Arxiv:2005.12299V2 [Hep-Th] 3 Jun 2020

CERN-TH-2020-080

QCD thermalization: Ab initio approaches and interdisciplinary connections

J¨urgenBerges∗ Institute for Theoretical Physics, Heidelberg University, Philosophenweg 16, D-69120 Heidelberg, Germany

Michal P. Heller† Max Planck Institute for Gravitational Physics (Albert Einstein Institute), Am M¨uhlenberg 1, D-14476 Potsdam, Germany

Aleksas Mazeliauskas‡ Theoretical Physics Department, CERN, CH-1211 Geneva 23, Switzerland

Raju Venugopalan§ Physics Department, Brookhaven National Laboratory, Upton, New York 11973-5000, USA

Heavy-ion collisions at BNL’s Relativistic Heavy Ion Collider and CERN’s Large Hadron Collider provide strong evidence for the formation of a quark-gluon plasma, with temperatures extracted from relativistic viscous hydrodynamic simulations shown to be well above the transition temperature from hadron matter. Outstanding problems in QCD include how the strongly correlated quark- gluon matter forms in a heavy-ion collision, its properties off-equilibrium, and the thermalization process in the plasma. We review here the theoretical progress in this field in weak coupling QCD effective field theories and in strong coupling holographic approaches based on gauge-gravity duality. We outline the interdisciplinary connections of different stages of the thermalization process to non- equilibrium dynamics in other systems across energy scales ranging from inflationary cosmology, to strong field QED, to ultracold atomic gases, with emphasis on the universal dynamics of non-thermal and hydrodynamic attractors. We survey measurements in heavy-ion collisions that are sensitive to the early non-equilibrium stages of the collision and discuss the potential for future measurements. We summarize the current state-of-the art in thermalization studies and identify promising avenues for further progress.

hydrodynamic attractor

non-thermal attractor arXiv:2005.12299v3 [hep-th] 17 Aug 2021

∗ [email protected] † [email protected]; On leave from: National Centre for ‡ aleksas.mazeliauskas@cern.ch Nuclear Research, Pasteura 7, Warsaw, PL-02093, Poland § [email protected] 2

CONTENTS 1. Real-time simulations for fermions and gauge fields beyond the I. Big picture questions and outline of the review3 classical-statistical approximation 31 2. Real-time off-equilibrium dynamics of II. Hadron structure at high energies6 quantum anomalies 32 A. Quantum Chromodynamics7 B. QCD at small x and high parton densities7 V. Equilibration in QCD kinetic theory 33 C. Effective Field Theory for high parton A. The quasi-particle description of QCD densities: the Color Glass Condensate8 plasmas 33 D. Renormalization group evolution in the 1. Chiral kinetic theory 33 CGC EFT 10 B. Leading order kinetic theory 34 E. DIS and the dipole model 11 1. Elastic two-body scattering 34 F. RG evolution and geometric scaling 12 2. Fokker-Planck limit of elastic G. The state of the art in the CGC EFT 13 scatterings 35 3. Effective collinear one-to-two processes 35 III. Non-equilibrium QCD matter at high 4. Bethe-Heitler and LPM limits of collinear occupancy 14 radiation 36 A. Multi-particle production in strong fields 15 C. Bottom-up thermalization 36 B. The LO Glasma: classical gluon fields from 1. Initial conditions 36 shockwave collisions 16 2. Stage one: collisional broadening 37 C. Non-perturbative evolution of high 3. Stage two: collinear cascade 37 occupancy fields 17 4. Stage three: mini-jet quenching 37 1. Real time evolution of boost invariant 5. Numerical realization of bottom-up fields on the lattice 17 thermalization 38 2. Glasma flux tubes 18 D. Self-similar evolution in the high-occupancy 3. The IP-Glasma model 19 regime 39 D. The Glasma at NLO 20 1. Self-similar scaling 39 1. Dynamics of pη = 0 modes: QCD 2. Pre-scaling phenomenon 39 factorization and energy evolution 20 E. Extrapolation to stronger couplings 40 2. Dynamics of pη = 0 modes: plasma 1. Hydrodynamic attractors in QCD kinetic instabilities and6 the classical-statistical theory 40 approximation 21 2. Entropy production and initial energy density 41 IV. Far-from-equilibrium gluon and quark 3. Chemical equilibration of QGP 42 production: From plasma instabilities to 4. Equilibration of spatially inhomogeneous non-thermal attractors 22 systems 43 A. Non-equilibrium time evolution equations from the quantum effective action 23 VI. Ab initio holographic description of strong 1. Macroscopic field, spectral and statistical coupling phenomena 44 functions 23 A. Holography and heavy-ion collisions 44 2. Resummed evolution equations to leading B. Controlled strong coupling regime 44 order 24 C. Early times in Bjorken flow at strong B. Nonlinear evolution of plasma instabilities 24 coupling 46 1. Dynamical power counting 25 D. Hydrodynamic attractors in holography 47 2. Classical-statistical field theory limit 25 E. Holographic collisions 48 C. Non-thermal attractor 26 1. Planar shocks 49 1. Far-from-equilibrium universal scaling 27 2. Transverse dynamics in holography 50 2. Identifying the weak-coupling F. Other aspects of thermalization at strong thermalization scenario 27 coupling 50 3. Non-thermal attractors in scalar field 1. Non-conformal strongly-coupled QFTs 50 theories 28 2. Away from the strong coupling regime 51 D. Far-from-equilibrium separation of scales 3. Non-local correlators 51 and ultrasoft scale dynamics 29 1. Non-equilibrium evolution of the spatial VII. Signatures of non-equilibrium QCD 52 Wilson loop 29 A. Electromagnetic and hard probes 52 2. Effective condensate dynamics 30 B. Long-range rapidity correlations 53 E. Early-time fermion production and C. Bulk observables 53 quantum anomalies 30 D. Future prospects 54 3

VIII. Interdisciplinary connections 54 tions of the early stages of heavy-ion collisions has been A. Strong interactions: Unitary Fermi gas 55 made relatively recently. In particular, there is a grow- B. Highly occupied systems I: Preheating in ing realization that the far-from-equilibrium dynamics the early Universe 55 that characterizes early time physics is extremely im- C. Highly occupied systems II: Bose gases far portant for understanding collective phenomena in the from equilibrium 56 heavy-ion experiments [18–20]. This review summarizes D. Highly occupied systems III: our perspective on the theoretical and phenomenologi- Classicalization and unitarization of cal progress in this active research area and places these gravitational amplitudes 56 developments in a wider interdisciplinary context. E. Anomalous currents in non-equilibrium The QCD thermalization process represents an initial QED: Condensed matter systems and strong value problem in quantum field theory (QFT). It re- laser fields 57 quires understanding the many-body correlations in the F. Thermalization and entanglement 57 colliding hadrons, how such correlations influence multi- particle production as the collision occurs, and the sub- IX. Summary and Outlook 57 sequent effective loss of information of these many-body correlations during the thermalization process of the mat- Acknowledgments 59 ter produced. While von Neumann entropy is conserved in the unitary quantum evolution of a nuclear collision References 59 in isolation, observables of interest may nevertheless approach (local) thermal equilibrium. The characteristic time scales for the corresponding effective loss of infor- I. BIG PICTURE QUESTIONS AND OUTLINE mation and the extent to which the dynamics finally leads OF THE REVIEW to an approach to (local) thermal equilibrium for key observables are the central topics of this review. In particular, we will focus on the following key ques- Ultrarelativistic collisions of heavy nuclei at the BNL tions prompted by the dynamics of each stage of the Relativistic Heavy Ion Collider (RHIC) and the CERN spacetime evolution of quark-gluon matter in heavy-ion Large Hadron Collider (LHC) produce several thousand collisions1: particles in each event, generating the hottest and dens- est matter on Earth [1–8]. At the highest LHC energies, • What are the many-body correlations of strongly in- temperatures of the order of five trillion Kelvin are at- teracting matter in the colliding nuclei? tained [9]. Temperatures on this scale previously existed only at the earliest instants of our Universe, a 10th of a The colliding nuclei produce the initial state for microsecond after the Big Bang. Lattice gauge theory the subsequent thermalization process. In princi- studies [10] show strongly interacting matter at these ple, there can be different thermalization scenarios temperatures to be well over a crossover temperature for different initial conditions. Although many de- from hadron matter to a regime where the degrees of free- tails of the quantum evolution are lost quickly, it dom describing bulk thermodynamic quantities are the is crucial to classify the range of initial conditions fundamental quark and gluon fields of Quantum Chromo- (such as underoccupied versus overoccupied) lead- dynamics (QCD). The results of experimental and the- ing to a certain class of dynamical processes. oretical studies indicate that shortly after the heavy-ion In QCD, a proton (or any other nucleus) must be collision, the produced quark-gluon fields form a strongly viewed as a collection of short or long lived config- correlated state of matter, widely known as the quark- urations of partons (quarks, antiquarks, and glu- gluon plasma (QGP) [11]. ons), where each configuration carries the quantum The heavy-ion experiments at RHIC and LHC there- numbers of the proton. When the proton or nu- fore provide us with a unique opportunity to study terres- cleus is boosted to high energies, short lived config- trially the spacetime evolution of this non-Abelian QGP. urations typically containing large numbers of par- A striking finding from the RHIC and LHC experiments tons live much longer due to time dilation. It is is that the experimental data are consistent with a de- therefore more likely that a probe of the hadron scription of the QGP as a nearly perfect fluid with a at high energies will scatter off such many-body very low value of shear viscosity to entropy density ratio configurations of partons and that their decay will of η/s 0.2 (in natural units) [12]. These values are very dominate the physics of multi-particle production close to≤η/s = 1/(4π), a universal property of a class of in ultrarelativistic nuclear collisions. Learning how gauge theories with a large number of degrees of freedom precisely multi-particle production occurs requires at infinite coupling [13–16] that is described in terms of a dual gravity picture [17]. While our understanding of the thermal properties of QGP matter and the flow of the nearly perfect fluid has 1 For a complementary perspective on open questions in heavy-ion developed significantly, progress in theoretical descrip- collisions, we refer the reader to Ref. [19]. 4

a deep knowledge of the spatial and momentum include far-from-equilibrium attractors associated distributions of partons in the boosted nuclei, the with non-thermal fixed points [26–28]. nature of their correlations, and how these corre- Far-from-equilibrium hydrodynamic attractors are lations change with system size and with collision observed to emerge in both strong and weak cou- energy. pling [29, 30]. A related important set of questions concerns the use of effective theories like hy- • What is the physics of the first yoctosecond drodynamics for systems that are far away from (10 24 seconds) of the collision? − equilibrium. Yet another line of inquiry is to de- Parton configurations in a boosted nucleus have termine how features of the dynamics evolve be- their momenta distributed between a few fast tween the weakly coupled and strongly coupled modes and more plentiful soft modes. In a heavy- regimes. An intriguing possibility to consider is ion collision, these fast modes in each of the nuclei whether the topological properties of strongly cor- interact relatively weakly with the other nucleus related systems may help provide unifying descrip- and populate the “fragmentation regions” corre- tions at both weak and strong coupling. sponding to polar angles very close to the beam axes [21]. The slower degrees of freedom interact • Can we cleanly isolate signatures of quark-gluon more strongly with each other and produce strongly matter off-equilibrium? interacting gluon matter outside the fragmentation If matter in bulk locally equilibrates in heavy- regions. ion collisions, the only information of the non- This spacetime picture of nuclear collisions was equilibrium evolution that survives is what is im- developed in a groundbreaking paper by Bjorken printed as initial conditions for its subsequent hy- to describe the subsequent hydrodynamic flow of drodynamic evolution. The exceptions are elec- the quark-gluon plasma [22], albeit he did not ad- troweak and so-called “hard probes”; both of these dress how thermalization occurs in this scenario. are sensitive to the full history of the spacetime An interesting question in this regard is whether evolution of QCD matter. the strong interactions of the soft modes with each A significant development in recent years is the other are due to strong coupling or whether they vastly improved ability of the RHIC and LHC ex- may be due to the large occupancy of these soft periments to perform “event engineering” whereby modes. The answer to this question may also influ- final states can be studied by varying the “control ence the degree of transparency of the fast modes, parameters” corresponding to nuclear size, central- in particular a “limiting fragmentation” scaling ity of collision impact, and final state multiplicities phenomenon seen in data. (triggered thereby on typical versus rare event con- A spacetime scenario in which both soft and hard figurations) across wide ranges in energy and sys- modes in the nuclei interact very strongly and gen- tem size [31]. A challenging question is whether we erate hydrodynamic flow was suggested by Landau. can constrain the current state-of-the-art computa- It is conceivable that there is a transition between tional techniques to accurately reflect the system- these two spacetime pictures with energy [23, 24]; atics of this event engineering and, further, to use if so, can they be distinguished by phenomena such these to isolate empirically the out-of-equilibrium as limiting fragmentation [25]? dynamics.

• Is there a unifying theoretical description of quark- • Interdisciplinary connections gluon matter off-equilibrium? The study of the out-of-equilibrium dynamics of The quark-gluon matter formed in the first few yoc- strongly correlated systems is an important topic toseconds of the heavy-ion collision is very far from of significant contemporary interest in a number of equilibrium. A key question in its description is sub-fields of physics. As we later discuss, the ideas whether weak and strong coupling extrapolations and methods outlined in this review have significant to realistic values can lead to similar phenomenol- overlap with these fields. Can one exploit these ogy. interdisciplinary connections to make progress? A potentially rich line of inquiry is to isolate We will address the previously listed outstanding ques- which features of the non-equilibrium evolution of tions in two ab initio theoretical approaches to the prob- strongly correlated or coupled quark-gluon matter lem of thermalization in QCD. One approach, the Color are universal. One example is universal dynamics Glass Condensate Effective Field Theory (CGC EFT), in the approach to local thermal equilibrium gov- is applicable at very high energies corresponding to a erned by viscous hydrodynamics. Another example regime of very weak coupling α 1 and very high gluon S is universality in time dependence across a class of occupancies fg satisfying αSfg 1. This regime of weak non-equilibrium states for certain observables. In a coupling and high occupancies∼ in QCD is characterized weak coupling scenario, at high occupancies, these by a large emergent “saturation” scale that is much larger 5 than the intrinsic non-perturbative scales corresponding density matrix has to be parametrized at the initial scale to color confinement and chiral symmetry breaking. by physically plausible assumptions, a renormalization The CGC EFT employs weak coupling many-body group (RG) framework [39, 40] allows one to study sys- methods to separate (or factorize) these soft non- tematically the energy evolution of parton many-body perturbative modes from the harder modes of the order correlations as the hadron is boosted to higher energies. of the saturation scale. Specifically, the requirement that In SectionIII, we will outline the problem of multi- physics be independent of the scale separation between particle production in quantum field theory in the pres- soft and hard modes leads to renormalization group equa- ence of strong fields and discuss how this leads to a first tions that describe how such non-perturbative informa- principles description of the very early time evolution of tion provided as an input at a given energy scale changes the Glasma. Inclusive quantities such as multiplicities or as it evolves. As one approaches asymptotic energies, the energy densities, and their spacetime correlations, can be factorization of the hard and soft scales becomes increas- computed systematically in the Glasma in powers of the ingly robust and many of the properties of quark-gluon coupling αS 1 at sufficiently high energy. At leading matter can be computed systematically. The quark-gluon order in this power counting, the Glasma fields are highly matter in this limit is called the Glasma [32, 33]. occupied classical fields, with magnitude 1/αS. The other ab initio approach to thermalization is in the At next-to-leading order (NLO), we discuss how quan- limit of strong ’t Hooft coupling of α N , as the tum fluctuations, co-moving with the colliding nuclei, S c → ∞ number of colors Nc . In this limit, holographic ap- can be absorbed into the density matrices describing proaches based on gauge-gravity→ ∞ duality [17, 34, 35] are their non-perturbative many-body color charge distribu- robust and can be used to obtain exact results in non- tions. In contrast, non-comoving quantum fluctuations Abelian gauge theories, with the best understood exam- produced after the collision in the Glasma are unstable ple being = 4 superconformal Yang-Mills theory. and display quasi-exponential dynamical growth [41]. We N Neither of these theoretical approaches to the problem later describe how the physics of these unstable modes at very early proper times τ 1 log2(1/α ) is captured of thermalization are directly applicable to real world . QS S heavy-ion collisions at RHIC and LHC energies, where in a classical-statistical approximation of the quantum the relevant couplings are likely neither particularly weak evolution with given quantum initial conditions. nor infinitely strong. Thus data-theory comparisons rely SectionIV describes the non-linear time evolution of on phenomenological descriptions characterized by ex- far-from-equilibrium quark gluon matter for weak cou- trapolations of ab initio approaches well beyond their plings relevant at very high energies. The range of va- strict regimes of validity. By anchoring such phenomeno- lidity of classical-statistical field theory descriptions for logical models in fundamental theory in well controlled the evolution is discussed in terms of the two-particle- limits, their success or failure in comparisons to data irreducible (2PI) quantum effective action, which moti- can then be traced to a particular set of assumptions vates fully 3+1-dimensional numerical simulations of the in the extrapolations. We will clarify throughout the re- expanding Glasma fields. view whenever such phenomenological extrapolations are The lattice field theory simulations demonstrate the made. emergence of a non-thermal attractor described by a self- We will begin in SectionII by discussing the structure similar gluon distribution, whose dependence on momen- of matter within the colliding hadrons and heavy nuclei tum, and an overall cooling rate, are characterized by at high energies. After a brief introduction to QCD and universal numbers independent of the initial conditions. the associated parton picture of hadrons at high ener- Because the numerical simulations correctly describe dy- gies, we will focus our attention on what happens when namics in the infrared, the attractor solution helps one to the phase space density of partons in the wavefunctions of identify the right effective kinetic theory among several the colliding hadrons becomes large. Driving this physics competing options. is an emergent energy-dependent close packing “satura- Kinetic theory increasingly captures the relevant dy- tion” scale QS [36], which grows with energy and nu- namics of the thermalization process as the system ex- clear size, allowing for a systematic weak coupling de- pands and cools. In SectionV, we discuss the leading scription of the properties of saturated partons in high order kinetic theory framework, going progressively from energy QCD. Specifically, we will discuss the CGC EFT, elastic 2 2 scatterings to effective collinear 1 2 pro- wherein the high energy hadron is described as a coherent cesses and↔ taking special note of interference and↔ plasma state of static color sources and dynamical gluon fields. instability effects. Phenomenological extrapolations to The saturation scale is manifest in the CGC EFT, allow- realistic couplings can also be explored in the language ing one to describe strongly correlated many-body parton of hydrodynamic attractors, where the dependence on correlations in the hadron wavefunctions [37, 38]. the coupling is replaced by the kinematic viscosity η/s. Non-perturbative soft modes of the high energy nuclei, For values of the kinematic viscosity extracted from hy- their color charge distributions, and many-body correla- drodynamic simulations of heavy-ion collision, reason- tions thereof, are represented by a density matrix at a able predictions are obtained for entropy production [42], given energy scale that is much smaller than those of the as well as for hydrodynamic and chemical equilibration hard weakly coupled modes. While this non-perturbative times [43, 44]. 6

