of many-body systems. Re- This review addresses recent theoretical progress in an im- cently, it has been recognized that the concept of dissipative state portant and uprising class of dynamical non-equilibrium phases preparation in quantum optics [52, 53] can be developed into a of quantum matter, which emerge in driven open quantum sys- many-body context, both theoretically [54–57] and experimen- tems, where a Hamiltonian is not the only resource of dynamics. tally [58–60]. Suitably tailored dissipation then does not neces- This concerns both non-equilibrium stationary states, but also sarily act as an adversary to subtle quantum mechanical correla- the dynamics of such ensembles. Strong motivation for the ex- tions such as phase coherence or entanglement [61–66]. In con- ploration of non-thermal stationary states comes from a recent trast, it can even create these correlations, and dissipation then surge of experiments in diverse areas: in cold atomic gases [6– represents the dominant resource of many-body dynamics. In 8], hybrid light-matter systems of Bose-Einstein condensates particular, even topologically ordered states in spin systems [67] placed in optical cavities are created [9, 10], or driven Ryd- or of fermionic matter [68? ] can be induced dissipatively ([56] berg ensembles are prepared [11]; light-driven semiconductor and [70–72], respectively). These developments open up a new heterostructures realize Bose-Einstein condensation of exciton- arena for many-body physics, where the quantum mechanical polaritons [12, 13]; coupled microcavity arrays are pushed to microscopic origin is of key importance despite a dominantly a regime demonstrating both strong coupling of light and mat- dissipative dynamics. ter [14] and scalability [15]; large ensembles of trapped ions implement varieties of driven spin models [16, 17]. All those Summarizing, this is a fledging topical area, where first results systems are genuinely made of quantum ingredients, and share underpin the promise of these systems to exhibit genuinely new in common that coherent and driven-dissipative dynamics oc- physics. In the remainder of the introduction, we will discuss cur on equal footing. This creates close ties to typical setups in in some more detail the three major challenges which emerge in quantum optics. But on the other hand, they exhibit a continuum these systems. The first one concerns the identification of novel of degrees of freedom, characteristic for many-body physics 1. macroscopic many-body phenomena, which witness the micro- scopic driven open nature of such quantum systems. Second, we anticipate the theoretical machinery, which allows us to perform the transition from micro- to macrophysics in a non-equilibrium 1 This includes the cases of extended spatial continuum and lattice systems. In context in practice. Third, we describe some representative ex- both cases, a continuum of momentum modes obtains, underlying the charac- perimental platforms, which motivate the theoretical efforts, and teristics of many-body problems at long wavelength. in which the predictions can be further explored. 3
A. New phenomena breaks detailed balance [26]. In particular, in three dimen- sions and close to the critical point of driven-dissipative Bose- As pointed out above, one of the key goals of the research re- condensation, a degeneracy of critical exponents indicates a uni- viewed here is the identification of new macroscopic many-body versal asymptotic thermalization, in the sense of an emergent phenomena, which can be uniquely traced back to the micro- thermal FDR. A similar phenomenology is observed in a disor- scopic driven open nature of such systems, and do not have an dered multimode extension of the Dicke model, see Sec. III B. immediate counterpart in equilibrium systems. This underlines the strongly attractive nature of the thermal equi- The driven nature common to the systems considered here can librium fixed point at low frequencies. Still in these systems, always be associated to the fact that the underlying microscopic non-equilibrium conditions leave their traces in the dynamical Hamiltonian is time dependent, with a time dependence relating response of the system, in terms of information that does not to external driving fields such as lasers. When such an ensemble enter the FDR at leading order. For example, the critical behav- (i) in addition has a natural partition into a “system” and a “bath” ior is characterized by a fine structure in a new and indepen- — a continuum of modes well approximated by harmonic oscil- dent critical exponent, which measures decoherence, and whose lators with short memory —, (ii) the system-bath coupling is value distinguishes equilibrium and non-equilibrium dynamics, weak compared to a typical energy scale of the “system” Hamil- see Sec.IVB. tonian and (iii) linear in the bath creation and annihilation oper- ators (so that they can be integrated out straightforwardly), then Instead of emergent thermal behavior indicating the fadeout of an effective (still microscopic) description in terms of combined non-equilibrium conditions upon coarse graining, also the oppo- Hamiltonian and driven-dissipative Markovian quantum dynam- site behavior is possible. For example, low dimensional (d ≤ 2) ics of the “system” ensues. The “system” dynamics obtains by bosonic systems at low noise level, such as exciton-polaritons tracing out the bath variables. well above threshold, are not attracted to the equilibrium fixed The resulting effective microdynamics is “non-equilibrium,” point as their three dimensional counterparts, but rather flow to in a sense sharpened in Sec. II D 1. This not only concerns the the non-equilibrium fixed point of the Kardar-Parisi-Zhang [73] time evolution, but also holds for the non-equilibrium stationary universality class, see Sec.IVC. This can be interpreted as a states. More precisely, the above situation implies an explicit universal and indefinite increase of the non-equilibrium strength, breaking of detailed balance, since the “system” energy is not which is triggered even if the violation of detailed balance at the conserved due to the explicitly time-dependent drive. microscopic level is very small. What do we actually mean by “detailed balance” and “ther- mal equilibrium?” In an operational sense, the principle of de- Universal non-equilibrium phenomena can also occur in the tailed balance states that there is a partition invariance for the time evolution of driven open systems. For example, intriguing temperature (or, more generally, the noise level) present in the scaling laws describing algebraic decoherence [46], anomalous system: an arbitrary bipartition of the system can be chosen, one diffusion [74], or glass-like behavior [45, 75–77] have been iden- part can be traced out, and the resulting subsystem will be at the tified in the long time asymptotics of driven spin systems close same temperature (noise level) as the total system. This parti- to the stationary states. Conversely, the short time behavior of tion invariance is the condition for a globally well-defined tem- driven open lattice bosons shows universal scaling laws directly perature characteristic for systems in thermal equilibrium. More witnessing the non-equilibrium drive, see Sec.V. This scaling formally, thermal equilibrium can be detected by means of so- can be related to a strongly pronounced non-equilibrium shape called fluctuation-dissipation relations (FDRs). These connect of the time-dependent distribution function in the early stages of the two fundamental observables in physical systems — correla- evolution, and be traced back to conservation laws of the driven- tion and response functions (see Sec.IIB). In the case of thermal dissipative generator of dynamics. equilibrium, the connection is dictated by the particle statistics alone. It is then given by the Bose- and Fermi-distributions, re- The above discussion mainly focuses on the difference be- spectively. Deviations from this universal form, which has only tween equilibrium and non-equilibrium systems on the macro- two free parameters (temperature and chemical potential, relat- scopic level of observation. Another direction, still much less ing to