In SectionVI, we provide an overview of holography derlying the non-thermal attractor in the Glasma, which based strong coupling approaches to thermalization in has been widely discussed in the context of reheating in gauge theories. Our focus is on the conceptual features, the early Universe following inflation [26, 50]. The latter universal mechanisms, and predictions from these stud- in turn is, in the perturbative high-momentum regime, ies. In particular, ab initio holographic computations a relativistic generalization of weak wave turbulence in predict the applicability of hydrodynamics over a time fluids [51]. In the non-perturbative infrared regime, the scale set by the local energy density, when the expanding Glasma attractor is nearly identical to that of overoccu- matter in heavy-ion collisions settings is characterized pied cold atomic gases, sharing the same scaling functions by a large spatial anisotropy in its energy-momentum and exponents in a wide spectral range [52]. This is sug- tensor [45–47]. This is at variance with the common pre- gestive of a classification of far-from-equilibrium systems sumption of local thermal equilibrium in applying hydro- into universality classes analogous to those for critical dynamics; in a paradigm shift, the transition to hydro- phenomena close to equilibrium [53]. An exciting de- dynamic flow is now referred to as hydrodynamization velopment with cross-disciplinary potential is the use of rather than thermalization [48]. state-of-the-art cold atom experiments to provide deep We will discuss, in particular, phenomenological at- insight into such universal dynamics [54–56]. tempts to apply these ideas to model heavy-ion collisions The search for effective theories far from equilibrium in the context of (1+1)-dimensional boost invariant flow is also a major research direction in the theory of com- where hydrodynamization and hydrodynamic attractors plex systems ranging from understanding entanglement were first discovered. We will also cover work on more to information loss and thermalization of closed quan- realistic holographic descriptions of heavy-ion collisions tum many-body systems, with insights to be gained from that model confinement, the breaking of conformal in- “tabletop” atomic and condensed matter systems [57]. variance, the running of the coupling, and large Nc sup- On the other end of the energy scale are the connections pressed non-local correlations. to black holes and string theory with respect to general SectionVII is devoted to a discussion of signatures questions regarding the scrambling of information [58, 59] of non-equilibrium dynamics in heavy-ion data. While and the unitary dynamics underlying black hole forma- electromagnetic and high transverse momentum strongly tion and evaporation [60–65]. interacting final states are sensitive to early time dynam- Finally, the role of topology in heavy-ion collisions has ics, significant contributions to their rates accrue from all interdisciplinary connections in the chiral magnetic effect stages of the spacetime evolution of the system. Measure- which is now observed in condensed matter systems [66]. ments of long range correlations among high momentum Continual advances in laser technology also offer great final states offer promise in isolating the early time non- promise in the precision study of anomalous currents off equilibrium dynamics of the Glasma from the late stage equilibrium. hydrodynamic flow. This can be achieved by “event en- We end the review in SectionIX with a brief summary gineering” the response of these final states to variations and outlook toward future developments in our under- in energy and system size. We will also discuss how bulk standing of thermalization in QCD. As the outline sug- observables, in combination with these final states, can gests, thermalization in QCD is a rich field with many constrain thermalization scenarios. research directions and we have had to make choices in A striking example of the role of topology in heavy- our presentation due to space limitations. An impor- ion collisions is the Chiral Magnetic Effect (CME) [49] tant topic that we do not address is the off-equilibrium corresponding to a vector current along the direction of dynamics of QCD matter in the vicinity of a critical an external magnetic field that is induced by topological point [67–69]. Another is the related topic of hydro- transitions. The CME is primarily an early time effect; dynamic fluctuations [70, 71]. Other noteworthy omis- in this case as well, event engineering of multi-particle sions in our presentation include the discussion of holo- correlations offers the possibility of uncovering its role. graphic deep inelastic scattering [72–74], holographic In SectionVIII, we will address the question of the in- hard probes [75–79] and features of linear response the- terdisciplinary connections of the thermalization process ory [80–82]. Some aspects of holographic approaches that in heavy-ion collisions to that of other strongly corre- we omit or treat only partially were discussed in [48, 83– lated systems across energy scales. A striking similarity 86]. of strongly correlated flow in heavy-ion collisions to that of unitary Fermi gases was already noted shortly after the discovery of the QGP perfect fluid. The Glasma likewise II. HADRON STRUCTURE AT HIGH shares common features with other overoccupied systems ENERGIES across energy scales, from inflationary dynamics in the early Universe to a quantum portrait of black holes as The initial value problem of the thermalization process highly occupied graviton states to those of overoccupied in hadron-hadron collisions requires a deep understand- ultracold Bose gases. ing of the structure of QCD matter in the wavefunctions A concrete example of the influence of interdisciplinary of the colliding hadrons. The spacetime picture since ideas is that of the turbulent thermalization process un- the early days of QCD is that the highly Lorentz con- 7 tracted large x valence partons in the ultrarelativistic Because QCD is a confining theory, it is not analyti- hadron wavefunctions go through unscathed in the colli- cally tractable in general and numerical methods are es- sion, while their accompanying small x “fur coat of wee- sential to uncover its properties. Euclidean lattice Monte parton vacuum fluctuations” [87] interacts strongly to Carlo methods can be applied to compute, with good ac- form hot and dense matter [22]. The wee parton phase curacy, “static” properties of the theory such as the mass space distributions evolve with energy and nuclear size; spectrum of hadrons, magnetic moments, and thermody- their properties determine key features of the bulk prop- namic properties of QCD at finite temperature [89, 90]. erties of the matter produced after the collision. These methods are, however, very limited in deter- In this section, after a brief introduction to QCD and mining dynamical “real time” features of theory because the parton model of hadrons at high energies we will dis- of the contributions of a large number of paths to the cuss significant developments in the description of hadron QCD path integral in Minkowski spacetime. There are wavefunction in the CGC EFT. In particular, we will ad- promising approaches to surmount this difficulty such dress how the semi-hard saturation scale QS arises in the as steepest descent Lefshetz thimble methods but they nuclear wavefunctions, which justifies their description as are currently applicable only to problems in 1+1 dimen- highly occupied gluon shockwaves. As the largest scale sions [91]. Likewise, quantum computing offers an al- in the problem, it not only sets the scale for many-body ternative paradigm for computing real time dynamics, correlations in these shockwaves, and in the Glasma mat- but its applicability to QCD likely remains far in the ter produced after the collision, but subsequently also future [92]. determines the thermalization time and the initial tem- One should note that the production of high trans- perature of the quark-gluon plasma. verse momentum and massive particles (jets and heavy quarkonia being two notable examples) can be computed with high precision in perturbative QCD (pQCD) [93]. A. Quantum Chromodynamics This is because these processes correspond to very short transverse distances and asymptotic freedom tells us that Quantum Chromodynamics (QCD), the modern the- the QCD coupling αS is weak at these scales. ory of the strong force in nature, is a nearly perfect theory, with the only free parameters being the quark masses [88]. The Lagrangian of the theory can be writ- B. QCD at small x and high parton densities ten compactly as A great success of pQCD is the QCD parton model [94], 1 a µν,a X f µ f wherein the complex dynamics of quark and gluon fields QCD = F F + Ψ¯ (iγ Dµ,ij mf δij)Ψ . L −4 µν i − j in hadrons can, at high energies and large momentum f (1) resolutions, be viewed as that of a weakly interacting gas a a a abc b c of partons (single-particle quark, antiquark, and gluon Here Fµν = ∂µAν ∂ν Aµ gf AµAν is the QCD field − − a states). The cleanest way to access this sub-nucleon strength tensor for the color gauge fields Aµ that live in the adjoint representation of SU(3), with a = 1, , 8 structure is in the deeply inelastic scattering (DIS) of and f abc the structure constants of the gauge group.··· electrons or other leptons off nucleons and nuclei, wherein The quark fields live in the fundamental representation a virtual photon emitted by the electron strikes a quark of SU(3) and are labeled with their color and flavor in- or antiquark inside the hadron. f For the thermalization process of interest in this re- dices Ψ where the color index i = 1, , 3 and f de- i view, the asymptotic high energy (or “Regge”) limit of notes the flavors of quarks with masses ···m . Finally, the f DIS is most relevant. This limit corresponds to the Dirac matrix γµ is contracted with the covariant deriva- Bjorken DIS variable x Q2/s 0 where Q2 is the tive D = ∂ δ + igta Aa , with ta the generators of Bj µ,ij µ ij ij µ ij squared four-momentum transfer∼ and→ s is the squared SU(3) in the fundamental representation. center-of-mass energy. In the parton model, x x, The theory is rich in symmetry. The structure of the Bj where x is the light cone fraction of the momentum≈ of Lagrangian is dictated by the invariance of the quark and the hadron carried by the struck parton2. At small x, gluon fields under local SU(3) color gauge transforma- or equivalently at high energies, the number of partons tions. In addition, for massless quarks the theory has a in the hadron proliferate rapidly, as first observed in global chiral SU(3) SU(3) symmetry, global baryon L R DIS experiments at the HERA collider in Germany [95– number U(1) and axial× charge U(1) symmetries, and V A 100]. This growth is consistent with the predictions the quark and gluon fields are invariant under scale trans- of the pQCD Dokshitzer-Gribov-Lipatov-Altarelli-Parisi formations. The Lagrangian is invariant under discrete (DGLAP) [101–104] evolution equations. parity, charge, and time reversal symmetries. All of these symmetries, except that of local SU(3) color, are broken by vacuum or quantum effects that give rise to all the emergent phenomena in the theory, includ- 2 In hadron-hadron collisions, it is more appropriate to speak in ing confinement, asymptotic freedom, quantum anoma- terms of momentum fractions, so we henceforth use x instead of lies and the spontaneous breaking of chiral symmetry. xBj. 8

2 2 QS, where parametrically, for Q = QS the maximal occupancy is equated to the gluon phase space density as 1 xG (x, Q2 ) = A S . (2) α (Q ) 2(N 2 1)πR2 Q2 S S c − A S Fig.1 illustrates the gluon saturation phenomenon and the interpretation of QS as the emergent “close packing” scale.

FIG. 1. Transverse hadron profile resolved in scattering with 2 fixed squared momentum transfer Q and increasing center- C. Effective Field Theory for high parton densities: of-mass energy √s. The requirement for proliferating soft the Color Glass Condensate gluons to have maximal occupancy 1/αS generates the close packing saturation scale QS . Adapted from [110]. Since the usual formalism of pQCD relies on two-body and higher twist distributions being small, an alterna- The mathematical basis of the parton model in QCD tive framework is necessary to understand the physics follows from the observation that if one picks a lightcone3 of gluon saturation and the emergence of the saturation gauge A+ = 0 and quantizes the quark and gluon fields scale in the nuclear wavefunction at high energies. Fortu- of QCD along a light front surface (say, x+ = 0), the itously, the problem of high parton densities can be for- Hamiltonians of the free quark and gluon fields share mulated as a classical effective field theory on the light the same vacuum4 as the fully interacting theory [108]. front, which as noted greatly simplifies the problem of This allows one to construct the hadron wavefunction as heavy-ion collisions at high energies. a linear combination of a complete set of multi-parton To understand this better, we will outline here an ex- eigenstates, each of which is an eigenstate of the free plicit construction performed for nuclei with large atomic QCD Hamiltonian. number A 1 [111–113]. An important ingredient in In this lightcone framework, the parton distribution this construction in the infinite momentum frame (IMF) functions measured in DIS experiments can be inter- P + of the nucleus is a Born-Oppenheimer separa- → ∞ preted as one-body states of quarks and gluons that carry tion in time scales between the Lorentz contracted large a lightcone momentum fraction x = k+/P +, where k+ is x (k+ P +) “valence” modes and the noted “wee fur” ∼ the parton’s lightcone momentum and P + is the lightcone of small x (k+ P +) gluons and “sea” quark-antiquark momentum of the hadron. As first argued in [36, 109], pairs. For partons of transverse momentum k , lightcone ⊥ two-body “higher twist” gluon distributions, in a light- lifetimes are given by 5 cone operator product expansion (OPE), grow as the + + 2 2 1 2k 2xP square [xGA(x, Q )] of the leading twist gluon distribu- τwee = = 2 2 2 k− k ≡ k tion. For a fixed Q , these two-body distributions be- ⊥ ⊥ come as large as the leading twist one-body distribution 2P + τ τ τ , (3) as x 0. valence ≈ k2 −→ wee valence Importantly,→ the net effect of such many-body contri- ⊥ butions6 is opposite that of the leading term, softening suggesting that the valence parton modes are static over the growth in the gluon distribution. When the gluon the times scales over which wee modes are probed. How- ever, one cannot integrate the valence sources completely phase space density is maximal, of the order of 1/αS, all n-body lightcone distributions contribute equally. This out of the theory because they are sources of color charge saturation of gluon distributions in a nucleus of radius for wee partons and must couple to these in a gauge in- R , corresponds to the generation of the saturation scale variant manner. A Note further that, since wee partons have large lightcone wavelengths (λ 1/k+ = 1/xP +), they can re- wee ∼ √ solve a lot of color charge provided that their transverse 3 ± 0 3 Lightcone coordinates are k = (√k ±k )/ 2 and lightcone fields wavelength is not too large. The inequality are defined as A± = (A0 ± Az)/ 2; we work here in the metric g = 1; g = −1, where i and j represent the two transverse 1 1 R m ±,∓ i,j λ λ A N , (4) coordinates. wee ∼ k+ ≡ xP + valence ≡ P + 4 In lightcone quantization, this argument requires a careful treatment of k+ = 0 vacuum modes [105]. For a perturbative treat- where on the r.h.s the Lorentz contraction factor is ment of lightcone wavefunctions, it may be sufficient to project + P /mN (with mN the nucleon mass), suggests that wee out such modes [106, 107]. 1/3 7 5 partons with x A− resolve partons all along the In OPE language, these higher twist contributions are suppressed by powers of 1/Q2. 6 These include the screening of bremsstrahlung gluons by real and virtual gluons, as well as the recombination of softer gluons into 7 −1 harder gluons. Wee partons with wavelength k⊥ ≤ ΛQCD ∼ 1 fm see no color 9

1/3 longitudinal extent 2 RA A in units of the inverse The first term in Eq. (8) is the Yang-Mills action in the nucleon mass. ∼ QCD Lagrangian given in Eq. (1). The dynamics of wee These charges will be random since they are confined gluons in the CGC is specified by this term. The sec- to different nucleons and do not know about each other. ond term8 denotes the coupling of the wee gluon fields A wee parton with momentum k resolves an area in the to the large x color charge densities ρ, which we have ar- transverse plane (∆x )2 1/k2 ⊥. The number of valence gued are static lightcone sources. Because the sources are partons that it interacts⊥ ∼ simultaneously⊥ with is eikonal sources along the lightcone, their gauge invariant coupling to the wee fields is described by the path Nvalence 2 k k(∆x )2 = (∆x ) , (5) ordered exponential along the lightcone time direction ⊥ 2 ⊥ R + ,a a ≡ πRA U , = exp ig dx A− T . Physically, U corre- −∞ ∞ P 1/3 sponds to the color rotation of the color sources in the which is proportional to A since Nvalence = 3A in background of the wee gluon fields. QCD. For a large nucleus with k 1, one can show The weight functional in the effective action (or the for N 2 that the most likely color charge representa- c ≥ Gaussian random color charges in Eq. (6), in what is now tion that the wee gluons couple to is a higher dimensional called the McLerran-Venugopalan (MV) model [111, 112, √ classical representation of the order of k [114]. 116], can equivalently be written as9 Thus, wee partons couple to ρ, the classical color Z a a charge per unit transverse area of large x sources. On 2 ρ (x )ρ (x ) W + [ρ] = exp d x ⊥ ⊥ . (9) average, since the charge distributions are random, the Λ 2 − ⊥ 2µA wee partons will couple to zero charge; however, fluctuations locally can be large. These conditions can be For each configuration of ρ’s in Eq. (7), the saddle represented as point of the effective action is given by the Yang-Mills (YM) equations a a b 2 ab (2) ρ (x ) = 0 , ρ (x )ρ (y ) = µA δ δ (x y ) , µν,a ν+ a h ⊥ i h ⊥ ⊥ i ⊥ − ⊥ DµF = δ δ(x−) ρ (x ) , (10) (6) ⊥ 2 2 2 g A where a = 1, ,Nc 1 and µA = 2πR2 is the color whose solution is the non-Abelian analog of the ··· − A charge squared per unit area. For a large nucleus (A Weizäcker-Williams (WW) fields in classical electro- 2 1/3 2 dynamics. The chromo-electromagnetic gluon field 1), µA A ΛQCD is a large scale. Since it is the ∝ 2 strengths are singular on the nuclear sheet of width largest scale in the problem, αS(µA) 1. This result is + ∆x− 2Rm /P and zero (pure gauge) outside. remarkable because it provides a concrete example sug- ∼ N gesting that QCD at small x is a weakly coupled EFT The gauge field solutions in lightcone gauge are given wherein systematic computations of its many-body prop- by A− = 0 and erties are feasible. k 1 k Acl = V (x−, x ) V †(x−, x ) , (11) We can now combine the previous kinematic and dy- ig ⊥ ∇ ⊥ namical arguments and write the generating functional where k = 1, 2 are the transverse coordinates and V = for the small x effective action as − hR x 1 i ( Λ+ R ) exp dz− 2 ρ˜(z−, x ) . This solution of the equa- Z R + iS + [A,ρ] j A ⊥ ⊥ [dA]δ(A )e Λ − · P −∞ ∇ [j] = [dρ] WΛ+ [ρ] + . tions of motion requires path ordering of the sources in R Λ + iS + [A,ρ] Z [dA]δ(A )e Λ x− [116, 119]. Further, theρ ˜ that appears in the solution µ (7) is the color charge density in Lorenz gauge ∂µA0 = 0, Here Λ+ denotes the longitudinal momentum scale that + 1 where one has the solution Acl0 = 2 ρ˜(x−, x ), Acl0 − = separates the static color sources from the dynamical ∇⊥ ⊥ Acl0 = 0. In fact, since the Jacobian of the transforma- gauge fields and the gauge invariant weight functional tion⊥ [dρ] [dρ˜] is simple [119], many-body distributions WΛ+ [ρ] describes the distribution of these sources at the → + in the lightcone gauge can be computed in terms of color scale Λ , with its path integral over ρ normalized to charges in Lorentz gauge, a natural choice from the anal- unity. ogy to WW fields [120]. The CGC effective action can be written in terms of As a simple example, the correlator of gauge fields in the sources ρ and the fields A as a large nucleus can be computed analytically in the MV Z 1 4 a µν,a model by averaging the solution in Eq. (11) with the S + [A, ρ] = d x F F Λ 4 µν weight functional W : Z i 2 Z + d x dx−δ(x−)Tr ρ U , [A−] . (8) AA = [dρ˜]Acl.[˜ρ]Acl.[˜ρ]WΛ+ [˜ρ] . (12) Nc ⊥ −∞ ∞ h i

charge at all since color is conﬁned (in nucleons!) on this scale. 8 This term can alternatively [115] be written as It is only wee partons with k⊥ ΛQCD that see color charges Tr [ρ log(U−∞,∞)]. from diﬀerent nucleons along the longitudinal direction. 9 Sub-leading terms were discussed in [117, 118]. 10

W takes place on much larger time scales. This two- stage averaging process clarifies how one reconciles gauge invariance with the presence of a colored condensate. Fast hadron fast partons (ρ) The CGC classical equations possess a “color memory” effect [122] corresponding to the large gauge trans- formation V of a quark after interacting with the gluon small-x gluons (A[ρ]) shockwave, generating a transverse momentum kick p Q to the quark that can be measured in DIS experi-⊥ ∼ Dipole S γ ∗ ments [123]. Remarkably, this is exactly analogous to the inertial displacement of detectors after the passage FIG. 2. DIS in the dipole picture. The virtual photon emit- of a gravitational shockwave [124]. This gravitational ted by the electron splits into a qq¯ dipole that scatters off memory is deeply related to asymptotic spacetime sym- dynamical small x gauge fields coupled to the static large x metries and soft theorems in gravity and may also hold lightcone sources. From [110]. useful lessons for QCD.11

Equation (12) can be further Fourier decomposed to ex- D. Renormalization group evolution in the CGC dN EFT tract the number distribution of wee gluons 2 and d k⊥ 10 expressed in terms of QS. Specifically, for the occupa- 3 2 (2π) dN QS We have discussed a classical EFT for large nuclei tion number φ = 2 2 2 one obtains φ 2 2(N 1) πR d k⊥ k c − ∝ ⊥ and Gaussian sources where the separation between fields for k QS, However, for k QS the distribu- ⊥ ⊥ (wee partons) and sources (valence sources) was picked tion is modified substantially from the WW distribution: randomly to be at the momentum scale Λ+. Physical ob- 1 φ log(QS/k ). This softened infrared distribution ∼ αS ⊥ servables such as the inclusive cross-section in Eq. (13) in the CGC EFT provides a simple explanation of gluon should not depend on Λ+. This invariance is the essence saturation. of the renormalization group and we will later sketch how We are now in a position to understand the term Color it is realized in the EFT; a detailed discussion can be Glass Condensate (CGC) [37, 110], which is used to de- found in [110]. scribe the ground state properties of a hadron or nucleus The important point to note is that real and virtual at very high energies. Color is obvious since the state is quantum fluctuations in the classical background field of composed primarily of a large number of gluons and sea the target, while apparently suppressed by (αS) are ac- quark-antiquark pairs. It is a glass because these small x + + O tually αS log(Λ /Λ0 ) (1) from the phase space inte- gluons and sea quarks are generated by random sources ∼ O + gration of these modes when Λ = Λ+e 1/αS (or, equiv- with lifetimes much longer than the characteristic time 0 − alently, when x = x e 1/αS ). These large NLO scales of the scattering. This explains the structure of wee val. − contributions can be absorbed into the form of the LO Eq. (7), where the path integral over the curly brackets + cross-section in Eq. (13) at the scale Λ by redefining the is performed first for fixed color charge distributions and 0 weight functional WΛ+ [˜ρ] WΛ0+ [˜ρ0]. Hereρ ˜0 =ρ ˜+δρ˜ is then averaged over an ensemble of such distributions. Fi- → + nally, the state is a condensate because gluons have oc- the new color source density at Λ0 that incorporates the color charge density δρ˜ induced by quantum fluctuations cupation numbers φ 1/αS, with momenta peaked at + + ∼ between Λ and Λ0 . k QS. ⊥To∼ take a specific example, consider the inclusive cross- One can thus write section in the DIS scattering of a virtual photon on the Z dσ = [ ρ˜ ] W 0+ [˜ρ ] dˆσ [˜ρ ] , (14) nucleus, illustrated in Fig.2. In the CGC EFT, it is h LO+NLOi D A Λ A LO A expressed as the cross-section for a fixed distribution of sources convoluted with an ensemble of such sources: where Z h + + i WΛ0+ [˜ρA] = 1 + log(Λ /Λ0 ) LLx WΛ+[˜ρA] , (15) dσ = [ ρÃ] W + [˜ρA] dˆσ[˜ρA] . (13) H h i D Λ with the quantum fluctuations absorbed as we discuss Thus, on the time scale t 1/Q of the probe, it resolves a shortly. condensate of gluons with∼ a well-defined number density of longitudinal modes down to x x 1. Because ∼ Bj of time dilation [see Eq. (3)] the averaging over ρA with 11 An “infrared triangle” between asymptotic symmetries, memory and soft theorems in gravity [125] also allows for an elegant interpretation of the infrared structure of QED [126, 127]. While color confinement implies that such universal features do not ap- 10 2 2 In the MV model, this defines QS = cAµA, where the coefficient ply to QCD in general, an emergent QS ΛQCD suggests that cA is determined numerically [121]. they may be applicable in the Regge limit. 11

Since the l.h.s of Eq. (14) should not depend on the of the valence modes. The RG evolution also predicts arbitrary “factorization scale” Λ+, the derivative of that the width of the wee gluon cloud will shrink with both l.h.s and r.h.s with respect to it should be zero. increasing boost (or rapidity) relative to an “observer” From Eq. (15), one can therefore deduce the Jalilian- quark-antiquark pair, albeit at a slower rate than their Marian–Iancu–McLerran–Weigert–Leonidov–Kovner larger x counterparts. Thus, in the CGC EFT the scale (JIMWLK) RG equation [39, 40, 128] for the overlap of the wave functions in the thermalization process is set by 1 rather than the Lorentz contracted QS ∂ 1 W [˜ρ ] = W [˜ρ ] , (16) width of the valence quarks given by + . ∂Y Y A HLLx Y A P where the JIMWLK Hamiltonian [129] E. DIS and the dipole model Z 1 δ ab δ LLx = χ (x , y )[˜ρ] , (17) a ⊥ ⊥ b H 2 x⊥,y⊥ δρ˜ (x ) δρ˜ (y ) Here and in SectionIIF we will concretely relate the ⊥ ⊥ CGC EFT to the structure functions that are measured describes the evolution of the gauge invariant weight in DIS. These comparisons are essential for precision tests + functional W with rapidity Y = log(Λ /Λ+) of the CGC EFT picture of high energy nuclear wavefunc- 0 ≡ log(x0/x) once the non-perturbative initial conditions for tions. They also play an important role in constraining W are specified at an initial x0. the saturation scale and the shadowing of nuclear distri- The Hamiltonian is computed in the CGC EFT, with butions that are key to determining the initial conditions ab a b χ (x , y )[˜ρ] = δρ˜ (x )δρ˜ (y ) ρ˜ the two-point func- for early time dynamics in heavy-ion collisions. These ⊥ ⊥ h ⊥ ⊥ i tion of induced charge densities12 in the classical back- connections will become more evident in SectionIIIC. ground field of the hadron. Note that with this compu- The inclusive cross-section can be expressed in full gen- µν tation of LLx, the solution of Eq. (16) resums leading erality as dσ = L W , where L is the well-known H h i µν µν logarithms αS log(x0/x) (LLx) to all orders in perturba- lepton tensor [134] representing the squared amplitude tive theory. Thus this powerful RG procedure extends for the emission of a virtual photon with four-momentum the accuracy of computations of the cross-section from qµ and W µν is the spin-averaged DIS hadron tensor that, dσLO+NLO dσLO+LLx . for a nucleus in the IMF, can be reexpressed as [135, 136] h The JIMWLKi → h RG equationi can equivalently be ex- + Z Z pressed as a hierarchy of equations (the Balitsky- µν 1 P 2 4 iq x W = Im d X dX− d x e · JIMWLK hierarchy independently derived in [130]) for 2π mN ⊥ × the expectation value of an operator O: µ x x ν x x Tr γ S (X− + ,X− )γ S (X− ,X− + ) , h A 2 − 2 A − 2 2 i * Z + ∂ O Y 1 δ ab δ (19) h i = a χ (x , y ) b O[α] , dY 2 x ,y δα (x ) ⊥ ⊥ δα (y ) ⊥ ⊥ ⊥ ⊥ Y where SA(x, y) = i ψ(x)ψ¯(y) A is the quark propagator (18) − hµ i 13 a 1 a in the gauge fields A of the nucleus. where α = 2 ρ˜ . Remarkably, Eq. (18) has the form of ∇⊥ In the CGC, the leading contribution is obtained by a generalized Fokker-Planck equation in functional space, replacing the full QCD background field with the sat- where Y is “time” and χ is the diffusion coefficient [129]. µ µ urated classical background field: A Acl, where There is no known analytical solution to the JIMWLK µ → Acl are the non-Abelian WW fields in Eq. (11). In equation; as we later discuss, it can be solved numerically. 14 A− = 0 gauge , the momentum space quark propaga- However, good approximations exist in different limits. tor in the classical background field is remarkably sim- In a “weak field” (and leading twist) limit gα 1, one ple, given by [135] SAcl (p, q) = S0(p) q(p, q)S0(q), where recovers for the number distribution (and the correspond- Tip/ the free Dirac propagator is S = 2 and (q, p) = ing occupation number φ) extracted from Eq. (12), the 0 p +iε Tq R i(q⊥ p⊥) z⊥ 1 celebrated LLx Balitsky-Fadin-Kuraev-Lipatov (BFKL) (2π)δ(p− q−)γ− e− − · V ± (z ) is the ef- ± − z⊥ ⊥ equation [131, 132] of pQCD. Another mean field “ran- fective vertex corresponding to the multiple scattering of dom phase” approximation [129, 133] allows one to eval- the quark (or antiquark) off the shockwave background uate the occupation number φ in the “strong field” limit field represented by the eikonal path ordered phase V of gα 1. introduced after Eq. (11). The∼ longitudinal extent of the wee gluon cloud gener- 1 1 ated by the RG evolution has a width x− = + . k ∼ QS 1/αS 1 This is much more diffuse relative to the width e− QS 13 The second average in Eq. (19) corresponds to averaging overρ ˜. We employ the relativistic normalization hP |P i = P + 3 3 P + R 2 − − (2π) δ (0) ≡ d X⊥dX , where X and X⊥ are mN mN 12 R R 2 center-of-mass coordinates. Note that here and henceforth in this section, = d x⊥ and 14 x⊥ The solution of the YM equations is identical in this case to the R R 2 2 = d x⊥d y⊥. x⊥,y⊥ solution in Lorenz gauge. 12

The DIS structure function is simply related to the the dipole Wilson line correlator but also the quadrupole inclusive cross-section. Plugging the dressed CGC prop- and higher point correlators that appear in semi-inclusive agator into Eq. (19), one can show, to this order of accu- ﬁnal states in e+A and p+A collisions. racy, that it can be expressed as [135]

F. RG evolution and geometric scaling 2 Z 1 Z 2 Q 2 2 F2(x, Q ) = dz Ψγ∗ qq¯ σqqA¯ (x, Q ) . 4π2α → em 0 r⊥| | The MV model is valid for a large nucleus at rapidi- (20) ties when the bremsstrahlung of soft gluons is not sig- Equation (20) can be simply interpreted to be the con- nificant, namely, for αSY 1. The classical expressions volution of the probability of the virtual photon to split we derived have no x dependence.≤ For moderate x, one into a quark-antiquark pair (which can be computed in can introduce x dependence in framework along the lines QED [137]) with the “dipole” scattering cross-section of of the IP-Sat model that we discussed. However when the quark-antiquark pair to scatter off the nucleus. For αSY 1, the model is no longer applicable. In this the impact parameter b = (x +y )/2, the cross-section ⊥ ⊥ ⊥ regime, the RG evolution of the S-matrix in Eq. (22) is is given by described by the Balitsky-JIMWLK hierarchy in Eq. (18) Z which, in addition to the coherent multiple scattering ef- 2 σqqA¯ = 2 d b Y (b , r ) , (21) fects in the MV model, captures the previously discussed ⊥N ⊥ ⊥ real and virtual quantum corrections. where the forward scattering amplitude Y (b , r ) = Substituting the expectation value of the correlator of N ⊥ ⊥ 1 Y (b , r ), with the S-matrix the Wilson lines in Eq. (22) into the Balitsky-JIMWLK ⊥ ⊥ − S hierarchy in Eq. (18), leads, for Nc,A 1, to the closed 1 form Balitsky-Kovchegov (BK) [116, 130] equation for Y (r ) = Tr V (x )V †(y ) Y . (22) S ⊥ Nc h ⊥ ⊥ i the RG evolution in the rapidity of the dipole scattering amplitude: One can compute the S-matrix explicitly in the MV Z 2 model, which gives[116, 135, 136], ∂ Y (x , y ) (x y ) N ⊥ ⊥ =α ¯ ⊥ − ⊥ ∂Y S (x z )2(z y )2 2 2 2 z⊥ π r AxGN (x, 1/r ) h ⊥ − ⊥ ⊥ − ⊥ Y (r ) = exp αS ⊥ ⊥ , (23) S ⊥ − 2N πR2 Y (x , z ) + Y (y , z ) Y (x , y ) c A × N ⊥ ⊥ N ⊥ ⊥ − N ⊥ ⊥ i where GN denotes the gluon distribution in the proton Y (x , z ) Y (z , y ) . (24) 1 −N ⊥ ⊥ N ⊥ ⊥ at the scale 2 . r⊥ One can expand the exponential for very small values The BK equation is the simplest RG equation to capture the physics of gluon saturation. For 1, the of r , and one observes that the dipole cross-section is NY nearly⊥ transparent to the color of the small dipoles. As non-linear term in the last line of Eq. (24) can be ignored r grows, the S-matrix decreases; the saturation scale is and the equation reduces to the linear BFKL equation as defined⊥ as the value of r at which the S-matrix has a anticipated previously. In this limit, the amplitude has value that is significantly⊥ smaller than what one would the solution anticipate in pQCD. While there is some freedom in set- ρ ρ2 Y (r ) exp ωα¯sY , (25) ting this scale, its growth with decreasing x is determined N ⊥ ≈ − 2 − 2βα¯SY by the growth in the gluon distribution. where ω = 4 log 2 2.77, β = 28 ζ(3) 33.67 and The MV result in Eq. (23) is the QCD Glauber 2 2 ≈ ≈ ρ = log(1/r ΛQCD). This solution gives the rapid model [138] which gives the survival probability of a ⊥ dipole after multiple independent scatterings off the nu- “Markovian” growth of the dipole cross-section in rapid- cleus. It can be refined by introducing an impact pa- ity due to the copious production of softer and softer rameter distribution inside the proton [139], the so-called gluons. However, when 1 the non-linear term arising impact-parameter-dependent saturation (IP-Sat) model, NY ∼ which can be further extended to model the S-matrix for from the fusion and screening of soft gluons completely the nuclei [140, 141]. saturates the growth of the dipole cross-section. If we The IP-Sat model provides very good agreement with impose a saturation condition Y = 1/2 for r = 2/QS on Eq. (25), the argument of theN exponential vanishes⊥ for a wide range of small x DIS data on e+p scattering at 2 2 HERA [142]. The latter constrains the parameters of ρs = log(QS/ΛQCD), with this model, which in turn is an essential ingredient of Q2 = Λ2 ecα¯S Y where c = 4.88 . (26) the IP-Glasma model of the initial conditions for heavy- S QCD ion collisions. We will discuss the IP-Glasma model in Further, if we write ρ = ρS + δρ, where δρ = Sec. IIIC3. 2 2 log(1/r QS), we find that [148] An advantage of the MV model formulation is that ⊥ 2 2 γs one can compute with relative ease [143–147] not only Y r QS , (27) N ≈ ⊥ 13

corrections is clearly seen in Fig.3. 1.4 Initial Condition These give [155], Running Coupling 1.2 Fixed Coupling 2 2 hp i Qs,running αS = ΛQCD exp 2b0c(Y + Y0) , (28) 1.0

0.8 where b0 is the coefficient of the logarithm in the one loop φ QCD β-function15. The running coupling results are well 0.6 approximated by a power law increase of the amplitude 0.4 consistent with the HERA data. Further, qualitative fea- 0.2 tures of geometric scaling persist, albeit the window for geometrical scaling is significantly smaller [157]. 0.0 10-2 10-1 100 101 102 103 104 For a large nucleus at the saturation boundary Y0 2 1/3 1/3 ∝ k2 log (A ), one recovers the A scaling of the saturation scale in the MV model from Eq. (28) for Y0 Y .A striking result for Y Y is that the saturation scale FIG. 3. Unintegrated gluon distribution (in units of the trans- 0 verse area) vs the squared transverse momentum (normalized for a fixed impact parameter becomes independent of A. to its value at the peak of the initial condition curve) from In the asymptotic Regge limit, strongly correlated gluons the solution of the Balitsky-Kovchegov equation. The differ- in the nuclear wavefunctions lose memory of the initial ent curves represent increasing rapidities (left to right) for conditions whereby they were generated. fixed and running coupling. From [151].