the typical conserved quantities energy and particle num- developed, concerns the distinction between classical and quan- ber), provide a necessary requirement for non-equilibrium con- tum effects. Again, although the quantum mechanical descrip- ditions. tion is necessary at a microscopic level, the persistence of quan- In non-equilibrium stationary states, no such general form ex- tum effects at the macroscale is not guaranteed. This is mainly ists. We will encounter a concrete and simple example in the due to the Markovian noise level inherent to such quantum sys- context of the driven Dicke model (a cavity photon coupled to tems. Nevertheless, systems with suitably engineered driven- a collective spin) in Sec. III A, where the form of the FDR de- dissipative dynamics show typical quantum mechanical phe- pends on the observable we are choosing (e.g., the position or nomena such as phase coherence [54, 55, 61, 63, 64, 78], en- momentum correlations and responses). tanglement [58–60, 62], or topological order [56, 66, 70–72]. On the other hand, thermal FDRs can emerge at long Especially fermionic systems, which do not possess a classical wavelength, even though the microscopic dynamics manifestly limit, are promising in this direction. 4
B. Theoretical concepts and techniques techniques, such as diagrammatic perturbation theory including sophisticated resummation schemes. But it also encompasses The development of theoretical tools needed to perform the non-perturbative approaches, which often capitalize on the flex- transition from micro- to macro-physics in driven open quantum ibility of the functional integral when it comes to picking the systems is still in progress, as a topic of current research. A rea- relevant degrees of freedom for a given problem. This is the son for the preliminary status of the theory lies in the fact that challenge of emergent phenomena, whose solution typically is two previously rather independent disciplines — quantum op- strongly scale dependent [86]. Familiar examples include an tics and many-body physics — need to be unified on a technical efficient description of emergent Cooper pair or molecular de- level. grees of freedom in interacting fermion systems, or vortices which conveniently parameterize the long-wavelength physics Quantum optical systems are well described microscopically of interacting bosons in two dimensions. The description of the in terms of Markovian quantum master equations, which treat change of physics with scale was given its mathematical foun- coherent Hamiltonian and driven-dissipative dynamics on equal dation in terms of the renormalization group already mentioned footing. To solve such equations both for their dynamics above, yet another tool developed and most clearly formulated in and their stationary states, powerful techniques have been de- a functional language. Finally, the functional integral based on vised. This comprises efficient exact numerical techniques for a single scalar quantity — the system’s action, which encodes small enough systems, such as the quantum trajectories ap- all the dynamics on the microscopic scale — is a convenient proach [48, 79, 80]. But it also includes analytical approaches framework when it comes to the classification of symmetries and such as perturbation theory for quantum master equations, e.g., associated conservation laws, and their use in devising approxi- in the frame of the Nakajima-Zwanzig projection operator tech- mation schemes respecting them. nique [81, 82], or mappings to P, W, or Q representations [83], casting the problem from a second quantized formulation into To put it short, while the driven open many-body systems partial differential equations. are well described by microscopic master equations, the tradi- A characteristic feature of traditional quantum optical sys- tional techniques of quantum optics cannot be used efficiently tems is the finite spacing of the few energy levels which play – at least not in the case where the generic complications of a role. When considering systems with a spatial continuum of many-body systems start to play a role. Conversely, their driven degrees of freedom instead, the energy levels become continu- open character makes it impossible to approach these problems ous. This does not mean that the microscopic modelling in terms in the framework of equilibrium many-body physics. This sit- of a quantum master equation is inappropriate: for the driven- uation calls for a merger of the disciplines of quantum optics dissipative terms in such an equation, the assumption of spa- and many-body physics on a technical level. On the numerical tially independent dissipative processes (such as atomic loss or side, progress has been made in one spatial dimension recently spontaneous emission) is still valid as long as the emitted wave- by combining the method of quantum trajectories with power- length of radiation is well below the spatial resolution at the ful density matrix renormalization group algorithms [87–90], scale where the microscopic model is defined. Indeed, in this see [48] for an excellent review on the topic. For more analyti- situation, destructive interference of radiation justifies the de- cal approaches, the Keldysh functional or path integral [91–96] scription of driven dissipation in terms of incoherent processes. is ideally suited (but see [97] for a recent systematic perturbative However, under these circumstances the smallness of a micro- approach to lattice Lindblad equations with extensions to sophis- scopic expansion parameter no longer guarantees the smallness ticated resummation schemes [98], and [99–101] for advanced of the associated perturbative correction. Here the reason is that variational techniques). Conceptually, the latter captures the in perturbation theory, one is summing over intermediate states most general situation in many-body physics — the dynamics with propagation amplitudes down to the longest wavelengths. of a density matrix under an arbitrary temporally local generator This can lead to infrared divergences in naive perturbation the- of dynamics. We refer to [102, 103] for an introduction to the ory — a circumstance that found its physical interpretation and Keldysh functional integral. In our context, it can be derived by technical remedy in equilibrium in terms of the renormalization a direct functional integral quantization of the Markovian quan- group [84, 85]. We emphasize that it is precisely this situation of tum master equation. This procedure results in a simple transla- long wavelength dominance which underlies much of the univer- tion table from the master equation to the key player in the as- sality, i.e., insensitivity to microscopic details, which is encoun- sociated Keldysh functional integral, the Markovian action. At tered when moving from the microscale to macroscopic observ- first sight, the complexity of the non-equilibrium Keldysh func- ables in many-particle systems. tional integral is increased by the characteristic “doubling” of The modern framework to understand many-particle prob- degrees of freedom compared to thermal equilibrium. However, lems in thermodynamic equilibrium is in terms of the functional it should be noted that it is precisely this feature which relates integral formulation of quantum field theory. The spectrum of the Keldysh functional integral much closer to real time (von its application covers a remarkable range of energy scales, from Neumann) evolution familiar from quantum mechanics. We will ultracold atomic gases to condensed matter systems with strong demonstrate this in Sec.IIA, making the Markovian action a correlations to quantum chromodynamics and the quantum the- rather intuitive object to work with. Furthermore, when prop- ory of gravity. It provides us with a well-developed toolbox of erly harnessing symmetries, the complexity of calculations can 5 often be made comparable to thermodynamic equilibrium. Most ory. The reason is the finite Markovian noise level that such importantly however, the powerful toolbox of quantum field the- systems exhibit generically, as explained in Sec.IIC. The prefix ory is opened in this way. It is thus possible to leverage the full “semi” refers to the fact