G. The state of the art in the CGC EFT 2 4 2 for Q < QS/ΛQCD, where γs = 0.63 is the BK anomalous dimension. In this section, we have outlined a description of the This “geometrical scaling” of the forward scattering wavefunction of a high energy nucleus in the CGC EFT, 2 2 emphasizing a qualitative understanding of gluon satura- amplitude means that Eq. (21) scales with Q /QS(x) alone instead of with x and Q2 separately. Remarkably, tion and key related analytical results. There have been this phenomenon was observed at HERA, providing a significant developments since in the CGC EFT. strong hint for the saturation picture [149]. Moreover, On the formal side, the Balitsky-JIMWLK framework 2 4 2 for the LLx evolution of n-point Wilson line correlators the wider scaling window Q < QS/ΛQCD stretching beyond QS provides a first principles explanation for a so- has been extended to NLLx [158–162]. For the two-point called “leading twist shadowing” of nuclear parton distri- dipole correlator, which satisfies the LLx BK equation, butions relative to those in the proton [150]. Such shad- the formalism has been extended to NLLx [163] and for owed parton distributions are used to compute the rates = 4 supersymmetric Yang-Mills even to NNLLx in N of hard processes in heavy-ion collisions; understanding a recent tour de force computation [164]. The BFKL or their microscopic origins is therefore important for quan- BK kernel, however, receives large collinear contributions tifying hard probes of thermalization. that need to be resummed in so-called small x resumma- The BK equation in a reaction-diffusion approxi- tion schemes for quantitative predictions [165–168]. mation can be formally mapped into a well-known While as we have discussed there are good analytical equation in statistical physics, the Fischer-Kolmogorov- approximations, a full analytical solution of the BK equa- Petrovsky-Piscounov (FKPP) equation [152]. In this tion does not exist. Numerical simulations have, however, context, geometrical scaling appears as a late-time so- been known for some time for the LLx BK equation [169], lution of a non-linear equation describing a traveling the LLx+running coupling BK equation [170, 171], and wavefront of constant velocity. In Fig.3, we show nu- even more recently the full NLLx equation implementing merical results for the unintegrated gluon distribution collinear resummation [172, 173]. In particular, it was 2 2 πNck⊥ R + 2 ik⊥ r⊥ 2 shown in [173] that this NLLx framework provides very φ(k ) = ∞ d r e · [1 Y (r )] , which 2αS 0 ⊥ ⊥ good agreement with the HERA data. displays⊥ this traveling wave front structure,− N with the evo- Numerical simulations of higher point correlators in lution of the peaks of the wavefronts representing the the Balitsky-JIMWLK hierarchy have also been per- evolution of Q2 with rapidity. The correspondence of S formed. As noted, Eq. (18) has the form of a func- high energy QCD to reaction-diffusion processes is very tional Fokker-Planck equation. This can therefore be rich; specific applications to DIS have been discussed re- reexpressed as a Langevin equation in the space of Wil- cently [153, 154]. 2 son lines [129, 175], allowing one to simulate the rapid- QS in Eq. (26) [and the amplitude in Eq. (27)] grows ity evolution of two-point Wilson line correlators [176] very rapidly with rapidity, much faster than in the HERA data. However this is significantly modified by running coupling corrections, which are part of the next- 15 to-leading logarithms in x (NLLx) contributions to QCD Sub-leading corrections in Y to QS have been computed to high evolution. The significant effect of these running coupling order [156]. 14

III. NON-EQUILIBRIUM QCD MATTER AT low energy high energy 1 1 1 HIGH OCCUPANCY 0.8 0.5 0.5 0.6 The CGC EFT provides us with powerful tools to 0.4 1/Q address multi-particle production in heavy-ion collisions 0 s 0 1/Qs 0.2 y[fm] 0 from first principles; the key organizing principle is the -0.5 -0.5 -0.2 kinematic separation in the hadron wavefunction between -0.4 static color sources at large x and small x gauge fields. -1 -1 -0.6 -1 -0.5 0 0.5 1 -1 -0.5 0 0.5 1 In the following, we will sketch the elements of the for- x[fm] x[fm] malism to follow the thermalization process through the overlap of two CGCs. To apply this EFT framework to thermalization, one FIG. 4. Solution of the JIMWLK equation for the correlator † first needs to understand how to compute from first prin- of Wilson lines V (x⊥)V (y⊥) probed by the DIS dipole [174]. ciples multi-particle production in the presence of strong As the nucleus is boosted from low energy (or rapidity) to fields17. The quark-gluon matter formed in this process is high energy, the regions with large values of these correlator the Glasma [32, 33], a nonequilibrium state with high oc- shrink spatially, corresponding to larger values of QS . cupancy [f (1/αS)]; this state decays and eventually thermalizes.∼ The O description of the temporal evolution of the Glasma can be classified systematically in weak as well as four-point quadrupole and sextupole16 correla- coupling into LO, NLO, etc. tors [174, 179, 180]. Figure4 shows a result for the dipole Following our discussion of multi-particle production, correlator from these simulations. Unfortunately, a sim- we will describe the temporal evolution of the Glasma at ilar Langevin representation is not known at present for LO. This corresponds to the solution of classical Yang- the NLLx JIMWLK Hamiltonian. Mills equations with CGC initial conditions for the fields Precision computations require not just higher order using both analytical approaches (valid for transverse computations of the JIMWLK kernel but higher or- momenta greater than the saturation scale) and a non- der computations of process dependent “impact factors” perturbative real time approach employing Hamilton’s analogous to pQCD computations of coefficient functions equation on the lattice. The LO solutions are indepen- that are convoluted, order-by-order, with the DGLAP dent of rapidity, with the dynamics of the corresponding splitting functions [181]. For inclusive DIS, analytical “flux tube” structures occuring entirely in the transverse expressions exist for the virtual photon impact factor plane of the collision. We will next discuss the IP-Glasma 2 Ψγ∗ qq¯ in Eq. (20)[182]. Recently, NLO impact fac- model of heavy-ion collisions, which combines the LO | → | tors were computed for DIS exclusive diffractive light classical solutions with constraints on QS from DIS ex- vector meson production [183] and DIS inclusive pho- periments on the proton and on nuclei. ton+dijet production [184, 185]. Numerical implemen- However the LO description of the Glasma is limited tation of these results remains a formidable task and an because the classical fields are unstable to NLO quantum essential component of precision studies of gluon satura- fluctuations that break boost invariance, growing expo- tion at the future Electron-Ion Collider (EIC) [186, 187]. nentially in the square root of the proper time. As we An outstanding problem at small x is the impact pa- later discuss, a careful treatment of such NLO modes rameter dependence of distributions. The BFKL kernel shows that the dominant contributions can be resummed at large impact parameters contributes a Coulomb tail and absorbed into a classical-statistical description of the 1/b2 ; the conformal symmetry of the kernel and ge- evolution. A key difference from the prior LO description ometric∼ ⊥ scaling suggest a particular dependence of the is that the resummed classical-statistical evolution is now saturation scale on the impact parameter [188]. The in 3+1 dimensions, involving both transverse and longi- Coulomb tail is however not regulated by saturation tudinal degrees of freedom. This distinction is of funda- and violates the Froissart bound on the asymptotic be- mental importance in the subsequent description of the havior of total cross-sections [189]. This is cured non- thermalization process in weak coupling. perturbatively only by the generation of a mass gap in In SectionIV, we will discuss how this classical- QCD. The Coulomb tail may be less of a problem in large statistical description fits into the general weak coupling classification of the evolution of quantum fields and shall nuclei with ΛQCDRA 1 because the contribution of the Coulomb tail may be suppressed relative to protons, for outline the power counting that delineates the applicability of this approximation and its subsequent matching which ΛQCDRA 1. ∼ to kinetic theory. We will also describe in SectionIV

17 A well-known example of such a formalism is e+e− pair pro- 16 These are probed in semi-inclusive DIS [145] and in proton- duction in strong electromagnetic ﬁelds [190]; another is that of nucleus collisions [146, 177, 178]. Hawking radiation from the Black Hole horizon [191]. 15

universal features of the Glasma that makes its study interesting in its own right.

A. Multi-particle production in strong ﬁelds

To compute multi-particle production systematically in the collision of the CGC gluon “shockwaves”, we C + will begin with the first principles Lehmann-Symanzik- - Zimmermann (LSZ) formalism in QFT. For simplicity, 3 we consider here a self-interacting φ scalar theory; our FIG. 5. Top panel: multi-particle production from cut discussion extends straightforwardly to the Yang-Mills “vacuum-vacuum” graphs connecting time dependent sources case. of the two nuclei after the collision. From [37]. Bottom In the LSZ formalism, the amplitude for n-particles in panel: the Schwinger-Keldysh closed time contour on which the “out” state generated from the “in-vacuum” can be the sources and fields are defined. expressed as Z " n result as [198] 1 Y 4 ipi xi p p 0 = d x e · h 1,out ··· n,out| ini Zn/2 i i=1 1 n Pn = exp (i [J+] i [J ]) J+=J−=J , (31) # n!D V − V − | 2 2 δ ∂xi + m exp (i ) . (29) × δJ(xi) V with

Z 2 2 ∂2 + m2 Here p1 pn denote the momenta of the produced par- 0 ∂xi + m yi δ δ ··· = ZG+ (x, y) . ticles and the “in-out” vacuum-amplitude 0out 0in = − Z Z δJ (x) δJ (y) h | i D x,y + exp(i ), where is the sum of all connected vacuum- (32)− V V 3 vacuum diagrams coupled to external sources. An illus- R 4 0 R d pi ip (x y) Here x = d x, G+ (x, y) = (2π)32E e · − tration of multi-particle production for the problem at − pi ≡ hand is shown in Fig.5. θ(p0)δ(3)(x y) and Z is the residue of the pole of the − In QFT computations, one usually sets J = 0 after the renormalized propagator. functional differentiation and 0out 0in is a pure phase. The action of the operator can be understood as fol- 2D 2 h2 | i (∂x +m ) When J is physical, 0out 0in = exp( 2 Im ) = 1. In lows. The “+” piece with i δ acts on a par- computing multi-particle|h | productioni| in− this context,V 6 it Z δJ+(x) ticular diagram in the connected sum of vacuum-vacuum is useful to employ18 the Schwinger-Keldysh (SK) QFT connected diagrams [J ] by removing a source J and formalism [196, 197]. One introduces + and vertices + + then amputating theV renormalized propagator to which it with opposite signs of the coupling in Feynman− diagrams, is attached. The same procedure is followed for the “ ” and likewise for the sources J . The corresponding “+” piece; the two amputated propagators are then sewn− to- and “ ” fields live on the upper± and lower segments gether by the renormalized “cut” propagator ZG0 . of a closed− time contour ranging forward in time from + Computing P in a theory with physical sources− is dif- t = on the upper contour and back to on the n ficult because one also has to compute the disconnected lower−∞ contour, as shown in Fig.5. Time ordered−∞ “++” vacuum-vacuum graphs for each n. However if we define (anti-time ordered “ ”) Green’s functions “live” on the a generating functional F (z) = P znP , Eq. (31) gives upper (lower) contour,−− and the mixed + “Wightman” n n functions connect the upper and lower contours.− F (z) = exp (z ) exp (i [J+] i [J ]) J+=J−=J , (33) Following the LSZ formalism, the probability to pro- D V − V − | duce n-identical particles is and successive differentiation of Eq. (33) with respect to n 3 z (and setting z = 1) generates the n-particle correla- 1 Y d pi P = p p 0 2 , (30) tors n(n 1)(n 2) . These moments do not re- n n! (2π)32E |h 1,out ··· n,out| ini| h − − · · · i i=1 pi quire one to compute the disconnected vacuum-vacuum graphs, since they also appear in the normalization of Pn 2 2 2 19 where Epi = pi + m . Plugging the expression for the and therefore cancel out in the moments. amplitude in Eq. (29) into the r.h.s, one can express the

19 Such cancellations are seen in the Abramovsky-Gribov-Kancheli 18 For other discussions of the SK formalism in the context of the (AGK) rules [199] that implement the combinatorics of CGC and the Glasma, see [192–194]. For a recent discussion in cut/uncut vacuum-to-vacuum graphs in Reggeon ﬁeld the- the context of thermal ﬁeld theory, see [195]. ory [200]. 16

This is illustrated by expressing the r.h.s of Eq. (33) where repeated indices are summed over. Note also that R for z = 1 as (x) µ[ρ1, ρ2](x) and m = 0 in ∆x,y. An integration Aby parts≡ A exp (i SK[J+,J ]) = exp ( ) exp (i [J+] i [J ]) , Z Z V − D V − V − (34) 4 ip x 2 3 ip x d x e · ∂x µ(x) = d x e · (∂0 iEp) µ(x) , where now i [J ,J ] represents the sum over all A x0 + − A SK + → ∞ vacuum–to–vacuumV connected− graphs that live on the SK (38) closed time contour. One can then express the inclusive shows that Eq. (37) can be computed by solving µ multiplicity as [198] the classical YM equations in Eq. (10) [with J = µ+ µ + 2 δ δ(x−)ρ1(x ) + δ −δ(x )ρ (x ) and µ(x) x0= = ⊥ ⊥ A | −∞ Z 0] to determine µ(x). N = ZG0 (x, y) [Γ (x)Γ (y) + Γ (x, y)] , A + + + J±=J For the following discussion, it will be convenient to h i − − − x,y introduce the (τ, η, x ) coordinate system, where the (35) ⊥ p 0 2 3 2 with the amputated one-point and two-point Green’s proper time τ = (x ) (x ) and the spacetime ra- 1 x0+x3 − 2 functions in the Schwinger-Keldysh formalism defined, pidity η = 2 log( x0 x3 ), and gµν = diag(1, τ , 1, 1). respectively, as A convenient gauge− to solve the YM equations− − in− the forward lightcone is the Fock-Schwinger gauge τ 2 + + 20 A ≡ R δi SK R R δ i SK x − + x− = 0. In this gauge , the solution Γ (x) = ∆x V ;Γ+ (x, y) = ∆x ∆y 2 V , A A ± δJ (x) − δ J+(x)J (y) to the YM equations are manifestly boost invariant: ± −(36) µ(τ, η, x ) µ(τ, x ) and one obtains [204–206], 2 2 A ⊥ ≡ A ⊥ R ∂x+m with ∆x = . ig Z i = Ai + Ai ; η = [Ai ,Ai ] , (39) In summing over all the nodes of all the trees connect- A 1,cl. 2,cl. A 2 1,cl. 2,cl. ing Γ+(x) to the sources, the time (anti-time) ordered i η + Feynman propagators in each tree on the upper (lower) with ∂τ = 0 and ∂τ = 0 at τ = 0 . This solu- A A SK contour are recursively converted to retarded propa- tion is obtained by matching the delta-functions on the gators: GR = G++ G+ G + G . This is equiv- lightcone wedges in Fig.6. − − ≡ − − −− alent to solving the classical equations of motion with Since the gauge fields are functionals of ρ1,2, the full retarded boundary conditions when J = J ! A further average inclusive multiplicity in the Glasma is obtained important result is that the renormalized± cut propagator by averaging over many nuclear collisions, each with its 21 Γ+ is obtained by solving the small fluctuation equa- distribution of color sources in the two nuclei . This can tions− of motion in the classical background, also as an be expressed as initial value problem with retarded boundary conditions. Z d N LO MV MV As previously discussed, the classical fields, and hh ii = [Dρ1][Dρ2]WYbeam Y [ρ1]WYbeam+Y [ρ2] dY d2p − sources thereof, of the colliding CGCs are static shock- ⊥ waves; as such, they do not spontaneously decay and are d N LO h i [ρ1, ρ2] , (40) thus part of the nuclear wavefunction. After the colli- × dY d2p ⊥ sion, the colored sources become time dependent. Thus, 2 µ µ where Ybeam = log(√s/mN ) is the beam rapidity and Γ in Eq. (35) corresponds to ∂x ,cl. where ,cl. is MV MV ± A± A± WYbeam Y (WYbeam+Y ) are the weight functionals in the the time dependent (1/g) Glasma field in the forward MV model− in Eq. (9) and at LO are independent of lightcone. The two-pointO function Γ (x, y) in Eq. (35) + Ybeam Y (Ybeam + Y ). is (1) and therefore NLO in the power− counting for the − O With the initial conditions in Eq. (39), the YM equa- inclusive multiplicity in the Glasma. The formalism can tions for τ = 0+ can be solved perturbatively to lowest be extended to higher orders in α . Its generalization to ρ1 ρ2 S non-trivial order in ( 2 2 ); in this “dilute-dilute” ap- higher multiplicity moments was developed in [201]. O ∇⊥ ∇⊥ proximation, one obtains the following for identical nuclei: 6 4 2 B. The LO Glasma: classical gluon fields from d N LO 2 g µA 2Nc(Nc 1) hh ii = πRA − (p , Λ) . (41) shockwave collisions dY d2p (2π)4 p4 L ⊥ ⊥ ⊥ This result, which agrees with the pQCD bremsstrahlung Since at LO in our power counting only the prod- 2 µ 2 ν formula first derived by Gunion and Bertsch [207] is valid uct Γ+(x)Γ (y) ∂x +∂x in Eq. (35) contributes, one obtains− for a≡ fixedA distributionA− of lightcone sources ρ ,1,2 = ρ1,2 (where 1, 2 denote the two nuclei) [202] ± 20 µ A perturbative solution was also found in Lorenz gauge ∂µA = Z d N LO 1 µ 0 [203]. R R ν 21 h 2i [ρ1, ρ2] = 3 ∆x ∆y ελελ µ(x) ν (y) , Owing to color confinement at distance scales 1/ΛQCD, one re- dY d p 16π x,y A A 1/Λ ⊥ quires R QCD d2x ρa = 0 for each such configuration. (37) 0 ⊥ 1,2 17

C. Non-perturbative evolution of high occupancy t + x− x ﬁelds (3) η = cst. A = ? µ 1. Real time evolution of boost invariant ﬁelds on the lattice τ = cst.

(1) (2) z While analytical results for the inclusive multiplicity Aµ = pure gauge 1 Aµ = pure gauge 2 are available only in limited kinematic regions, the YM equations for shockwave collisions can be solved numeri- ρ1,2 n (4) cally to all orders (( 2 ) )[213, 214] to obtain the full O ∇⊥ Aµ = 0 non-perturbative result for Eq. (40)[215–219]. Hamil- ton’s equations are solved in the Fock-Schwinger gauge τ = 0 with the initial conditions at τ = 0 speci- Afied by Eq. (39). To preserve gauge invariance, lattice FIG. 6. Spacetime diagram of gauge field configurations. Be- gauge theory techniques can be adapted to this prob- fore the collision, the gauge fields are pure gauge solutions with zero field strength. (In the text, the pure gauge solution lem. The boost invariance of the LO shockwave gauge i fields provides a significant simplification whereby the of the right moving nucleus is denoted by A1,cl. and that of i (3+1)-dimensional [(3+1)-D] Kogut-Susskind QCD lat- the left moving nucleus is denoted by A2,cl..) After the collision, the gauge field solution ( i,η in the text) correspond to tice Hamiltonian [220] can be “dimensionally reduced” finite field strengths in the Glasma.A From [32]. to the (2+1)-D form [214]

2 X g a i i 2τ aH = tr E E + 2 (Nc Re tr U1,2) x τ g a − 2 for p g µA and (p , Λ) is a logarithmically diver- # ⊥ L ⊥ τ a X 2 gent function, screened at Λ ΛQCD. + tr π2 + tr Φ Φ˜ . (42) ≈ a τ − i From our dipole model discussion [see Eq. (23) and i 2 2 the related discussion], QS GA(x, p ), where p is the ⊥ In Eq. (42) the trace refers to SU(2) color and the sum momentum conjugate to the∝ dipole size.⊥ This suggests is over all discretized cells with lattice spacing a in the that Eq. (41) (employing Q2 µ2 , as noted in foot- S ∝ A transverse plane. For clarity, we have omitted the cell note 10) can be generalized to a “k factorization” form i ⊥ d N LO R 2 2 2 index j for all quantities in Eq. (42). Further, the E with dYhh d2iip αS dk φA(x1, k )φB(x2, (k p ) ). Here ⊥ ∝ ⊥ ⊥ ⊥ − ⊥ i 1, 2 are the components of the transverse electric φ (x,k2 ) ∈ { } A,B ⊥ field living on each site; discretizing the initial conditions k2 is the Fourier transform of the dipole scat- ⊥ gives Ei = 0 at τ = 0. The spatial plaquette of link tering amplitude22 in each of the hadrons discussed in variables U i, Sec.IIF. This k factorization formula [36, 210] is widely j used in phenomenological⊥ studies of hadron-hadron col- j 1 2 1 2 † † lisions. U1,2 = Uj Uj+ê1 Uj+ê2 Uj , (43) The dilute-dilute analytical approximation for shock- (where +e î indicates a shift from j by one lattice site in wave collisions can be generalized to compute the inclu- ρ1 the i = 1, 2 transverse direction) represents the squared sive multiplicity to lowest order ( 2 ) in one of the O ⊥ longitudinal magnetic fields in the Glasma. In Eq. (42), ρ2 n ∇ sources but to all orders (( 2 ) ) in the other. In this we represent Aη(τ, x ) as an adjoint scalar field Φ be- O ⊥ ⊥ “dilute-dense” case as well,∇ the inclusive gluon multi- cause, as a result of boost invariance, it transforms co- plicity can be expressed as a k -factorized convolution variantly under η-dependent gauge transformations: of the unintegrated gluon distributions⊥ in the projectile 2 2 ˜ j i i and target. It is valid for Q (x ) Q (x ), cor- Φ = U Φj+êi U † . (44) S,1 1 S,2 2 i j j responding to the forward (or backward) kinematic regions of the shockwave collision where the parton mo- Finally, π = Eη = Φ˙ /τ in Eq. (42) represents the longi- mentum fractions are x x . Alternately, it can be a tudinal electric field. 1 2 good approximation in proton-nucleus collisions, where Details pertaining to the numerical simulations of the Q2 A1/3Q2 [211, 212]. real time evolution of gauge fields were given in [214, S,A ∼ S,p 218]. In the early work, only uniform sheets of nuclei were considered with constant (x independent) values of QS. These were subsequently relaxed to consider finite nuclei [221, 222]; more realistic simulations with event- by-event simulations of RHIC and LHC collisions were 22 This distribution is distinct from the WW-distribution and co- later developed in the IP-Glasma model that we shall incides with it only for large k⊥ [208, 209]. discuss shortly [223]. 18

As anticipated, the numerical results reproduce the perturbative result in Eq. (41) at large k QS. How- ever, unlike in that expression, there is⊥ no logarithmic 4 factor (k , ΛQCD). At momenta k < QS, the 1/k distributionL ⊥ is modified to a form that⊥ is well fit by⊥ a Bose-Einstein exponential distribution [219]. Even more remarkably, the non-linear dynamics generates a plasmon mass23 that screens the momentum distribution in the FIG. 7. Glasma flux tubes: Boost invariant LO Glasma con- infrared [216, 226]. The energy density is therefore well- figurations of transverse size 1/Q at τ = 0+ with paral- defined at all proper times without infrared or ultraviolet S lel Eη and Bη, corresponding to finite Chern-Simons charge. divergences [227]. Such configurations decay rapidly and are unstable to quantum fluctuations. From [230].

2. Glasma flux tubes At the earliest times after the collision τ = 0+, as noted, An interesting consequence of the LO Glasma solution only the longitudinal Eη and Bη = Fxy fields are non- is that the Weizäcker-Williams plane polarized E and B zero. Equation (47) then immediately gives T = and = . Thus, at the earliest times theP pressureE in fields in the colliding CGCs become purely longitudinal PL −E + the Glasma is purely transverse; after initial transverse immediately after the collision at τ = 0 ; Eη,Bη = 0 6 dynamics, the longitudinal pressure 0 from below and Ei,Bi = 0. It was pointed out in [228] that this PL → configuration satisfies the identity by τ 1/QS. Since the Glasma at LO is conformal, the energy∼ density satisfies = 2 at this time. E PT α Z Stringy models capture essential features of confining Q = S d4x Tr E B , (45) CS 2π η · η dynamics in QCD [231]. In high energy collisions, they have a long history and capture the bulk features of the αS R 3 0 spectrum of multi-particle production [232, 233]; they where the topological charge QCS = 16π d x K and Kµ is the Chern-Simons current. A neat interpreta- underlie event generators such as PYTHIA [234]. These tion [32, 229] of this result is that the YM equations at models however screen color at distance scales 1/ΛQCD τ = 0+ can be expressed as E = ρ and B = ρ , and carry only electric flux and no magnetic flux; par- ∇· el. ∇· mag. where ρel., ρmag. are respectively electric and magnetic ticle production is assumed to arise from the Schwinger charge densities24 on the gluon shockwaves after the col- mechanism [232]. It is remarkable nevertheless to observe lision. that similar stringy solutions emerge from the more fun- As sketched in Fig.7, the induced electric and mag- damental framework of classical YM equations. netic charges generate a “stringy” Glasma flux tube [230] Motivated by this stringy picture, we expect the num- of chromo-electromagnetic fields that is uniform in rapid- ber of gluons per unit rapidity to equal the number of 2 ity stretching between the fragmentation regions of the flux tubes [S /(1/QS)] times the gluon occupancy in ⊥ 2 3 nuclei and are color screened [222] on transverse distance a flux tube (2(Nc 1)/α¯S/(2π) ) multiplied by a non- perturbative coefficient− of (1). Extracting the num- scales 1/QS. O One≥ can straightforwardly compute the energy den- ber density from the correlator of gauge fields at τ 26 ∼ sities and pressures in the Glasma from the different 1/QS [216], one indeed finds that 25 components of the stress-energy tensor . We obtain 2 2 dNLO 2 (Nc 1) QSS = 2 T + L where ⊥ = cN −3 , (48) E P P dY (2π) α¯S 1 xx yy 2 T (T + T ) = Tr Fxy + Eη where S is the transverse area of the collision,α ¯S = P ≡ 2 ⊥ αSNc/π, and cN is a gluon liberation coefficient [235] 2 ηη 1 2 2 2 L τ T = Tr F + E Tr Fxy + E (47). estimated from the numerical simulations to be c = 1.1 P ≡ τ 2 ηi i − η N with 10% accuracy [121]. The YM simulations can also be extended to compute two particle correlations in the Glasma [236]: 23 This plasmon mass is parametrically larger than the confining d2N conn. κ dN dN scale; its properties were investigated recently using a number of LO = 2 LO LO , approaches [224, 225]. 2 2 2 2 2 2 dY1d p dY2d k (Nc 1)QSS dY1d p dY2d k 24 These induced charge densities are proportional to the commu- ⊥ ⊥ − ⊥ ⊥ ⊥ ij i j ij i j (49) tators δ [A1,cl.,A1,cl.] and [A1,cl.,A1,cl.] respectively. 25 Note that

µν µα νβ γδ 1 µν αγ βδ T = −g g g Fαγ Fβδ + g g g Fαβ Fγδ. (46) 4 26 Here and henceforth for simplicity the path integral over gauge ﬁelds (moot at LO) and over sources (hhii) is implicit. 19

27 2 2 where κ2 is a non-perturbative constant . Again the 4/r + Q0, with Q0 fixed by the HERA inclusive data. numerical simulations bear out the Glasma flux tube in- The⊥ gluon distribution xG(x, Q2) is parametrized at the 2 2 terpretation: the likelihood that two particles are cor- initial scale Q0 and then evolved up to the scale Q using related is suppressed by the number of flux tubes, and LO DGLAP-evolution. We define the nuclear saturation q 2 non-factorizable color connected graphs are suppressed scale QS = 1/ r ,s at r = r ,s for which the argument by (1/N 2). Perturbative arguments suggest that this ⊥ ⊥ ⊥ O c of the exponential in Eq. (50) equals one-half. To obtain picture can be extended to n-particle cumulants and that the spatial dependence of Q , one self-consistently solves the n-particle multiplicity distribution that generates S x = 0.5 QS(x , x)/√s for every x . these cumulants is a negative binomial distribution [237]. The result⊥ of this procedure is a⊥ lumpy distribution of For n-particle multiplicities, this expectation is confirmed 2 QS(x , x) denoting the sub-nucleon structure of the nu- by non-perturbative numerical simulations [238]. cleus.⊥ Since the IP-Sat model is a simple generalization of the MV model, one can extract the variance of the color 2 2 x 2 x 3. The IP-Glasma model charge density g µA( ) at each x from QS( , x)[121]. One then samples random⊥ color charges ρa⊥(x ) on a transverse lattice: ⊥ In the discussion thus far, color charge fluctuations on 2 2 the scale 1/QS provide the only structure in the collid- a b ab kl 2 g µA(x ) ρk(x )ρl (y ) = δ δ δ (x y ) ⊥ , (51) ing gluon shockwaves. However, nucleon distributions in h ⊥ ⊥ i ⊥ − ⊥ Ny nuclei are not uniformly smooth and can fluctuate from event to event. These fluctuations in nucleon positions where the indices k, l = 1, 2,...,Ny label the Ny points are extremely important for understanding key features of representing the width of the nucleus in x−. The path of the data such as the azimuthal moments vn of the flow ordered Wilson line in the dipole model S-matrix (see distributions at low momenta [239, 240]. Another im- (22)) is discretized as portant ingredient in the realistic modeling of heavy-ion collisions is the dependence of the saturation scale in the Ny A(B) ! Y ρk (x ) nuclei on x (or, equivalently, √s), which describes the VA(B)(x ) = exp ig ⊥ , (52) ⊥ − ∇2 m2 variations of particle multiplicites in energy and rapidity k=1 T − at RHIC and the LHC. We will outline here the IP-Glasma model [223, 238, where m is a infrared cut-off and A and B distinguish the 241, 242], and improvements thereof, which incorporates color charge distributions in the two colliding nuclei. The the fluctuations in the nucleon positions to construct corresponding dipole distributions in each of the incom- event-by-event lumpy color charge distributions and cor- ing nuclei for a particular configuration of color sources responding gluon field configurations in the LO Glasma is shown in the top panel of Fig.8. framework. As we will also discuss, the energy depen- To each lattice site j, one then assigns two SU(Nc) dence of these configurations at a given Y or √s is de- matrices V(A),j and V(B),j, each of which defines a pure i termined by the saturation scales in the two nuclei. gauge configuration with the link variables U(A,B),j = An essential input is the dipole cross-section of the pro- V V † , where +e ˆ indicates a shift from j (A,B),j (A,B),j+e î i ton. We consider here the IP-Sat model [139, 140] which, by one lattice site in the i = 1, 2 transverse direction. as discussed in Sec.IIE, is an impact parameter depen- i The link variables in the future lightcone Uj that are dent generalization of the MV model. As noted, high input into Eqs. (43) and (44) are determined [214] from precision combined data from the H1 and ZEUS collab- solutions of the lattice classical Yang-Mills equations at orations [243, 244] are used to constrain the parameters τ = 0, of the model and excellent fits are obtained [142]. The dipole cross-section for each nucleus at a given x n a h i i i tr t U + U (1 + U †) is constructed by taking the product of the S-matrices (A) (B) i i i io corresponding to the dipole cross-sections of overlapping (1 + U ) U † + U † = 0 , (53) nucleons at a given spatial location x . It can be ex- − (A) (B) ⊥ pressed as [141] a where t are the generators of SU(Nc) in the fundamen- A tal representation. (The cell index j is omitted here.) dσ π2 2 2 2 PA i 1 dip r⊥ αS (Q )xG(x,Q ) Tp(x⊥ xT ) 2 = 1 e− 2Nc i=1 − , The N 1 equations in Eq. (53) are highly non-linear 2 d2x − c − ⊥ (50) and are solved iteratively for Nc = 3. With these initial conditions, Hamilton’s equations corresponding to where Tp stands for the Gaussian thickness function for each of the A nucleons in each nucleus and Q2 = Eq. (42), are solved to compute inclusive quantities in the LO Glasma. The lower panel of Fig.8 shows the result for the energy density in the transverse plane at τ = 1/QS 27 The results have a weak dependence on the ratio m/QS , where The IP-Glasma model gives a good description of bulk m is an infrared lattice regulator. features of distributions at RHIC and the LHC [241, 20

shall discuss now, quantum ﬂuctuations that are parametrically (1) and that contribute to Γ+ in Eq. (35) play a largeO role both before (pη = 0 modes)− and after (pη = 0 modes) the collision28. We discussed the former previously6 in the context of the small x evolution of the hadron wavefunctions. We will discuss here the role of these modes after the collision. The pη = 0 modes appear only after the collision; as we shall subsequently6 discuss, they play a fundamental role in the thermalization of the Glasma.

1. Dynamics of pη = 0 modes: QCD factorization and energy evolution

At NLO [ (1) relative to the leading (1/α ) contri- O O S FIG. 8. Top panel: collisions of nuclei with sub-nucleon color bution] for the inclusive multiplicity in Eq. (35), one of charge fluctuations determined by the IP-Sat model. Bottom the two terms is the amputated small fluctuations prop- panel: the LO energy density in the Glasma at τ = 1/QS . agator Γ+ and the other is a one loop correction to Γ From [238]. (or equivalently,− the classical field). The pη = 0 modes lie± close to the beam rapidities Ybeam; before the collision, they can be visualized as the± fur of wee gluon modes ac- 242]. In particular, when matched with the MUSIC companying the valence partons moving along the light relativistic viscous hydrodynamic code [245], the IP- cone. Glasma+MUSIC model provides an excellent description After the collision, the valence partons are stripped of of the multiplicity distributions, the inclusive central- the small x wee gluon modes that then populate pη = 0. ity, and the p distributions, as well as the vn distri- η 6 ⊥ The surviving p = 0 modes are valence modes and the butions in heavy-ion collisions putting strong constraints quasi-static cloud of large x partons than accompany on the extracted transport coefficients of the quark-gluon them into the fragmentation region of the nuclear col- plasma [246, 247]. lision. Thus pη = 0 modes after the collision are likely There have been several developments since. First, not very interesting from the perspective of thermaliza- the model has been extended to include JIMWLK evo- tion at central rapidities. lution of the sources ρ(x ) ρ(x , x∓) for nuclei with Before the collision, all one has are the pη = 0 modes. ⊥ → ⊥ large P ±, enabling one to study rapidity correlations of These modes are further separated into sources and fields produced gluons [151, 248] and 3D evolution of the LO with the latter dynamically absorbed into the former Glasma fields [248–250]. Further, the extension of the IP- via the “small x” evolution of the weight functionals Glasma+MUSIC model to hadron-hadron and hadron- WYbeam Y[ρ1,2] of each of the comoving nuclei. This, nucleus collisions [251] indicates that sub-nucleon shape however,± requires a factorization of the quantum fluctu- fluctuations in the Glasma are essential to understand- ations of each of the two nuclei from each other. ing final state contributions to two and multi-particle The resulting factorized form of Eq. (40) can be cumulants of azimuthal anisotropies for high multiplic- proven to leading logarithmic accuracy in x [260, ity events in small systems [252], the so-called “ridge” 261]. An important ingredient in the proof is correlations [253]. the structure of the cut propagator G+ (u, v) Data on incoherent diffraction from HERA are sensi- k2 − ∝ 2 + ik+(u− v−)+i ⊥ (u+ v+) R d k⊥dk 2k+ tive to such non-perturbative “shape” fluctuations [254– k+ e − − . If the spacetime 256]; the framework developed here allows one to con- points u and v reside on one of the nuclei, say, moving + strain the latter with HERA data and in the future along x , then u− v− and one of the phases vanishes. ≈ + likely more precisely with the EIC. Numerical simula- The other phase oscillates rapidly when k 0, giv- ing a convergent contribution. However, for k→+ it tions suggest that long range two particle correlations → ∞ in the Glasma [257] when combined with hydrodynamic converges to unity, and one obtains a logarithmic diver- + + flow can explain the systematics of high multiplicity az- gence dk /k that is the source of the large logarithms imuthal moments in small systems [258, 259]. resummed in the small x evolution of the nucleus. In the case where quantum fluctuations in the two nuclei could “talk” to each other before the collision, the D. The Glasma at NLO