that phase coherence may still persist power of sophisticated techniques from equilibrium field theory in such circumstances — the situation is comparable to a Bose- to driven open many-body systems. Einstein condensate at finite temperature. Recently however, sit- Relation to classical dynamical field theories — The quan- uations have been identified where the drastic simplifications of tum mechanical Keldysh formulation reduces to the so-called the semi-classical limit do not apply. In particular, this occurs Martin-Siggia-Rose (MSR) functional integral in the (semi-) in systems with dark states — pure quantum states which are classical limit, in turn equivalent to a stochastic Langevin equa- dynamical fixed points of driven-dissipative evolution [54, 110]. tion formulation [102]. This statement will be made precise and In these cases, classical dynamical field theories are inappropri- discussed in Sec.IIC. A large amount of work has been dedi- ate, which calls for the development of quantum dynamical field cated to this limit in the past. On the one hand, this concerns theories [28]. These developments are just in their beginnings. dynamical aspects of equilibrium statistical mechanics, and we In this review, we concentrate on systems composed of refer to the classic work by Hohenberg and Halperin [1] for an bosonic degrees of freedom. However, it is also possible to ad- overview. This work shows that, while the static universal crit- dress spin systems in terms of functional integrals [111], and ical behavior is determined by the symmetries and the dimen- simple models systems have been analyzed in this way, see [112] sionality of the problem, the dynamical critical behavior is sen- and Sec. III A. More sophisticated approaches to spin systems sitive to additional dynamical conservation laws. This leads to were elaborated in the context of multimode optical cavities a fine structure, defining dynamical universality classes which in [113], and systematically for various symmetries for lattice are denoted by models A–J [1]. These models also provide a systems in [114]. Fermi statistics is also conveniently imple- convenient framework to describe the statistics of work, sum- mented in the functional integral formulation. This is relevant, marized in Jarzynski’s work theorem [104, 105] and Crooks’ re- e.g., for driven open Fermi gases in optical cavities [21, 22, 115] lation [106]. On the other hand, non-equilibrium situations are or lattices [116], or dissipatively stabilized topological fermion captured as well. Here, we highlight in particular genuine non- matter [70–72]. equilibrium universality classes, which are not smoothly con- nected to the equilibrium models. Among them is the problem of reaction-diffusion models [107] including directed percola- C. Experimental platforms tion [4], which is relevant to certain chemical processes (for an implementation of this universality class with driven Rydberg The progress in controlling, manipulating, detecting and scal- gases, see [50]). Another key example is surface growth, de- ing up driven open quantum systems to many-body scenarios has scribed by the Kardar-Parisi-Zhang equation [73], giving rise been impressive over the last decade. Here we sketch the basic to a non-equilibrium universality class which is at the heart of physics of three representative platforms, and indicate the rele- driven phenomena such as the growth of bacterial colonies or vant microscopic theoretical models in the frame of the Marko- the spreading of fire. vian quantum master equation. In later sections, we will trans- In the same class of approaches ranges the so-called Doi-Peliti late this physics into the language of the Keldysh functional inte- functional integral [108, 109], which is a functional representa- gral. For each of these platforms, excellent reviews exist, which tion of classical master equations, and may be viewed as an MSR we refer to at the end of Sec.ID together with further litera- theory with a specific, highly non-linear appearance of the field ture on open systems. The purpose of this section is to give variables. It reduces to the conventional MSR form in a leading an overview only, and to put the respective platforms into their order Taylor expansion of the field non-linearities. A compre- overarching context as driven open quantum systems with many hensive overview of models, methods, and physical phenomena degrees of freedom. in the (semi-)classical limit is provided in [5]. We also note that the usual mean field theory, where corre- lation functions are factorized into products of field amplitudes 1. Cold atoms in an optical cavity, and microcavity arrays: and which is often used as an approximation to the quantum driven-dissipative spin-boson models master equation in the literature, corresponds to a further for- mal simplification of the semi-classical limit. Here the effects of Cavity quantum electrodynamics (cavity QED), with its fo- noise are neglected completely. Conversely, the semi-classical cus on strong light-matter interactions, is a growing field of re- limit represents a systematic extension of mean field theory, search, which has experienced several groundbreaking advances which includes the Markovian noise fluctuations. This level of in the past few years. Historically, these systems were devel- approximation is referred to as optimal path approximation in oped as few or single atom experiments, detecting the radiation the literature on MSR functional integrals [5, 102]. properties of atoms, which are strongly coupled to a quantized In many cases, even though the microscopic description is light field. The focus has recently been shifted towards loading in terms of a quantum master equation, at long distances the more and more atoms inside a cavity. Thereby, not only single Keldysh field theory reduces to a semi-classical MSR field the- particle dynamics in strong radiation fields can be probed, but 6 also collective, macroscopic phenomena, which are driven by for N → ∞, it features a phase transition, which spontaneously light-matter interactions. For an excellent review on this topic, breaks the Ising symmetry. Crossing the transition, the system covering important experimental and theoretical developments, enters a superradiant phase, characterized by finite expectation x see [10]. In the experiments, cold atoms are loaded inside an values hai , 0, hσi i , 0. This describes condensation of the optical or microwave cavity, for which the coherent interaction cavity photons, i.e., the formation of a macroscopically occu- between the atomic internal states and a single cavity mode dom- pied, coherent intra-cavity field, and a “ferromagnetic” ordering inates over dissipative processes [117]. The atoms absorb and of the atoms in the x-direction. emit cavity photons, thereby changing their internal states. Due Although the Dicke model is a standard model√ for cavity QED to this process, the spatial modulation of the intra-cavity light experiments in the ultra-strong coupling limit Ng > ωzω0, it field induces a coupling of the cavity photons to the atomic in- has been realized only very recently in cold atom experiments, ternal state as well as their motional degree of freedom [118]. where an entire BEC was placed inside an optical cavity [9]. It In this way, cavity photons mediate an effective atom-atom in- has been shown that this setup maps to a Dicke model, with a teraction, which leads to a back-action of a single atom on the “collective” spin degree of freedom [120]. Here, the detuning motion of other atoms inside the cavity. This cavity mediated of the pump laser was chosen such that the atoms effectively re- interaction is long-ranged in space and represents the source of main in the internal ground state, but acquire a characteristic re- collective effects in cold atomic clouds within cavity QED ex- coil momentum when scattering with a cavity photon. This scat- periments. One hallmark of collective dynamics in cavity QED tering creates a collective, motionally excited state, which re- has been the observation of self-organization of a Bose-Einstein places the role of an individual, internally excited atom. The ex- condensate in an optical cavity. This is accompanied by a Dicke perimental realization of a superradiance transition in the Dicke phase transition via breaking of a discrete Z2-symmetry of the model is usually inhibited, since the required coupling strength underlying model [9, 119–121]. by far exceeds the available value of the atomic dipole coupling. Although the dominant dynamics in these systems is coherent, However, for the BEC in the cavity, the energy scales of the ex- there are dissipative effects which cannot be discarded. These cited modes are much lower than the optical scale of the atomic are the loss of cavity photons due to imperfections in the cav- modes. In this way, the superradiance transition indeed became ity mirrors, or spontaneous emission processes of atoms, which experimentally accessible. This was inspired by a theoretical emit photons into transverse modes. These effects modify the proposal using two balanced Raman channels between different dynamics of the system on the longest time scales. Therefore, internal atomic states inside an optical cavity, which reduced the they become relevant for macroscopic phenomena, such as phase effective level splitting of the internal states to much lower en- transitions and collective dynamics. For instance, dissipative ef- ergy scales [129]. fects have been shown theoretically [122, 123] and experimen- In addition to the unitary dynamics represented by the Dicke tally [124] to modify the critical exponent of the Dicke transi- model, the cavity is subject to permanent photon loss due to im- tion compared to its zero temperature value. This illustrates that perfections in the cavity mirrors. For high finesse cavities, the for the analysis of collective phenomena in cavity QED experi- coupling of the intra cavity photons to the surrounding vacuum ments, the dissipative nature of the system has to be taken into radiation field is very weak, and the latter can be eliminated in a account properly [112, 125]. Born-Markov approximation [81]. This results in a Markovian Typically, for a cavity field which has a very narrow spectrum, quantum master equation for the system’s density matrix the atomic internal degrees of freedom can be reduced to two internal states, whose transitions are nearly resonant to the pho- ∂tρ = −i[HD, ρ] + Ldρ, (2) ton frequency. The operators acting on these two internal states can be represented by Pauli matrices, making them equivalent where ρ is the density matrix for the intra cavity system and Ld to a spin-1/2 degree of freedom. A very important model in the adds dissipative dynamics to the coherent evolution of the Dicke framework of cavity QED is the Dicke model [126–128], which model. For a vacuum radiation field, it is given by describes N atoms (i.e., two-level systems) coupled to a single L † − 1 { † } quantized photon mode. This is expressed by the Hamiltonian dρ = κ LρL 2 L L, ρ . (3) (here and in the following we set ~ = 1) The Lindblad operator L acting on the density matrix describes N N pure photon loss (L = a) with an effective loss rate κ; the lat- † ωz X z g X x † HD = ω0a a + σi + √ σi a + a . (1) ter depends on system specific parameters, but is typically the 2 N i=1 i=1 smallest scale in the master equation (2)[10]. z Here, ω0 is the photon frequency, σ = |1ih1| − |0ih0| repre- In generic cavity experiments, there are also atomic sponta- sents the splitting of the two atomic levels with energy differ- neous emission processes. The atoms scatter a laser or cavity x ence ωz. σ = |0ih1| + |1ih0| describes the coherent excitation photon out of the cavity, and this represents a source of deco- and de-excitation of the atomic state proportional to the atom- herence. This process has been considered in Ref. [112], and photon coupling strength g. The Dicke model features a dis- leads to an effective decay rate of the atomic excited state and crete Z2 Ising symmetry: it is invariant under the transforma- therefore to a dephasing of the atoms, described by additional † x † x z tion (a , a, σi ) 7→ (−a , −a, −σi ). In the thermodynamic limit, Lindblad operators Li = σi [130]. However, these losses are 7 at least three orders of magnitude smaller than the cavity de- using multiple qubits [140]. An advantage of such schemes is cay rate and typically not considered [9, 125]. The basic model that the symmetries of the underlying dynamics are preserved for cavity QED with cold atoms is therefore represented by the or less severely corrupted in this way (cf. the discussion in Dicke model with dissipation, formulated in terms of the master Sec.IID). In other platforms, such as exciton-polariton systems equation (2). Its dynamics is discussed in Sec. III, including the (cf. the subsequent section), incoherent pumping is more natural Dicke superradiance transition. from the outset. The Dicke model Hamiltonian takes the form HD = Hc + Hs + All the systems discussed here represent genuine instances of Hcs, where Hc,s,cs represent the bosonic cavity, spin, and spin- driven open many-body systems. Besides the coherent drive, boson sectors, respectively. There are many directions to go be- they undergo dissipative processes, which have to be taken into yond the Dicke model with single collective spin, still keeping account for a proper understanding of their time evolution and the basic feature of coupling spin to boson (cavity photon) de- stationary states. A generic feature of these processes is their grees of freedom. One direction – relevant to future cold atom locality. For the effective spin degrees of freedom, the typical − experiments – is to consider multimode cavities instead of a sin- processes are qubit decay (local Lindblad operators Li = σi ) and z gle one. In particular, in conjunction with quenched disorder, dephasing (Li = σi ). For the bosonic component, local single- intriguing analogies to the physics of quantum glasses can be photon loss is dominant, Li = ai. established in this way [113, 131]. Here, the global coupling of Under various circumstances, such as a low population of all spins to a single mode gi ≡ g is replaced by random couplings the excited spin states, the latter degrees of freedom can be gi,`, where the index ` now refers to a collection of cavity modes. integrated out. In this limit, their physical effect is to gen- The many-body physics of such an open system is discussed in erate Kerr-type bosonic non-linearities, giving rise to driven Sec. III. open variants of the celebrated Bose-Hubbard model. These The basic building block of the Dicke model is the spin-boson models can even be brought into the correlation dominated † x term of the form Hsc ∼ (a + a )σ , i.e. a Rabi type non- regime [24, 132, 133, 141]. Oftentimes, these approximations on linearity that preserves the Z2 symmetry of Hc,s. In the con- the spin sector actually apply, and it is both useful and interesting text of circuit quantum electrodynamics, a natural many-body to study these effective low frequency bosonic theories instead generalization of Hamiltonians with a spin-boson interaction is of the full many-body spin-boson problems, see Refs. [142–146] to consider entire arrays of microcavities (instead of consider- for recent work in this direction. ing many modes within a single cavity). These cavities can be coupled to each other by single photon tunnelling processes between adjacent cavities, giving rise to Hubbard-type hopping † 2. Exciton-polariton systems: driven open interacting bosons terms ∼ Jai a j + h. c., where i, j now label the spatial index of the cavities. In cirquit QED, strong non-linearities can be gen- erated, e.g., by coupling to adjacent qubits made of Cooper pair Exciton-polaritons are an extremely versatile experimental boxes [132, 133]. This gives rise to many-body variants of the platform, which is documented by the richness of physical phe- Rabi model [134], whose phase diagrams have been studied re- nomena that have been studied in these systems both in the- cently [135–137]. Furthermore, for the implementation of lat- ory and experiment. For a comprehensive account of the sub- tice Dicke models with large collective spins, the use of hybrid ject, we refer to a number of excellent review articles [13, 147, quantum systems consisting of superconducting cavity