Thus far we have focused on the leading order dynam- 28 η ics of classical ﬁelds (1/g) in the Glasma. As we p is the Fourier conjugate of the spacetime rapidity η. A ≡ O 21

spacetime points u and v reside respectively on the light- be written as cones of the incoming nuclei corresponding to u± v± = − 6 Z 3 Z 2 0. The phases therefore oscillate rapidly when k± X d k 3 ip x 0 λ µ → ∞ d x e · (∂x iEq) µ aλ0k , 3 0 and there are no logarithmic divergences from such con- 0 (2π) 2Ek x − λ,λ →∞ tributions. The only possible region in which such fluc- (55) tuations may contribute is where the nuclei overlap. The µ + 1 1 where aλ0ak(x) is a small fluctuation field of (1) about area of this region is x x− = + − ; such con- µ Oµ a ik x P P s with the plane wave initial condition e 0 T e , tributions are therefore suppressed by the∼ squared c.m. λ · whereA T a are the SU(3) generators in the adjoint rep- energy. resentation30. Note that the previous structure is analo- Thus, the factorized form in Eq. (40) at LLx is satis- MV gous to that of Eq. (38) except that the classical field is fied to high accuracy, and one can replace W [ρ1,2] Ybeam Y replaced by the small fluctuation field. The latter obeys with W [ρ ], where the latter satisfies± the Ybeam Y 1,2 the small fluctuation equations of motion, and its solu- JIMWLK equation± in Eq. (16). This allows one to go tion can be expressed as beyond the boost invariant MV expression and treat the dynamical evolution in Y of the weight functionals in the Z µ 3 µ two nuclei. While our arguments suggest that the factor- a (x) = d u [a(y) Ty] (x) , (56) + · A ization theorem can be extended to NLLx, a formal proof τ=0 is lacking. where Ty is a linear operator that corresponds to a shift As Ybeam increases with increasing energy, the W ’s in of the initial data on the classical fields and their deriva- Eq. (40) describe the energy evolution of the inclusive tives [260, 262], multiplicity29. Running coupling corrections, which are δ δ part of the NLLx contributions, improve the accuracy of a(y) T = aµ(y) + (∂ν aµ(y)) , (57) the computations significantly. In the future, one may · y δ µ(y) δ(∂ν µ) anticipate using the NLLx JIMWLK Hamiltonian as a A A systematic improvement in describing energy evolution on the initial spacelike surface at τ = 0+. and rapidity correlations in heavy-ion collisions. The key insight provided by Eq. (56) is that, to com- Details of the factorization of the W ’s, and their en- pute the small fluctuation field at a spacetime point x in ergy evolution, are crucial to phenomenology because the forward lightcone, it is sufficient to know the small they dictate concretely the dependence of final state ob- fluctuation field at τ = 0+ rather than solve the small servables (such as the energy density and the correlators fluctuation equations on a time-dependent background. thereof) on the saturation scales in the wavefunctions of We will return to this point shortly. the colliding nuclei. Plugging Eq. (56) into Eq. (55) and then into Eq. (54), one obtains " # η dN Z Z 2. Dynamics of p = 0 modes: plasma instabilities and the NLO T T T 6 2 = [δ (y) y] + [Γ2(y, z) y z] classical-statistical approximation dY d p Σ A · Σ ,Σ · ⊥ y y z τ=0+ dNLO The pη = 0 modes are generated right after the colli- , (58) 6 × dY d2p sion when the sources become time dependent and pro- ⊥ R 3 duce gluon modes away from the rapidities of the beams. where Σy = d y denotes the initial spacelike surface At NLO, their contribution to the gluon spectrum for a τ = 0+ and fixed distribution of color sources can be written as [202] X Z d3k Z Γ2(y, z) = a+kλ(y)a kλ(z) , (59) dNNLO 1 4 4 ip (x y) 2 2 X λ λ 3 − = d x d y e · − ∂ ∂ (2π) 2Ek dY d2p 16π3 x y µ ν λ ⊥ λ µ ν µ ν is the small fluctuation propagator evaluated on this sur- [ (x)δ (y) + δ (x) (y) + G+ (x, y)] , (54) × A A A A − face31. λ where µ is a gluon polarization vector of helicity λ. The This NLO result, however, is not suppressed paramet- first two terms in Eq. (54) represent the NLO contribu- rically by (α ) relative to the LO result because the O S tion to Γ+(x)Γ (y) in Eq. (35), with δ the one-loop LO Glasma is very unstable to small fluctuations: correction to the− classical field [Aρ , ρ ] and the A ≡ A 1 2 last term representing Γ+ , which first appears at NLO. δ (x) √γinst.τ − Ty (x) A ge , (60) We first consider the cut propagator term G+ in A ∼ δ (y) ∼ Eq. (54). Its contribution to the NLO multiplicity− can A

30 For compactness, we will suppress color indices henceforth. 29 This LLx result is implicitly assumed in the 3+1-D IP-Glasma 31 Discussions of the computation of this propagator at τ = 0+ can simulations [248]. be found in [262–264]. 22

0.0001 tiplicity distribution at early times in the Glasma: 1e-05 2 c0+c1 Exp(0.427 Sqrt(g µ τ)) 2 Z 1e-06 c +c Exp(0.00544 g dN η 0 1 µ τ) hh ii = [Dρ1][Dρ2]WYbeam Y [ρ1]WYbeam+Y [ρ2]

L 2 3 1e-07 dY d p −

µ ⊥

4 Z 1e-08 dNLO / g [Da] F [a] 2 [ + a] . (62) ηη 1e-09 × dY d p A T

2 ⊥ τ 1e-10 This result of course applies to other inclusive quantities, max 1e-11 such as the components of the stress-energy tensor given 1e-12 in Eq. (46). 1e-13 In the classical-statistical approximation, the one loop 0 500 1000 1500 2000 2500 3000 3500 2 correction to the classical field (δ ) is suppressed at early g µ τ A times relative to the G+ term that we consider here. In general, the classical-statistical− approximation does not FIG. 9. Growth of the maximally unstable Fourier mode of 2 ηη 2 account for the full quantum evolution of the Glasma the longitudinal pressure PL = τ T . Note that since g µ 3 2 −5 ∝ fields. In SectionIV, we will discuss the dynamical power QS , the results are in units of Q /g , with g 10 and S ∼ Lη = 1.6. From [41]. counting of quantum fields within the framework of the two-particle irreducible (2PI) effective action that specifies the range of validity of the classical-statistical approximation and the nature of the corrections beyond, as where γinst., parametrically of the order of QS, denotes well as numerical results from the implementation of this the growth rate of the instability. This exponential approximation and the consequences thereof. growth of small fluctuations in Eq. (56) with √τ is clearly demonstrated in Fig.9 using (3+1)-D numerical simulations of the YM equations for an η-dependent fluctua- IV. FAR-FROM-EQUILIBRIUM GLUON AND tion a(η) on top of the boost invariant Glasma back- QUARK PRODUCTION: FROM PLASMA ground [41, 265]. The very small values of g in the plot32 INSTABILITIES TO NON-THERMAL are chosen to ensure that the classical-statistical approx- ATTRACTORS imation is satisfied in the numerical simulations. This point is discussed further in SectionIV. We have seen in the sectionIII that the overoccu- The existence of such instabilities was previously pre- pied Glasma is unstable with respect to small quan- dicted [266] and studied with the context of a finite tem- tum fluctuations that break longitudinal boost invari- perature hard thermal loop effective field theory [267, ance. As noted, the growth of fluctuations is caused 268]. They are understood to be analogous to the Weibel by primary (Weibel-like [18]) instabilities [41, 265, 271]. instabilities familiar in plasma physics [269]; for a recent However, there are also secondary instabilities that arise review, we refer the reader to [18]. due to the nonlinear interactions of unstable modes [272]. As a result of the instability, the exponentially growing The fluctuations that are initially small grow with time small fluctuations can become of the order of the LO clas- and an over-occupied plasma emerges on a time scale 1 2 1 2 1 sical field for τ log . In a so-called classical- QSτ log (αS− ). ∼ γinst. αS ∼ statistical approximation [270], these leading instabilities At this stage, details about the initial spectrum of can be resummed to all orders, modifying Eq. (58) as fluctuations are effectively lost as a consequence of the strongly nonlinear evolution. The apparent loss of information at such an early stage gives rise to decoher- dN Z dN resum = [Da]F [a] LO [ + a] , (61) ence toward a more isotropic equation of state in this dY d2p dY d2p A ⊥ ⊥ prethermalization regime [273–275]. Subsequently, a universal scaling behavior emerges far from equilibrium with h i R 1 increasing anisotropy [52], which is described in terms where F [a] exp Σ Σ a(y)Γ2− (y, z)a(z) . ∼ − y z of non-thermal attractor solutions [27, 28], representing To conclude our discussion of the classical-statistical the first stage of the “bottom-up” thermalization sce- approximation, as a final step we need to perform the nario [276, 277]. average of the color sources to obtain the inclusive mul- In the following, we will describe how this nonlinear behavior emerges, starting with the underlying quantum field theory, formulated as an initial value problem in time. Essential aspects of the far-from-equilibrium quan-

32 2 tum evolution can be approximated by a controlled weak- At RHIC (LHC) energies, g µ ∝ QS ∼ 1 − 2 GeV on the x-axis of Fig.9. With these values, τ 10 fm, the typical life time coupling expansion around the full (non-perturbative) of such a collision. However, for g ∼ 10−5, from QCD running, classical-statistical theory, which was first pointed out g2µ is larger than the Planck scale. The takeaway message from in the context of scalar field theories [270, 278, 279] and Fig.9 is the functional form of the fit and not the absolute values. then extended to include fermions [280–284]. 23

In strong field QCD, this corresponds to an expansion evolution equations for connected one and two-point cor- 2 in αS g /(4π), where the leading order contribution relation functions follow from the stationarity of the 2PI includes≡ the full classical-statistical theory of gluons de- effective action [290, 291] scribed in Sec.III. The next-to-leading order contributions take into account the back-reaction of the quarks i 1 i 1 Γ[ ,G] = S[ ] + tr ln G− + tr G0− ( ) G onto the gluons and encode important quantum effects A A 2 2 A + Γ [ ,G] + const , (63) such as anomalies. The non-equilibrium time evolution 2 A of gluons with dynamical quarks was been studied nu- 1,µν 2 a b where iG− (x, y; ) δ S[ ]/δ (x)δ (y) is the merically on the lattice in Refs. [285–287]. 0;ab µ ν inverse propagator withA Lorentz≡ A indicesA µ, νAand color in- Such an expansion around the full classical-statistical 2 dices a, b = 1,...,Nc 1 for SU(Nc) gauge theories with field theory breaks down on the time scale QSτ classical action S[ ].− Here Γ [ ,G] contains all two- 3/2 ∼ A 2 A αS− [27, 276], where typical gluon occupancies become particle irreducible contributions, which leads to the self- of the order of unity. To continue further and capture the µν ab energy Πab (x, y) 2iδΓ2[ ,G]/δGµν (x, y). Higher n- late-time evolution toward local thermal equilibrium, one point correlation functions≡ A can be obtained from Γ[ ,G] employs a resummed perturbative description of quan- by functional differentiation with respect to theA fields tum field theory in an on-shell approximation. This also once the solutions for and G are known. underlies the effective kinetic theory that we will discuss 1. Macroscopic field,A spectral and statistical functions in Sec.V. The range of validity of both approximation schemes, The full quantum evolution equation for the macro- the expansion around the classical-statistical theory at scopic field is obtained from the stationarity of Γ[ ,G] early times and effective kinetic theory employed at with respect to variations in (x) and is given by A late times with their common overlap at intermedi- A 1 ate times [288, 289], can be efficiently discussed using δS[ ] i δG− ( ) δΓ [ ,G] = J µ(x) tr 0 G 2 . the two-particle irreducible (2PI) quantum effective ac- aA a a A aA δ µ(x) − − 2 δ µ(x) − δ µ(x) tion [290, 291] on the closed time path [292, 293]. A A A (64) For our discussion of the evolution equations for two- point functions, it is convenient to introduce spectral and A. Non-equilibrium time evolution equations from statistical components by the quantum effective action ab ab i ab 0 0 Gµν (x, y) Fµν (x, y) ρµν (x, y) sgn (x y )(65) Quantum evolution equations can be formulated in ≡ − 2 C − terms of expectation values of field operators, such as the where the spectral function ρ(x, y) is associated with the macroscopic field (x) and the connected two-point cor- A expectation value of the commutator of two fields and the relation function or propagator G(x, y) on the closed time statistical function F (x, y) by the anti-commutator for contour introduced in Sec.III. In practice, the space- 33 C bosons [293]. A similar decomposition can be done for time evolution of the one-point, two-point or higher-point the self-energy, Π(x, y) iΠ(0)(x)δ(x y)+Π(F )(x, y) correlation functions cannot be computed for the full iΠ(ρ)(x, y)sgn (x0 y0≡)/ −2, where Π(0)− describes a lo-− quantum theory without approximations. However, one cal contributionC to− the self-energy. With this notation, can formally write exact evolution equations, which pro- the equations for spectral and statistical two-point cor- vides an efficient starting point justifying the applicabil- relation functions, which follow from the stationarity of ity of systematic expansion schemes. Γ[ ,G] with respect to variations in G, can be written Writing for simplicity only the gauge field part, the asA [293]

Z x0 h 1,µγ (0),µγ i cb (ρ),µγ cb iG0−,ac (x; ) + Πac (x) ργν (x, y) = dz Πac (x, z)ργν (z, y) , A − y0 Z x0 Z y0 h 1,µγ (0),µγ i cb (ρ),µγ cb (F ),µγ cb iG0−,ac (x; ) + Πac (x) Fγν (x, y) = dz Πac (x, z)Fγν (z, y) + dz Πac (x, z)ργν (z, y) . (66) A − t0 t0

33 In terms of the Keldysh components of the propagator employed 0 0 G−−(x, y) = F (x, y) + iρ(x, y)sgn(x − y )/2, in Sec.III, this reads 0 0 G+−(x, y) = F (x, y) + iρ(x, y)/2, G++(x, y) = F (x, y) − iρ(x, y)sgn(x − y )/2, and G−+(x, y) = F (x, y) − iρ(x, y)/2. 24

R b R b 0 R 3 p In Eq. (66) we denote a dz a dz d z g(z) with statistical two-point correlation functions read given initial time t and g as≡ the determinant− of the met- 0 1,µγ cb iG− (x; ) ρ (x, y) = 0 , ric. The inverse propagator enters Eq. (66) as 0,ac A γν 1,µγ cb 1 1 iG0−,ac (x; ) Fγν (x, y) = 0 . (68) 1,µν ac γα µν cb iG− (x; ) = ( g)− 2 D ( )( g) 2 g g D ( ) A 0,ab A − γ A − α A 1 1 In Eq. (68) sub-leading contributions are suppressed by ac γν µα cb abc µν ( g)− 2 D ( )( g) 2 g g D ( ) gf ( ) 2 − − γ A − α A − Fc A at least a factor of g relative to the leading contribution. At this order the evolution of the Glasma background ab ab abc c with the covariant derivative Dµ ( ) = δ ∂µ gf µ fields decouples from that of the fluctuations. The evo- a a a Aabc b c − A and µν ( ) = ∂µ ν ∂ν µ + gf µ ν as the field lution of vacuum fluctuations of the initial state is taken strengthF tensor.A A − A A A into account by Eq. (68) to linear order in the fluctua- The non-zero spectral and statistical parts of the self- tions. This was an important assumption in the deriva- energy Π(ρ/F )(A, F, ρ) on the r.h.s and the space-time tion in Sec.IIID2 and was exploited in Ref. [262, 264] to local part Π(0)(F ) on the l.h.s of this coupled set of equa- obtain the spectrum of initial fluctuations right after the tions make the evolution equations nonlinear in the fluc- collision. These approximations are therefore valid only tuations. In general, they contain contributions from the for evolution times short enough that the fluctuations interaction vertices of QCD, where in addition to the have parametrically small values. standard three- and four-vertices there is a three-gluon In general, it is difficult to find suitable approximation vertex associated with the presence of a non-vanishing schemes for the 2PI effective action in gauge theories be- field expectation value. The explicit expressions for the yond the linear regime [296]. However, it provides a for- derivatives on the r.h.s of Eq. (64) and the self-energy mal justification of a resummed coupling expansion of the contributions entering Eq. (66) were given to three loop quantum field theory around the full classical-statistical order (g6) in Ref. [294], and the corresponding expres- solution; as we will soon discuss, this scheme can be im- sions in co-moving (τ, η) coordinates can be found in plemented numerically on a lattice to describe dynamics Ref. [295]. The inclusion of quark degrees of freedom that are far from equilibrium. follows along the same lines and can also be found in Furthermore, as we shall also later discuss, the dif- Ref. [294]. ferent dynamical stages of the Glasma undergoing a non- The non-equilibrium initial conditions for the coupled equilibrium instability at early times can be conveniently evolution equations (64) and (66) can be formulated in understood analytically from power counting in the 2PI (τ, η) coordinates (and Fock-Schwinger gauge τ = 0) for effective action beyond the linear regime [272]. Not least, the Glasma initial conditions discussed in Sec.A III. The the 2PI effective action approach allows for efficient on- gauge field expectation values in Eq. (39) correspond to shell approximations employing a gradient expansion; the Glasma background fields, while the spectral and sta- these lead to effective kinetic equations describing non- tistical two-point functions describe the fluctuations. At equilibrium evolution at later times [297]. We will discuss all times the former satisfy the equal-time commutation these equations and their numerical solutions in Sec.V. relations

ρab (x, y) = 0 , µν x0=y0 B. Nonlinear evolution of plasma instabilities ab ab gµν 0 ∂x ρµν (x, y) 0 0 = δ p δ(~x ~y) , x =y − g(x) − In Sec. IIID2, we demonstrated that the highly − ab anisotropic state of the Glasma is unstable with re- 0 0 ∂ ∂ ρ (x, y) 0 0 = 0 . (67) x y µν x =y spect to small quantum fluctuations. In the language of Eq. (68), these correspond to the quasi-exponential growth of the statistical function, [41, 265, 271, 298, 299] 2. Resummed evolution equations to leading order h i ab p 2 Fµν (τ, τ, xT , yT , ν) exp Γ(ν) g µτ , (69) To isolate the leading contributions one has to take ∼ into account the strong external currents J (1/g) in 2 ∼ O where we recall that g µ QS and Γ(ν) is a function the Glasma, which induce non-perturbatively large back- of the order of unity for characteristic∝ modes ν that are ground fields (1/g). In contrast, the statistical Fourier coefficients with respect to the relative rapidity34 fluctuations FAoriginate ∼ O from the vacuum and are there- Z fore initially (1). The spectral function ρ encodes the ab dν ab iν(ηx ηy ) O F (x, y) = F (x, y, ν)e − . (70) equal-time commutation relations and is therefore para- µν 2π µν metrically (1) at any time. ConsideringO only the leading contributions in a weak coupling expansion, the evolution equation (64) reduces to the classical Yang-Mills equation for the classical 34 Here ν is equivalent to the momentum pη in the (τ, η) coordinate Glasma field , and the equations for the spectral and system. A 25

1. Dynamical power counting 2. Classical-statistical ﬁeld theory limit

The behavior of the quantum evolution beyond the lin- The evolution of the Glasma to later times than τocc is ear regime is captured by a dynamical power counting non-perturbative. While there are different ways to ad- scheme [300–303]. Self-energy corrections are classified dress this in scalar quantum field theories, with an exam- according to powers of the coupling constant g, of the ple being large-N resummation techniques [304, 305], for background field , and of the statistical fluctuations F . gauge theories the most frequently employed approach is A Thus, a generic self-energy contribution is of the order the classical-statistical approximation. The latter can be of gnF m lρk and contains the suppression factor from understood starting with the full quantum 2PI effective A powers of the coupling constant (n), as well as the en- action by a set of well-defined approximations. hancement due to a parametrically large background field One first notes that a given propagator line of a dia- (l) and large fluctuations (m). The “weight” of the spec- gram may be associated with either the statistical (F ) or tral function (k) remains parametrically of order one at the spectral (ρ) correlation function. The set of diagrams all times as encoded in the equal-time commutation re- included in the classical-statistical approximation can be lations, see Eq. (67). identified as those corrections that contain the most pow- For the strong macroscopic fields 1/g in the ers of the statistical function relative to powers of the A ∼ Glasma, sizable self-energy corrections occur once fluc- spectral function for each type of diagram [270]. This (n l)/m tuations grow as large as F 1/g − for charac- corresponds to resumming the leading effects of the in- ∼ teristic modes. This yields a hierarchy of time scales, stability to all orders in the coupling constant [262, 264]. where diagrammatic contributions with smaller values of Therefore, in contrast to expansions at fixed loop- r = (n l)/m become important at earlier times (since orders, the classical-statistical approach provides a con- − g 1) than contributions with larger values of r. trolled approximation scheme that is particularly well The quasi-exponential growth stops when fluctuations suited to problems involving large statistical fluctuations. become (1/g2), where they saturate. At (1/g2) Specifically, for the large F (1/g2) values encoun- O O the fluctuations lead to sizable contributions from ev- tered at the end of the plasma∼ O instability regime, ne- ery given loop-order and the perturbative power-counting glecting powers of ρ (1) compared to those of F scheme breaks down. The corresponding time scale may represents a systematic∼ weak-coupling O approximation of be estimated from the one-loop correction a system that is strongly correlated because of the large fluctuations. While leading order in this expansion corresponds to , the full non-equilibrium classical-statistical field theory for the gauge fields, genuine quantum corrections for the which has r = 2 (n = 2, l = 0, m = 1). Using the quasi- dynamics arise. As we will soon discuss, the dynamical exponential growth behavior [Eq. (69)] the factor of g2 ∼ evolution of quarks and anti-quarks represent a class of from the vertex is compensated for by the propagator such genuine quantum corrections [287]. line F (1/g2) at time ∼ O We can conclude from this discussion that for the g 1 1 2 2 far-from-equilibrium overoccupied Glasma there is a τocc log g− , (71) ∼ QS well-controlled mapping of the weak-coupling quan- which denotes the characteristic time for the end of the tum dynamics for correlation functions onto a classical- instability regime. statistical field theory. The latter can be simulated nu- The earliest time for nonlinear amplification to set in merically on a lattice. In principle, starting with large can be inferred from the diagram with the lowest value field amplitudes, the mapping involves two steps: (I) The of r. For our problem, this is realized by the one-loop field is separated into a large coherent part and a small contribution with r = 1 (n = 2, l = 0, m = 2), fluctuation part in which one linearizes the field evolution equations. The set of linearized equations is given by Eq. (68). (II) Although small initially, the fluctuations grow because of plasma instabilities. Once the fluctuations become sizable, the time evolution of the linearized which already becomes sizable when F (1/g), where equations is stopped and the results are used as input for the two propagator lines compensate for∼ the O 2 powers of a subsequent classical-statistical simulation that is fully the coupling. Again using the quasi-exponential growth non-linear. behavior [Eq. (69)] of the primary unstable modes, we A virtue of the two-step procedure of mapping the orig- find that the time at which this (1/g) correction be- inal quantum theory to the classical description is that comes important relative to the O(1/g2) in Eq. (71) is it has a well-defined continuum limit, enabling one to re- O τocc/4 in the weak-coupling limit. This is followed cover the full physical results for certain quantities in the ∼by a series of higher-loop corrections, all leading to a weak-coupling limit [306]. In scalar field theories, this is fast broadening of the primary unstable range in rapid- well tested by comparisons to fully quantum calculations ity wave number ν [272]. using 2PI effective action techniques [270] and likewise !"#$%&'(%)*+,-$,)%&#%&'(% #.("/#00123(4%5*6

QE(*+,-E-*H(+*!"!#$%$&$'(#)*%+$&$*%,*AE@*E@,*-./#,01)&234#*5# #$ 6')334#5(3)&3'&$*%,*193H3*X,DQ*)*G,-5*E*6127#81'9#)*30($%:***************!!""# 26 (*#('2:1#%&$ ############('&&$)1% "%&$ *1AE@*-9:-***** *********C ****5

% #" % #"

&' ' %"" ")' %)'() $! ' ' %"'() $! ' $ ! ! ! ! ! & $% $% # )* : +,

% -

* + , )% 7 8 : 3( . 4 (" * +4 /# $ 00 + - 2+, 12 9 Plasma - 3( * 7 $ 4 8 , , 8 +1 01 i%, & 0 &'<$($&7 &1 1$ % ,& %

" 3 #9

7 - ) ; $0 6 < %9 "< 1# # $9 & %+ #& 7< *+? % >*8 #$ =0 <4 & (' %7 1# ! @0 1<,+:,-=9@,,==*,:(*A(+**+(:+,--5* =#>(',4'#$%,&'<$($&$1,#(1'+#&*#'%#*612?*))30$1+##@3'29#:(3*%#0(',4'#2 −1 FIG. 10. Time evolution of the gluon distribution at early times 0 . QS τ . log (αS ) from next-to-leading order CGC initial *%#'#&$41#,)'(1#%"'#" '()−6% $! ########8A121#&A1#,7,&14#$,#,&2*%:(7#$%&12')&$%:#'%+##"' conditions [264] at very weak coupling (αS 10! ).( From Ref. [298]. ,&$((#5'2#52*4#1@3$($<2$34∼ # ! ! 5& !"#$#%&!'()*'(+*%,&-&.$""/(01#%&20+*"%0)&304 when scalar fields are coupled to fermions [307]. The taken as an additive contribution to the strong back- mapping was first applied in cosmology in the context of ground gauge fields. While primary unstable modes at post-inflationary scalar preheating dynamics [278, 279]. non-zero rapidity wave number exhibit quasi-exponential The two-step procedure is in practice replaced by a amplification first, secondary instabilities with enhanced simplified description whereby one already starts with growth rates set in with a delay for higher momentum the fully non-linear classical-statistical description from modes due to the previously described nonlinear pro- the initial time in the strong-field regime. This can cesses. Subsequently the instability propagates toward be well controlled, for a given regularization with lat- higher momenta until saturation occurs and the system tice spacing a in the weak coupling limit, by ensur- exhibits a much slower dynamics [265, 272]. This behaving that vacuum fluctuations from modes with mo- ior is similar to that observed in non-expanding gauge menta near the cutoff 1/a do not dominate the dy- theories [302, 303] and cosmological models for scalar namics. Several studies∼ have investigated the range of field evolution [300]. validity of this simplified “one-step” mapping of the original quantum theory onto the classical-statistical description–see for instance Ref. [308]; the limitations of the classical-statistical approximation were studied in de- C. Non-thermal attractor tail in Ref. [309] for scalar field theories. Figure 10 provides snapshots of the time evolution of The plasma instabilities lead to a far-from-equilibrium the gluon distribution for an analytically computed initial 2 1 state at time QSτocc log (αS− ), which exhibits an over- spectrum of fluctuations given in Ref. [264], already em- occupied gluon distribution∼ whose characteristic proper- ploying the fully non-linear classical-statistical description from the initial time in the strong-field regime. The non-equilibrium evolution is computed numerically using -2 the Wilson formulation of lattice gauge theory in real 10

-4 ν=4 time [298]. In addition to gauge invariant quantities, -4 µ ] 10 ν=12 Coulomb type gauge fixed distribution functions can be 2 ν=43 ν extracted for comparison to effective descriptions such as -6 =71 10 ν=94 kinetic theory. The definition of the distribution func- ν > ( τ , ν ) [g =200 L

P -8 tion shown in Fig. 10 employs the two-point correlation L 10

function of the gauge ﬁeld following Ref. [27]. While the

α (Qτ) β γ f(τ, pT , pz) = fS (Qτ) pT , (Qτ) pz . (73) In Fig. 12, we showed the predictions for various ther- α S malization scenarios for the momentum anisotropy with This scaling behavior is characteristic of the phenomenon decreasing occupancy. These thermalization scenarios of wave turbulence and has been observed in a variety are based on estimates in effective kinetic theory and of systems that are far from equilibrium [50, 313]. As differ primarily in how infrared momentum modes are shown in the left panel of Fig. 13, the moments of the treated. Clearly, these differences lead to very differ- 28

4 1.2 This would generate an attractor with a ﬁxed anisotropy -2 -3 x 10 x 10 such that ΛL/ΛT remains constant in time.

: 1 )

τ m=0 The selection of the appropriate eﬀective kinetic the- ,

z 3 0.8 ory using lattice simulation data represents the state of

=Q,p m=2 T the art and is the basis for the thermalization discus- f(p

2 2 0.6

g sion in Sec.V. A justiﬁcation of the kinetic description m 0.4 solely based on perturbation theory in its range of valid- / Q) z 1 ( p ity raises important open questions on how to incorporate Moments of the spectrum 0.2 the eﬀects of infrared modes.

0 0 -0.75 -0.5 -0.25 0 0.25 0.5 0.75

Longitudinal momentum: pz / Q 3. Non-thermal attractors in scalar ﬁeld theories 16

) m=0

τ 8 , Non-thermal attractors in overoccupied weakly cou- z 12 pled ﬁeld theories were studied earlier in the context of =Q,p : T 6 m=2

f(p cosmological (p)reheating and thermalization after in- 2

g 8 ﬂation in the early Universe [26, 50, 314]. A large m 4

/ Q) class of inﬂationary models employ scalar ﬁeld theo- z (p

γ ries, where an initially coherent inﬂaton ﬁeld decays due 4 +m

Rescaled moments 2 α - to non-equilibrium instabilities. These may originate ) τ

(Q from tachyonic or spinodal dynamics or parametric res- 0 0 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 onance [300, 315, 316]. The instabilities lead to overoc- τ γ Rescaled longitudinal momentum: (Q ) pz / Q cupied excitations, whose transient dynamics can exhibit self-similar evolution. FIG. 13. Left panel: moments of the single particle distri- The dynamics is in general spatially isotropic on large bution function as a function of longitudinal momenta. The scales, in contrast to the longituinal expansion relevant longitudinal spectra are evaluated at transverse momentum to heavy-ion collisions. To compare the two, if we im- p Q. The different curves correspond to different times of pose the isotropic case of no expansion with overoccu- the' evolution: Qτ = 750, 1000, 1500, 2000, 3000 (from top pied initial conditions for gauge fields, the gluon distri- to bottom). Right panel: the rescaled moments of the distribution function in the self-similar regime obeys f(t, p) = bution function are found to collapse onto a single curve when 4/7 1/7 t− fS(t− p) in three spatial dimensions. This is char- plotted as a function of the rescaled longitudinal momentum acteristic of an energy cascade toward higher momentum variable. From Ref. [27]. scale due to weak wave turbulence [310, 317, 318]. In the fixed box case for a relativistic real scalar field theory in the self-similar regime, the distribution func- ent paths in the thermalization process. As the sys- φ (d+1)/(2l 1) φ 1/(2l 1) tion obeys f (t, p) = t− − fS (t− − p) for l- tem evolves with decreasing occupancy from the initial vertex scattering processes [50]. For quartic (l = 4) self- 1 f αS− , classical-statistical field theory simulations ac- interactions, the exponents are identical to the gauge the- ∼ curately capture the physics of the infrared regime. This ory with the same geometry. However, in the presence may be used to distinguish whether a particular thermal- of spontaneous symmetry breaking the non-zero field ex- ization scenario is indeed realized, especially since lattice pectation value leads to effective three-vertex scattering simulations and effective kinetic theory have an overlap- processes off the macroscopic field. These analytical esti- 1 ping regime of validity when 1 < f < αS− . mates were numerically verified using 2PI effective action The gray lines in Fig. 12 indicate the different ther- techniques in Refs. [319, 320] for an N-component scalar malization scenarios put forward in Refs. (BMSS) [276], field theory with quartic self-interactions. In classical- (BD) [277], (KM) [311] and (BGLMV) [312]. Unlike the statistical simulations, which construct the ensemble av- BMSS scenario, which is consistent with the lattice simu- erages from individual runs with a non-zero field value, lation results and is discussed in detail in Sec.V, the BD the observed scaling exponents are consistent with the es- scenario considers the possibility that plasma instabili- timates in the presence of an effective three-vertex [50]. ties lead to an overpopulation f 1/α of modes with In Ref. [52] longitudinally expanding N-component ∼ S p . mD. The coherent interaction of hard excitations scalar field theories are analyzed starting from over- |with| the soft sector then causes an additional momen- occupied initial conditions. In the vicinity of the non- tum broadening such that the longitudinal momenta of thermal attractor, scaling behavior very similar to that hard excitations fall at a slower rate. A possible variant in a non-Abelian gauge theory is observed. The univer- of the impact of plasma instabilities for the subsequent sal scaling exponents and shape of the scaling function quantum evolution also underlies the KM scenario. In agree well with those obtained for the early stage of the the BGLMV scenario, elastic scattering is argued to be bottom-up thermalization process for gauge theories for highly efficient in reducing the anisotropy of the system. not too late times. Universality for large field fluctuations (III)

 Important new universality classes in weak-coupling regimes

out of equilibrium 29 Example: transient small QCD coupling with expansion in heavy-ion collisions

5 4.5 suggestion . 114 . Lett 114 4 3.5

log( 〈 W(t,A) 〉 ) 3 F t Q=100Q=100 2.5 0

061601, Editors‘ 2 SU(2) SU(3) 1.5 tQ=200 Berges et. al, Phys. Rev tQ=400 (2015) (2015) 6, 1 tQ=600

scaling function: rescaled momentum: Wilson Loop: -1/C 0.5 tQ=800 0 0 20 40 60 80 100 120 140 160 180 200 FIG. 14. The normalized distribution for the scalar the- ζ Scaled Area: (t/t )- A Q2/N ory fφ as a function of the rescaled longitudinal momentum 0 c at diﬀerent times in the self-similar regime compared to the gauge theory f . From Ref. [52]. g FIG. 15. Self-similar behavior of the spatial Wilson loop as a function of the time-rescaled area t−ζ A with universal ∼ scaling exponent ζ for gauge groups SU(Nc) with Nc = 2 As an example, Fig. 14 shows results for the N = 4 (circles) and Nc = 3 (triangles) [324]. component scalar theory for intermediate transverse momentum p Q/2, where the normalized scaling dis- T ∼ tribution as a function of the rescaled longitudinal mo- using the power law form of the occupation number dis- mentum is given. All data curves at diﬀerent times in tribution extracted in the perturbative regime. While all the scaling regime collapse onto a single curve using the characteristic momentum scales are initially of the same scaling exponents α = 2/3, and γ = 1/3. This scal- order Q , this suggests that during the self-similar evo- − S ing curve is seen to be indistinguishable from the cor- lution a dynamical separation of these scales K(t) responding scaling curve for non-Abelian gauge theory, mD(t) Λ(t) occurs as time proceeds. which shares the same scaling exponents. The results provide a striking manifestation of universality far from equilibrium. 1. Non-equilibrium evolution of the spatial Wilson loop

A proper description of the non-perturbative low mo- D. Far-from-equilibrium separation of scales and mentum regime can be based on gauge-invariant quan- ultrasoft scale dynamics tities. This should take into account that the infrared excitations of non-Abelian gauge theories are ex- The weakly coupled QCD plasma exhibits a hierar- tended objects, which can be computed from Wilson chy of scales in thermal equilibrium at high temperature loops [224, 302, 322, 324, 325]. At the magnetic scale, T , with the separation of hard momenta T dominat- spatial Wilson loops capture the long-distance behavior ∼ ing the system’s energy density, soft (electric screening of gauge fields , which is defined as or Debye) momenta gT , and ultrasoft (magnetic) mo- A 2 2 ∼ 1 R menta g T for g = 4παS 1. A similar separation i g i(z,t) dzi ∼ W = Tr e− C A , (74) of scales exists far from equilibrium in the vicinity of Nc P the non-thermal attractor, where for comparison we will consider the spatially isotropic case without longitudinal where the index i labels spatial components [326]. Here denotes path ordering along a closed line , and the expansion. P C Starting from over-occupied initial conditions, in this trace is in the fundamental representation of SU(Nc). The behavior of the spatial Wilson loop for large ar- fixed-box case the gluon distribution function in the self- 2 4/7 1/7 eas A 1/Q enclosed by the line reflects the long- similar regime obeys f(t, p) = t− fs(t− p) in three S C spatial dimensions [310, 317, 318]. Accordingly, the time- distance or infrared properties of the strongly corre- dependent hard momentum scale dominating the energy lated system. Like the large-distance behavior of the 1/7 spatial Wilson loop in a high-temperature equilibrium density is given by Λ(t) t . The Debye scale mD(t) q ∼ ∼ plasma, the spatial Wilson loop exhibits an area law in R 3 1/7 g d p f(t, p)/p t− decreases with time [28, 318, ∼ the overoccupied regime of the non-equilibrium plasma, 321–323]. i.e., log W A [224, 302, 322]. At even lower scales, the dynamics becomes non- However,− h herei ∼ the area-law behavior occurs in the self- perturbative for momenta K(t) where the occupancy similar regime of the non-equilibrium evolution. This is reaches 1/αS, and the perturbative notion of a gluon demonstrated in Fig. 15, which shows the logarithm of distribution∼ function becomes problematic in this ultra- the Wilson loop as a function of the time-rescaled area ζ soft regime. As suggested in Ref. [318], the evolution t− A with universal scaling exponent ζ [324, 325]. Re- 2/7 ∼ of the ultrasoft scale may be estimated as K(t) t− sults for the SU(2) and SU(3) gauge groups are both ∼ 30