arrays 148]. A Keldysh functional integral approach is discussed in coupled to solid-state spin ensembles have been proposed [138]. Refs. [149, 150], which provides both a microscopic deriva- Spontaneous collective coherence in driven-dissipative cavity ar- tion of an exciton-polariton model and a mean field analysis rays has been studied in [139]. including Gaussian fluctuations. At this point, we content our- In many physical situations (away from the ultra-strong cou- selves with a short introduction, with the aim of showing that pling limit), a form of the spin-cavity interaction alternative in a suitable parameter regime, exciton-polaritons very natu- to the Rabi term is more appropriate, with non-linear building rally provide a test-bed to study Bose condensation phenom- + † − block Hcs ∼ aσ + a σ . In fact, this form results naturally ena out of thermal equilibrium. Similar physics can also arise from the weak coupling rotating wave approximation of a driven in a variety of other systems, including condensates of pho- spin-cavity problem. For a single cavity mode and spin, the re- tons [151], magnons [152], and potentially excitons [153]. Re- sulting model is the Jaynes-Cummings model, which in contrast markably, even cold atoms could be brought to condense in a to the Dicke model possesses a continuous U(1) phase rotation non-equilibrium regime, where continuous loading of atoms bal- symmetry under a → eiθa, σ− → eiθσ−. ances three-body losses [154], or in atom laser setups [155–157]. Clearly, when such systems are driven coherently via a Hamil- A basic experimental setup for exciton-polaritons consists of † tonian Hd = Ω(a + a ) or suitable multimode generalizations a planar semiconductor microcavity embedding a quantum well thereof, both the U(1) and even the Z2 symmetries of the above (see Fig.1 (a)). This setting allows for a strong coupling of cav- models are broken explicitly. Coherent drive is usually the sim- ity light and matter in the quantum well, as originally proposed plest way to compensate for unavoidable losses due to cavity in [158]. The free dynamics of the elementary excitations of this leakage, although incoherent pumping schemes are conceivable system — i.e., of cavity photons and Wannier-Mott excitons — 8 is described by the quadratic Hamiltonian [13] (a) photon H0 = HC + HX + HC−X, (4) exciton e where the parts of the Hamiltonian involving only photons and excitons, respectively, take the same form, which is given by h (here the index α labels cavity photons, α = C, and excitons, α = X, respectively)2 Z dq X † Braggmirror QW Braggmirror Hα = ωα(q)a (q)aα,σ(q). (5) (2π)2 α,σ σ (b) upper ! † polariton Field operators aα,σ(q) and aα,σ(q) create or destroy a photon or exciton (note that both are bosonic excitations) with in-plane photon momentum q and polarization σ (there are two polarization excitation states of the exciton which are coupled to the cavity mode [13]). relaxation For simplicity, we neglect polarization effects leading to an ef- exciton fective spin-orbit coupling [13]. Due to the confinement in the transverse (z) direction, i.e., along the cavity axis, the motion of lower photons in this direction is quantized as qz,n = πn/lz, where n is polariton a positive integer, and lz is the length of the cavity. In writing emission the Hamiltonian (5), we are assuming that only the lowest trans- q verse mode is populated, which leads to a quadratic dispersion q 2 2 as a function of the in-plane momentum q = |q| = qx + qy: Figure 1. (a) Schematic of two Bragg mirrors forming a microcavity, in which a quantum well (QW) is embedded. In the regime of strong q q2 light-matter interaction, the cavity photon and the exciton hybridize and ω (q) = c q2 + q2 = ω0 + + O(q4). (6) C z,1 C 2m form new eigenmodes, which are called exciton-polaritons. (b) Energy C dispersion of the upper and lower polariton branches as a function of in- 0 plane momentum q. In the experimental scheme illustrated in this figure Here, c is the speed of sound, ωC = cqz,1, and the effective mass of the photon is given by m = q /c. Typically, the value of the (cf. Ref. [12]), the incident laser is tuned to highly excited states of the C z,1 quantum well. These undergo relaxation via emission of phonons and photon mass is orders of magnitude smaller than the mass of the scattering from polaritons, and accumulate at the bottom of the lower exciton, so that the dispersion of the latter appears to be flat on polariton branch. In the course of the relaxation process, coherence is the scale of Fig.1 (b). quickly lost, and the effective pumping of lower polaritons is incoher- Upon absorption of a photon by the semiconductor, an exciton ent. is generated. This process (and the reverse process of the emis- sion of a photon upon radiative decay of an exciton) is described by non-radiative decay rate of excitons, it is appropriate to think of exciton-polaritons as the elementary excitations of the system. Z dq X H = Ω a† (q)a (q) + H.c. , (7) In experiments, it is often sufficient to consider only lower C−X R 2 X,σ C,σ polaritons in a specific spin state, and to approximate the disper- (2π) σ sion as parabolic [13]. Interactions between exciton-polaritons where ΩR is the rate of the coherent interconversion of photons originate from various physical mechanisms, with a dominant into excitons and vice versa. The quadratic Hamiltonian (4) can contribution stemming from the screened Coulomb interactions be diagonalized by introducing new modes — the lower and between electrons and holes forming the excitons. Again, in the upper exciton-polaritons, ψLP,σ(q) and ψUP,σ(q) respectively, low-energy scattering regime, this leads to an effective contact which are linear combinations of photon and exciton modes. The interaction between lower polaritons. As a result, the low-energy dispersion of lower and upper polaritons is depicted in Fig.1 description of lower polaritons takes the form (in the following (b). In the regime of strong light-matter coupling, which is we drop the subscript indices in ψLP,σ)[13] reached when ΩR is larger than both the rate at which pho- Z " 2 ! # tons are lost from the cavity due to mirror imperfections and the † 0 ∇ † 2 2 HLP = dx ψ (x) ωLP − ψ(x) + ucψ (x) ψ(x) . (8) 2mLP While this Hamiltonian is quite generic and arises also, e.g., in 2 In Ref. [149, 150], a different model for excitons is used: they are assumed cold bosonic atoms in the absence of an external potential, the to be localized by disorder, and interactions are included by imposing a hard- peculiarity of exciton-polaritons is that they are excitations with core constraint. relatively short lifetime. In turn, this necessitates continuous re- 9 plenishment of energy in the form of laser driving in order to 3. Cold atoms in optical lattices: heating dynamics maintain a steady population. In Fig.1 (b), we consider the case in which the excitation laser is tuned to energies well above In recent years, experiments with cold atoms in optical lat- the lower polariton band. The thus created high-energy excita- tices have shown remarkable progress in the simulation of many- tions are deprived of their excess energy via phonon-polariton body model systems both in and out of equilibrium. A par- and stimulated polariton-polariton scattering. Eventually, they ticular strength of cold atom experiments is the unprecedented accumulate at the bottom of the lower polariton band. As a con- tuneability of model parameters, such as the local interaction sequence of multiple scattering processes, the coherence of the strength and the lattice hopping amplitude. This becomes possi- incident laser field is quickly lost, and the effective pumping of ble by, e.g., manipulation of the lattice laser and external mag- lower polaritons is incoherent. netic fields. It comes along with a very weak coupling of the A phenomenological model for the dynamics of the lower po- system to the environment, such that the dynamics can often be lariton field, which accounts for both the coherent dynamics gen- seen as isolated on relevant time scales for typical measurements erated by the Hamiltonian (8) and the driven-dissipative one de- of static, equilibrium correlations. However, more and more ex- scribed above, was introduced in Ref. [159]. It involves a dissi- periments start to investigate the realm of non-equilibrium phe- pative Gross-Pitaevskii equation for the lower polariton field that nomena with cold atoms, e.g., by letting systems prepared in a is coupled to a rate equation for the reservoir of high-energy ex- non-equilibrium initial condition relax in time towards a steady citations. However, for the study of universal long-wavelength state [37, 39, 163–165]. With these experiments, time scales are behavior in Sec.IV, this degree of microscopic modeling is ac- reached, for which the dissipative coupling to the environment tually not required: indeed, any (possibly simpler) model that becomes visible in experimental observables. Such dissipation possesses the relevant symmetries (see the discussion at the be- may even hinder the system from relaxation towards a well de- ginning of Sec.IV) will yield the same universal physics. Such fined steady state. a model can be obtained by describing incoherent pumping and A relevant example of a dissipative coupling is decoherence losses of lower polaritons by means of a Markovian master equa- of an atomic cloud induced by spontaneous emission of atoms tion:3 in the lattice [166, 167]. In this way, the many-body system is heated up, and therefore driven away from the low entropy ∂tρ = −i[HLP, ρ] + Ldρ, (9) state in which it was prepared initially. A detailed discussion of the microscopic physics and its long time dynamics can be where Ldρ encodes incoherent single-particle pumping and found in [45–47], see also the review article [48]. For bosonic losses, as well as two-body losses: atoms in optical lattices, the coherent dynamics is described by the Bose-Hubbard Hamiltonian [168, 169] Z † 2 X † U X Ldρ = dx γpD[ψ(x) ]ρ + γlD[ψ(x)]ρ + 2udD[ψ(x) ]ρ , H = −J b b + n (n − 1), (12) BH l m 2 l l (10) hl,mi l where which models bosonic atoms in terms of the creation and annihi- † lation operators [b , bm] = δlm in the lowest band of a lattice with 1 l D[L]ρ = LρL† − {L†L, ρ} (11) site indices l, m. The atoms hop between neighboring lattice sites 2 with an amplitude J, and experience an on-site repulsion U. The lattice potential V(x), which leads to the second quantized form reflects the Lindblad form, and γp, γl, 2ud are the rates of single- of the Bose-Hubbard model, is created by the superposition of particle pumping, single-particle loss, and two-body loss, re- counter-propagating laser beams in each spatial dimension. The spectively. The inclusion of the non-linear loss term ensures coherent laser field couples two internal atomic states via stimu- saturation of the pumping. An analogous mechanism is im- lated absorption and emission, which leads to the single particle plemented in the above-mentioned Gross-Pitaevskii description. Hamiltonian More precisely, in the spirit of universality, the above quantum 2 pˆ ∆ z x Ω(xˆ) master equation (9) and the above mentioned phenomenologi- Hatom = − σ − σ , (13) cal model reduce to precisely the same low frequency model for 2m 2 2 bosonic degrees of freedom upon taking the semiclassical limit where pˆ is the atomic momentum operator, ∆ is the detuning of in the Keldysh path integral associated to Eq. (9) (see Sec.IIC the laser from the atomic transition frequency, and Ω(xˆ) is the for its implementation), and integrating out the upper polariton laser field at the atomic position. For large detuning, the excited reservoir in the phenomenological model. state of the atom can be traced out, which leads to the lattice Hamiltonian pˆ 2 |Ω(xˆ)|2 Heff = + . (14) atom 2m 4∆ 3 While this approach captures the universal behavior, we note that non- This describes a lattice potential for the ground state atoms Markovian effects can be of key importance for other properties [160–162]. generated by the spatially modulated AC Stark shift. Adding 10
cial example for universality in out-of-equilibrium dynamics, which can be probed by cold atom experiments.
D. Outline and scope of this review
The remainder of this review is split into two parts. Figure 2. Illustration of the coherent (a) and the incoherent (b) con- Part1 develops the theoretical framework for the e fficient de- tribution to the dynamics stemming from the coupling of the atoms to scription of driven open many-body quantum systems. In Sec.II, a laser with amplitude |Ω| and large detuning ∆. (a) Via the AC Stark we begin with a direct derivation of the open system Keldysh effect, stimulated absorption and emission lead to a coherent periodic functional integral from the many-body quantum master equa- |Ω|2 potential with amplitude 4∆ . (b) Stimulated absorption and subsequent tion (Sec.IIA). We then discuss in Sec. II B 1 in detail a simple spontaneous emission lead to effective decoherence of the atomic state example: the damped and driven optical cavity. This allows the |Ω|2 with rate Γ 4∆ , where Γ is the microscopic spontaneous emission rate. reader to familiarize with the functional formalism. In particu- lar, the key players in terms of observables — correlation and re- sponse functions — are described. We also point out a number of an atomic interaction potential and expanding both the single exact structural properties of Keldysh field theories, which hold particle Hamiltonian (14) and the interaction in terms of Wan- beyond the specific example. Another example is introduced in nier states, the leading order Hamiltonian is the Bose-Hubbard Sec. II B 2: there we discuss the mean field theory of conden- model (12)[169]. sation in a bosonic many-body system with particle losses and In the semi-classical treatment of the atom-laser interaction, pumping. The semi-classical limit of this model and its valid- spontaneous emission events are neglected, as their probabil- ity are the content of Sec.IIC. This is followed by a discussion ity is typically very small (see below for a more precise state- of symmetries and conservation laws in the Keldysh formalism ment). They can be taken into account on the basis of optical in Sec.IID. In particular, we point out a symmetry that allows Bloch equations [166], which leads, after elimination of the ex- one to distinguish equilibrium from non-equilibrium conditions. cited atomic state, to an additional, driven-dissipative term in Finally, an advanced field theoretical tool — the open system the atomic dynamics. It describes the decoherence of the atomic functional renormalization group — is introduced in Sec.IIE. state due to spontaneous emission, i.e., position dependent ran- Part2 harnesses this formalism to generate an understand- dom light scattering. The leading order contribution to the dy- ing of the physics in different experimental platforms. We be- namics in the basis of lowest band Wannier states is captured by gin in Sec. III with simple but paradigmatic spin models with the master equation discrete Ising Z2 symmetry in driven non-equilibrium station- X ary states. In particular, we discuss the physics of the driven ∂tρ = −i[HBH, ρ] − γ [nl, [nl, ρ]], (15) open Dicke model in Sec. III A. This is followed by an extended l variant of the latter in the presence of disorder and a multi-
† mode cavity, which hosts an interesting spin and photon glass where nl = bl bl is the local atomic density, and ρ is the many- phase in Sec. III B. Sec.IV is devoted to the non-equilibrium 2 |Ω| U body density matrix. For a red detuned laser the rate γ = Γ 4∆2 is stationary states of bosons with a characteristic (1) phase ro- proportional to the microscopic spontaneous emission rate Γ and tation symmetry: the driven-dissipative condensates introduced the laser amplitude |Ω|. Note the suppression of the scale γ by in Sec. II B 2. After some additional technical developments re- a factor Γ/∆ 1 for large detuning, compared to the strength lating to U(1) symmetry in Sec.IVA, we discuss critical be- eff of Hatom. The coherent and incoherent contribution of the atom- havior at the Bose condensation transition in three dimensions. laser coupling to the dynamics is illustrated in Fig.2. In particular, we show the decrease of a parameter quantifying The dissipative term in the master equation (15) leads to an non-equilibrium strength in this case (Sec.IVB). The opposite energy increase hHBHi(t) ∼ t linear in time, and thus to heat- behavior is observed in two (Sec.IVC) and one (Sec.IVD) ing. Furthermore, it introduces decoherence in the number state dimensions. Finally, leaving the realm of stationary states, in basis, i.e., it projects the local density matrix on its diagonal in Sec.V we discuss an application of the Keldysh formalism to Fock space and leads to a decrease of the coherences in time [45– the time evolution of open bosonic systems in one dimension, 48]. Starting from a low entropy state at t = 0, the heating leads which undergo number conserving heating processes. We set up to a crossover from coherence dominated dynamics at short and the model in Sec.VA, derive the kinetic equation for the dis- intermediate times [43] to a decoherence dominated dynamics at tribution function in Sec.VB, and discuss the relevant approxi- long times [45–47]. In one dimension, both regimes have been mations and physical results in Secs.VC andVD, respectively. analyzed extensively both numerically (with a focus on deco- Conclusions are drawn in Sec.VI. Finally, brief introductions herence dominated dynamics) as well as analytically, and dis- to functional differentiation and Gaussian functional integration play several aspects of non-equilibrium universality, see Sec.V. are given in AppendicesA andB. Therefore, heating in interacting lattice systems represents a cru- Reflecting the bipartition of this review, the scope of it is 11 twofold. On the one hand, it develops the Keldysh functional in- ouvillian (sometimes this term is reserved for the second con- tegral approach to driven open quantum systems “from scratch”, tribution on the RHS of Eq. (16) alone). There are two contri- in a systematic and coherent way. It starts from the Marko- butions to the Liouvillian: first, the commutator term, which is vian quantum master equation representation of driven dissipa- familiar from the von Neumann equation, describes the coher- tive quantum dynamics [81, 82, 170], and introduces an equiv- ent dynamics generated by a system Hamiltonian H; the second alent Keldysh functional integral representation. Direct contact part, which we will refer to as the dissipator D 4, describes the is made to the language and typical observables of quantum op- dissipative dynamics resulting from the interaction of the sys- tics. It does not require prior knowledge of quantum field the- tem with an environment, or “bath.” It is defined in terms of a ory, and we hope that it will find the interest of — and be use- set of so-called Lindblad operators (or quantum jump operators) ful for — researchers working on quantum optical systems with Lα, which model the coupling to that bath. The dissipator has a many degrees of freedom. On the other hand, this review doc- characteristic Lindblad form [174, 175]: it contains an anticom- uments some recent theoretical progresses made in this concep- mutator term which describes dissipation; in order to conserve tual framework in a more pedagogical way than the original lit- the norm tr(ρ) of the system density matrix, this term must be ac- erature. We believe that this not only exposes some interesting companied by fluctuations. The corresponding term, where the physics, but also demonstrates the power and flexibility of the Lindblad operators act from both sides onto the density matrix, Keldysh approach to open quantum systems. is referred to as recycling or quantum jump term. Dissipation This work is complemetary to excellent reviews putting more occurs at rates γα which are non-negative, so that the density emphasis on the specific experimental platforms partially men- matrix evolution is completely positive, i.e., the eigenvalues of tioned above: The physics of driven Bose-Einstein condensates ρ remain positive under the combined dynamics generated by H in optical cavities is reviewed in [10]. A general overview of and D [176]. If the index α is the site index in an optical lattice driven ultracold atomic systems, specifically in optical lattices, or in a microcavity array, or even a continuous position label (in is provided in [48], and systems with engineered dissipation are which case the sum is replaced by an integral), in a translation described in [57]. Detailed accounts for exciton-polariton sys- invariant situation there is just a single scale γα = γ for all α tems are given in [13, 147, 148]; specifically, we refer to the associated to the dissipator. review [149] working in the Keldysh formalism. The physics of The quantum master equation (16) provides an accurate de- microcavity arrays is discussed in [14, 24, 134, 141], and trapped scription of a system-bath setting with a strong separation of ions are treated in [17, 171]. We also refer to recent reviews scales. This is generically the case in quantum optical systems, on additional upcoming platforms of driven open quantum sys- which are strongly driven by external classical fields. More pre- tems, such as Rydberg atoms [172] and opto-nanomechanical cisely, there must be a large energy scale in the bath (as com- settings [173]. For a recent exposition of the physics of quan- pared to the system-bath coupling), which justifies to integrate tum master equations and to efficient numerical techniques for out the bath in second-order time-dependent perturbation the- their solution, see [48]. ory. If in addition the bath has a broad bandwidth, the combined Born-Markov and rotating-wave approximations are appropri- ate, resulting in Eq. (16). A concrete example for such a set- Part 1 ting is provided by a laser-driven atom undergoing spontaneous emission. Generic condensed matter systems do not display such Theoretical background a scale separation, and a description in terms of a master equa- tion of the type (16) is not justified.5 However, the systems dis- cussed in the introduction, belong to the class of systems which II. KELDYSH FUNCTIONAL INTEGRAL FOR DRIVEN OPEN SYSTEMS permit a description by Eq. (16). We refer to them as driven open many-body quantum systems. Due to the external drive these systems are out of thermody- In this part, we will be mainly concerned with a Keldysh field namic equilibrium. This statement will be made more precise theoretical reformulation of the stationary state of Markovian in Sec. II D 1 in terms of the absence of a dynamical symme- many-body quantum master equations. As we also demonstrate, try which characterizes any system evolving in thermodynamic this opens up the powerful toolbox of modern quantum field the- equilibrium, and which is manifestly violated in dynamics de- ory for the understanding of such systems. scribed by Eq. (16). Its absence reflects the lack of energy con- The quantum master equation, examples of which we have servation and the epxlicit breaking of detailed balance. As stated already encountered in Sec.IC, describes the time evolution of a reduced system density matrix ρ and reads [81, 82]
! 4 We use the term “dissipation” here for all kinds of environmental influences X † 1 † ∂ ρ = Lρ = −i[H, ρ] + γ L ρL − {L L , ρ} , (16) on the system which can be captured in Lindblad form, including effects of t α α α 2 α α α decay and of dephasing/decoherence. 5 Davies’ prescription [177] allows one to also describe equilibrium systems where the operator L acts on the density matrix ρ “from both in terms of operatorial master equations, however with collective Lindblad sides” and is often referred to as Liouville superoperator or Li- operators ensuring detailed balance conditions (cf. also [82]). 12 in the introduction, the main goal of this review is to point out terms of a state vector evolution, even in the case of purely co- the macroscopic, observable consequences of this microscopic herent Hamiltonian dynamics.6 Instead, it is necessary to study violation of equilibrium conditions. the evolution of a state matrix, which transforms according to the integral form of the von Neumann equation in the second line of Eq. (17). Therefore, we have to apply the Trotter formula A. From the quantum master equation to the Keldysh functional and coherent state insertions on both sides of the density matrix. integral This leads to the doubling of degrees of freedom, characteris- tic of the Keldysh functional integral. Moreover, time evolution In this section, starting from a many-body quantum master can now be interpreted as occurring along two branches, which equation Eq. (16) in the operator language of second quantiza- we denote as the forward and a backward branches, respectively tion, we derive an equivalent Keldysh functional integral. We (cf. Fig.3 b)). Indeed, this is an intuitive and natural feature of focus on stationary states, and we discuss how to extract dynam- evolving matrices instead of vectors. ics from this framework in Sec.V. Our derivation of the Keldysh So far, we have concentrated on closed systems which evolve functional integral applies to a theory of bosonic degrees of free- according to purely Hamiltonian dynamics. However, we can dom. If spin systems are to be considered, it is useful to first per- allow for a more general generator of dynamics and still proceed form the typical approximations mapping them to bosonic fields, along the two-branch strategy. The most general (time local) and then proceed along the construction below (see Sec. III and evolution of a density matrix is given by the quantum master Refs. [112, 114]). Clearly, this amounts to an approximate treat- equation (16). Its formal solution reads ment of the spin degrees of freedom; for an exact (equilibrium) (t−t0)L N functional integral representation for spin dynamics, taking into ρ(t) = e ρ(t0) ≡ lim (1 + δtL) ρ(t0). (19) N→∞ account the full non-linear structure of their commutation rela- tions, we refer to Ref. [178]. For fermionic problems, the con- The last equality gives a meaning to the formal solution in terms struction is analogous to the bosonic case presented here. How- of the Trotter decomposition: at each infinitesimal time step, the ever, a few signs have to be adjusted to account for the fermionic exponential can be expanded to first order, such that the action anticommutation relations [102, 178]. of the Liouvillian superoperator is just given by the RHS of the The basic idea of the Keldysh functional integral can be de- quantum master equation (16); At finite times, the evolved state veloped in simple terms by considering the Schrodinger¨ vs. the is given by the concatenation of the infinitesimal Trotter steps. von Neumann equation, If we restrict ourselves to the stationary state of the system,7 but want to evaluate correlations at arbitrary time differences, i∂ |ψ(t)i = H|ψ(t)i ⇒ |ψ(t)i = U(t, t )|ψ(t )i, t 0 0 we should extend the time branch from an initial time in the dis- † ∂tρ(t) = −i[H, ρ(t)] ⇒ ρ(t) = U(t, t0)ρ(t0)U (t, t0), tant past t0 → −∞ to the distant future, t f → +∞. In analogy to (17) thermodynamics, we are then interested in the so-called Keldysh partition function Z = tr ρ(t) = 1. The trace operation contracts −iH(t−t0) where U(t, t0) = e is the unitary time evolution opera- the indices of the time evolution operator as depicted in Fig.3 tor. In the first case, a real time path integral can be constructed c), giving rise to the closed time path or Keldysh contour. Con- along the lines of Feynman’s original path integral formulation servation of probability in the quantum mechanical system is of quantum mechanics [179]. To this end, a Trotter decomposi- reflected in the time-independent normalization of the partition tion of the evolution operator function. In order to extract physical information, again in anal-
−iH(t−t0) N ogy to statistical mechanics, below we introduce sources in the e = lim (1 − iδtH) , (18) N→∞ partition function. This allows us to compute the correlation and response functions of the system by taking suitable variational with δ = t−t0 , is performed. Subsequently, in between the t N derivatives with respect to the sources. factors of the Trotter decomposition, completeness relations in After this qualitative discussion, let us now proceed with the terms of coherent states are inserted in order to make the (nor- explicit construction of the Keldysh functional integral for open mal ordered) Hamilton operator a functional of classical field systems, starting from the master equation (16). As mentioned variables. This is illustrated in Fig.3 a), and we will perform above, the Keldysh functional integral is an unraveling of Li- these steps explicitly and in more detail below in the context ouvillian dynamics in the basis of coherent states, and we first of open many-body systems. Crucially, we only need one set of field variables representing coherent Hamiltonian dynamics, which corresponds to the forward evolution of the Schrodinger¨ state vector. It is also clear that — noting the formal analogy of 6 Of course, the evolution of an M × M matrix, where M is the dimension of the the operators e−iH(t−t0) and e−βH — this construction can be lever- Hilbert space, can be formally recast into the evolution of a vector of length aged over to the case of thermal equilibrium, where the “Trotter- M × M. While such a strategy is often pursued in numerical approaches [48], ization” is done in imaginary instead of real time. it does in general not allow for a physical interpretation. In contrast to these special cases, the von Neumann equation 7 We assume that it exists. We thus exclude scenarios with dynamical limit for general mixed state density matrices cannot be rewritten in cycles, for simplicity. 13
a) |Ωi represents the vacuum in Fock space. (Note that accord- ing to this definition, which is usually adopted in the discus- (t0) t t t0 | i sion of field integrals [180], the state |ψi is not normalized.) U A key property of coherent states is that they are eigenstates of the annihilation operator, i.e., b |ψi = ψ |ψi, with the com- b) plex eigenvalue ψ. Clearly, this implies the conjugate relation ⇢(t0) hψ| b† = hψ| ψ∗.8 The overlap of two non-normalized coherent t t ∗ U U † t states is given by hψ|φi = eψ φ, and the completeness relation R ∗ ∗ reads 1 = dψdψ e−ψ ψ |ψi hψ|. c) + contour π The starting point of the derivation is Eq. (19), and we focus ⇢(t ) ⇢(t ) f 0 first on a single time step, as in the usual derivation of the co- tf =+ - contour t = 1 0 1 herent state functional integral [180]. That is, we decompose the time evolution from t0 to t f into a sequence of small steps of du- Figure 3. Idea of the Keldysh functional integral. a) According to the ration δt = (t f − t0)/N, and denote the density matrix after the Schrodinger¨ equation, the time evolution of a pure state vector is de- n-th step, i.e., at the time tn = t0 + δtn, by ρn = ρ(tn). We then −iH(t−t0) scribed by the unitary operator U(t, t0) = e . In the Feynman have functional integral construction, the time evolution is chopped into in- finitesimal steps of length δt, and completeness relations in terms of ρ = eδtLρ = (1 + δ L) ρ + O(δ2). (20) coherent states are inserted in between consecutive time steps. This n+1 n t n t As anticipated above, we proceed to represent the density matrix insertion is signalled by the red arrows. b) In contrast, if the state is mixed, a density matrix must be evolved, and thus two time branches in the basis of coherent states. For instance, ρn at the time tn can are needed. As explained in the text, the dynamics need not necessarily be written as be restricted to unitary evolution. The most general time-local (Marko- Z ∗ ∗ vian) dynamics is generated by a Liouville operator in Lindblad form. dψ+,ndψ+,n dψ−,ndψ−,n − ∗ − ∗ ρ = e ψ+,nψ+,n ψ−,nψ−,n c) For the analysis of the stationary state, we are interested in the real n π π time analog of a partition function Z = tr ρ(t f ), starting from t0 = −∞ × hψ |ρ |ψ− i |ψ i hψ− | . (21) and running until t f = +∞. The trace operation connects the two time +,n n ,n +,n ,n branches, giving rise to the closed Keldysh contour. As a next step, we would like to express the matrix element hψ+,n+1|ρn+1|ψ−,n+1i, which appears in the coherent state repre- collect a few important properties of those. Coherent states are sentation of ρn+1, in terms of the corresponding matrix element defined as (for simplicity, in the present discussion we restrict at the previous time step tn. Inserting Eq. (21) in Eq. (20), we ourselves to a single bosonic mode b) |ψi = exp(ψb†) |Ωi, where find that this requires us to evaluate the “supermatrixelement”