102 hard scale, classical-statistical lattice simulations results N=128, Qsa=1 Hard scale N=192, Q a=1/2 Debye scale are given for the Debye and the non-perturbative string 1 s 10 N=192, Qsa=1/3 String tension tension scale [322]. The results clearly demonstrate the Λ2 2 / Qs dynamical separation of scales as a function of time. 100 2 2 mD / Qs -1

Scales 10 2. Effective condensate dynamics -2 σ 2 10 / Qs The traced Wilson loop in Eq. (74) may be directly re- 10-3 lated to correlation functions of a gauge-invariant scalar 10 100 Time: Q t field [328–330]. In thermal equilibrium, this scalar s field serves as an order parameter for the confinement- deconfinement phase transition of the underlying gauge FIG. 16. Time evolution of the hard scale (Λ2), the elec- theory [331, 332]. In the self-similar scaling regime of the 2 tric screening scale (mD), and the spatial string tension (σ). non-thermal attractor, the dynamical evolution of the 2 Symbols for mD and σ denote lattice results for the different scalar order-parameter field modes toward the infrared lattice spacings employed, while the solid green line represents bears many similarities [325] to the dynamics of Bose con- a continuum extrapolation for the hard scale. The dotted densation in non-relativistic field theories far from equi- lines represent perturbative estimates for the hard and Debye librium [333–335]. Even quantitatively, the values for the scales, and a fit to the lattice results for the non-perturbative string tension scale. From Ref. [322]. infrared scaling exponents in the different theories agree well within errors [325]. The non-equilibrium infrared dynamics for scalars displayed. After we take into account the Casimir color starting from over-occupation has been studied in great 2 detail [26, 52, 320, 333–344]. The emergence of self- factors, normalizing the data points with CF = (Nc − similar scaling behavior is closely related to the existence 1)/(2Nc) discloses very similar behaviors for Nc = 2 of non-thermal fixed points [26, 345–347]. For scalar N- and Nc = 3 [324]. The scaling exponent ζ = 0.54 0.04 (stat.) 0.05 (sys.) agrees for both gauge groups to± component theories, the behavior can be approximately very good accuracy± [325]. This value of the scaling ex- described by a large-N effective kinetic theory to next- ponent for the ultra-soft scale √σ obtained from lattice to-leading order, which describes the perturbative higher simulations and the perturbatively motivated result for momentum regime as well as the non-perturbative in- the scaling of K(t)[318] are rather close, corroborating frared dynamics [341]. √σ K. Both relativistic and non-relativistic scalar theories The∼ positive value for ζ signals evolution toward larger can show the same infrared scaling and condensation length scales, with a growing characteristic area A(t) properties [334]. This is true even for the anisotropic dy- tζ . For large A/tζ one observes from Fig. 15 the gener-∼ namics of relativistic scalars with longitudinal expansion alized area-law behavior [324, 325] along the z-direction; the latter geometry is relevant in the context of heavy-ion collisions, and scalar and gauge log W A/tζ . (75) theories show very similar behaviors for higher momenta − h i ∼ in this case [313]. Because of the strong enhancement This implies a time-dependent string tension scale σ(t) = in the overoccupied infrared regime, the low momentum ζ ∂ log W /∂A t− . − h i ∼ modes exhibit essentially isotropic properties despite lon- In Ref. [322], this behavior is related to the rate of gitudinal expansion. topological transitions, the so-called sphaleron transition A remarkable development in this regard is that table- rate: top experiments with ultracold quantum gases have discovered universal transport processes toward the infrared Γ = C σ2 , (76) sphaleron starting from initial overoccupation of bosonic excita- where C is a number of the order of unity. The picture tions of trapped atoms [54, 55], which is similar to the that emerges is that the rate of topological transitions is case discussed here. This is discussed further in Sec.VIII. 4 large at early times (Γsphaleron QS) but subsequently decreases with time at a rate dictated∼ by the universal scaling exponent ζ. One expects this rate to converge E. Early-time fermion production and quantum from above to the thermal rate for sphaleron transitions anomalies in a high-temperature plasma [327]. We will return to the implications of these results for the evolution of anoma- In the high energy limit, strong gauge fields dominate lous currents in Sec. IVE. the earliest stages of the plasma’s space-time evolution. Figure 16 summarizes the behavior of the different However, the Bose enhancement from over-occupied glu- characteristic scales in the self-similar regime far from ons can lead to a rapid production of quarks with impor- equilibrium. Apart from the perturbative behavior of the tant phenomenological consequences for heavy-ion colli- 31 sions, such as direct photon production from the elec- A¯ A¯ A¯ trically charged quarks [348] or the breaking of classical symmetries due to anomalies, with a prominent exam- A¯ ple being the chiral magnetic effect [349, 350]. At early times these processes occur far from equilibrium and re- A˜ A˜ A¯ quire suitable techniques for their computation. We will now discuss these techniques and their consequences for FIG. 17. Illustration of rescaled “classical” three- and the production and evolution of fermions off-equilibrium. four-vertices, that are independent of the coupling. From Ref. [284].

1. Real-time simulations for fermions and gauge fields A˜ A˜ A¯ beyond the classical-statistical approximation g4 A˜ g4 Since identical fermions cannot occupy the same state, their quantum nature is in general highly relevant and a A˜ A˜ A˜ consistent quantum treatment of their dynamics is crucial. In the QCD Lagrangian in Eq. (1), quarks appear FIG. 18. Illustration of rescaled “quantum” three- and four- to be bilinear fields. Their real-time quantum dynamics vertices, that are g4. From Ref. [284]. may therefore be computed by numerically solving the ∼ operator Dirac equation coupled to the gluon fields. This can be achieved in an approximation where the This is illustrated in Fig. 17, which indicates the clas- gauge fields are treated using classical-statistical field sical three-vertex A¯2A˜ and four-vertex A¯3A˜ parts of ∼ ∼ theory and by employing a mode function analysis of the SYM, which are linear in the quantum field A˜. Figure 18 operator Dirac equation for quarks with available lat- gives the corresponding quantum three-vertex g4A˜3 4 3 ∼ tice simulation techniques [280–284]. For strong gauge and four-vertex g A¯A˜ parts of SYM, which are cubic fields 1/g, this approximate description amounts to in the quantum field∼ A˜ and suppressed by 2 powers of α ∼ S a systematic expansion of the quantum dynamics in relative to their classical counterparts. 2 αS g /(4π), where the leading order includes the full A similar analysis can be done for the Tr log ∆ 1[A,¯ A˜] ≡ − (non-perturbative) classical-statistical theory of gluons, contribution coming from the quark fluctuations. Ex- and the next-to-leading order takes into account back- panding this contribution in powers of the quantum field action of the quarks onto the gluons, which is controlled A˜ yields [284] by α N for N quark flavors. ∼ S f f This can be also directly understood from the path in- 1 2 4 3 Tr log ∆− [A,¯ A˜] g Tr jq[A¯]A˜ + g (A˜ ) . (79) tegral formulation of the quantum theory as described ∼ O in detail in Ref. [284] for Abelian and non-Abelian gauge ˜ theories with fermions on a lattice. Performing the Gaus- The linear term in A is proportional to the quark sian integration for the quark fields in QCD analytically vector-current in the presence of the classical gauge field jq[A¯][280–284]. yields a path integral for the gauge fields A± on the forward (+) and backward ( ) part of the closed time con- Correspondingly, in this formulation the limit g = 0 tour (see Sec.III) with an− effective action represents the classical-statistical field theory limit of pure Yang-Mills theory. In fact, the rescalings with the + 1 + + Seff [A ,A−] = Tr log ∆− [A ,A−] + iSYM[A ,A−] . gauge coupling employed in Eq. (78) reflect the fact that (77) for classical-statistical field theory the coupling can al- 1 + The term Tr log ∆− [A ,A−] arises from the Gaussian ways be scaled out by suitable field re-definitions, while 1 + integral over the quarks, where i∆− [A ,A−] denotes this not possible in the presence of quantum corrections. the inverse fermion propagator in the presence of the Since fermions are genuinely quantum, one cannot scale + gauge fields. Here SYM[A ,A−] is the Yang-Mills ac- out the coupling from their contributions, as seen in tion of the pure gauge theory evaluated on the upper Eq. (79), which starts at the order of αS. and lower branches of the closed time contour. According to the previous analysis, genuine quantum The power counting for strong gauge fields is most ef- corrections to the dynamics in pure Yang-Mills theory en- 2 ficiently done by a rotation of the -basis for the gauge ter only at the order of αS. Both the classical-statistical fields, splitting the gauge fields into± a “classical” part A¯ field contribution for the Yang-Mills part, and the lowest and a “quantum” one A˜, according to contribution from quark fluctuations to Seff , are linear in A˜. When we neglect higher-order corrections com- + 1 g 1 g A = A¯ + A,A˜ − = A¯ A.˜ (78) ing from terms with higher powers of A˜, the stationarity g 2 g − 2 condition δSeff [A,¯ A˜]/δA˜ = 0 yields the classical Yang- Expressed thus in terms of A¯ and A˜, the interaction terms Mills evolution equation for A¯ with the quark current of SYM can be similarly decomposed into classical and as a source term. This can be efficiently implemented quantum parts. numerically with sampling techniques using the Wilson 32 plaquette formulation on a lattice [280–284]. QS ΛQCD, so-called sphaleron transitions generate Numerical solutions of the non-equilibrium time evo- real-time transitions between configurations character- lution of gluons with dynamical quarks were obtained ized by integer valued topological charge that may be in Ref. [285] from (2+1) dimensional boost invariant separated by an energy barrier. simulations, in Ref. [286] in 3+1 space-time dimensions Interestingly, the boost invariant Glasma configura- for a non-expanding system, and in Ref. [287] for a re- tions discussed in Sec. IIIC2 do not correspond to in- alistic case with longitudinal expansion. The calcula- teger valued configurations of topological charge [228]; tions provide important first principles results on early sphaleron transitions therefore go hand in hand with quark production and the approach toward chemical the explosive growth of plasma instabilities that break equilibrium. The results for the gluon sector are in boost invariance, a phenomenon named “exploding line with earlier simulations without quarks as expected sphalerons” [355]. As noted in Eq. (76), the sphaleron at weak couplings, including self-similar scaling charac- transition rate is controlled by the spatial string tension teristic of the first stage of the bottom-up thermaliza- in the Glasma. tion scenario [27, 276]. Several properties of the quark While off-equilibrium topological transitions are an es- number distributions are carried over from the gluon sential ingredient, the CME in heavy-ion collisions is me- distributions, such as longitudinal momentum broaden- diated by the transport of quarks in this topological back- ing [287, 351]. ground and in the presence of external B fields. To ad- We also note recent work on the real-time propagation dress this problem of anomaly transport in such back- of heavy quarks in the Glasma that are important for a grounds, real-time lattice simulations were performed first-principles understanding of quarkonium production with dynamical fermions for 3 + 1 dimensional Abelian in heavy-ion collisions [352]. and non-Abelian gauge theories in Refs. [356, 357] for Classical-statistical lattice simulations cannot cor- given background gauge fields. In addition, transient rectly describe the late-time thermalization dynamics, anomalous charge production in strong-field QCD was when typical gluon occupancies become of the order of studied in Refs. [358, 359]. unity. The evolution may then be continued with effec- Anomalies have been investigated for Abelian theo- tive kinetic descriptions, as reported in Sec.VE3. ries off-equilibrium for the fully dynamical situation, including the back-reaction of the fermions onto the gauge fields, in one [360, 361], two [362] and three [363, 364] 2. Real-time off-equilibrium dynamics of quantum spatial dimensions. In Refs. [360, 361] dynamical topo- anomalies logical transitions in the massive Schwinger Model with a θ-term as a prototype model for CP-violation, are stud- The pair production of quarks and antiquarks lead to ied. A dynamical order parameter for quantum phase macroscopic manifestations of quantum anomalies, cor- transitions between different topological sectors is estab- responding to the breaking of classical symmetries by lished, that can be accessed through fermion two-point quantum effects. These may be observable in heavy-ion correlators. Using exact diagonalization techniques, the collisions in the form of a Chiral Magnetic Effect (CME) topological transitions have been shown to persist beyond whereby topological transitions in the very strong electro- the weak-coupling regime [360]. magnetic B fields at early times generate a vector current Quantum fluctuations lead to an anomalous violation in the direction of the B field [49, 353]. The prospects of parity symmetry in quantum electrodynamics for an for the discovery of this and related phenomena were re- even number of spatial dimensions, that was studied in viewed in Refs. [349, 350]. Ref. [362] using the previously described lattice simula- The key idea is that transitions between different topo- tion techniques. While the leading parity-odd electric logical sectors of the non-Abelian gauge theory can in- current vanishes in vacuum, a non-cancellation of the 0 duce a net axial charge asymmetry ja of light quarks, anomaly for strong electric fields off-equilibrium is ob- which can fluctuate on an event-by-event basis. In off- served with distinct macroscopic signatures. central heavy-ion collisions where strong electromagnetic The non-linear dynamics of the CME in QED was com- B~ -fields are present, this axial charge asymmetry can be puted in Ref. [363] using real-time lattice simulations. ~ 0 ~ converted into an electric current j jaB that is poten- For field strengths exceeding the Schwinger limit for pair tially observable. Since the large “magnetar∼ strength” production, one encounters a highly absorptive medium B fields die off very quickly after the collision [354], the with anomaly-induced dynamical refractive properties. CME is most pronounced at the earliest times after the An intriguing tracking behavior is found, in which the collision. system spends the longest time near collinear field con- The non-equilibrium dynamics of topological tran- figurations with maximum anomalous current. sitions in a highly occupied, albeit non-expanding, An interesting phenomenon observed in such simula- Glasma was studied in Ref. [322] by performing classical- tions of off-equilibrium QED plasmas is that of chiral in- statistical simulations and employing a cooling tech- stabilities proceeding through the primary and secondary nique to isolate infrared dominated topological transi- instabilities that we discussed previously culminating in a tions. Since gluon saturation generates a large scale self-similar turbulent magnetic helicity transfer to macro- 33 scopic length scales [364]; see also Ref. [365]. tion function f(X, p) 1 F (X, p) = i f(X, p) ρ(X, p), (80) − 2 ± V. EQUILIBRATION IN QCD KINETIC where “+” is for bosons and “ ” is for fermions and THEORY the quasi-particle momentum pµ −is the Fourier conjugate to the relative coordinate sµ. In general, there can be A. The quasi-particle description of QCD plasmas separate distributions for different color, spin and polarization components of the two-point correlation function. From the equation of motion for the statistical func- To solve the quantum equations of motion [Eq. (66)] tion one obtains the kinetic Boltzmann equation for the for the late time evolution toward local thermal equilib- distribution function, which is written as35 rium, an effective description with a well defined range of pµ∂ f(X, p) = C[f]. (81) validity at certain long time and distance scales is needed. µ − A well known example is kinetic theory, which describes The leading order collision term C[f] is obtained using the state of the system in terms of phase space distribu- a systematic power counting in the coupling constant; tions of particles. Such an effective kinetic description of this computation is non-trivial and various diagrammatic the plasma may be obtained from n-particle irreducible approaches have been employed to derive the relevant quantum effective action techniques by following along collision processes. For a systematic derivation of kinetic the lines of Refs. [294, 297, 366]. theory from the underlying field theory see [369–371] for The derivation of kinetic theory from the underlying the scalar case and [372, 373] for Abelian field theories. quantum field theory involves a series of approximations. For non-Abelian gauge theories at high temperatures, First, for the notion of particles with a well defined po- the leading order collision kernel appears at g4 order. sition and momentum between collisions to be valid, the However, in addition to elastic scattering processes, there de Broglie wavelength of the (quasi-)particles must be are collinear splitting processes that contribute at the small compared to the mean free path between collisions same order. The importance of the latter was recognized Likewise, quantum interference effects between succes- only later [374, 375]. The corresponding vertex correc- sive scattering events should not spoil a description in tions for the underlying quantum field theory can be for- terms of independent scatterings. For the weakly coupled mulated using higher nPI effective actions [294, 366]. QCD plasma at high temperature these questions were Once relevant physics processes are accounted for at addressed in a series of works culminating in the kinetic the given order, Eq. (81) describes the non-equilibrium theory formulation by Arnold, Moore and Yaffe [367]. evolution of QCD plasmas with the coupling constant g The phase space distribution functions employed in ki- as the only free parameter at high temperature (with the netic descriptions are derived from two-point correlation possible exception of heavy quark masses). In particular, functions of the underlying quantum field theory. In local one can use linearized kinetic theory to compute vari- thermal equilibrium, the system is locally homogeneous ous transport properties of the plasma around thermal and time independent. Therefore all two-point functions equilibrium: shear and bulk viscosities, conductivity, dif- can depend only on the relative coordinate sµ = xµ yµ. fusion and higher order transport coefficients [376–378]. For slow variations in space and time of the central− coor- For a recent comprehensive review on perturbative ther- dinates [Xµ = (xµ + yµ)/2], one considers the evolution mal QCD techniques in kinetic theory and beyond, see in X given by a gradient expansion of Eq. (66) for the Ref. [195]. As we will later discuss in detail, the QCD spectral function ρ and the statistical function F . To the kinetic theory also provides a phenomenologically suc- lowest order in gradients, the evolution equation for ρ is cessful picture of QCD thermalization in heavy-ion colli- not dynamical, and a quasi-particle description emerges sions [379–382]. For a complementary review, see [20]. from an on-shell spectral function ρ in the weak coupling limit [294]. 1. Chiral kinetic theory Here we consider the temperature T of the QCD plasma to be the single dominant energy scale in the problem. At leading order in the coupling, the self- In the rest of SectionV, we will discuss in detail the energy already receives contributions from an infinite equilibration of QCD in the framework of spin and color number of perturbative loop diagrams with hard (T ) averaged kinetic theory. Spin and color dependent kinetic internal momentum: Hard Thermal Loops (HTLs) [O368]. descriptions require extensions of phase space distribu- This results in quasi-particles acquiring a screening mass tions [383, 384]. Such theories must include a relativistic m gT . ∼ The equation of motion for the statistical function is solved by generalizing the Kubo-Martin-Schwinger 35 Keeping interactions with strong background gauge fields leads (KMS) relation to introduce a non-equilibrium distribu- to more general equations [18]. 34

covariant description of Berry curvature and of the dy- 1. Elastic two-body scattering namics of the chiral anomaly for spinning and colored particles in external background fields [385–387]. The 2 2 collision term for particle species s = a is Recent work in this direction includes a Wigner func- ↔ Z 3 3 3 tion approach [388–392], chiral effective field theory [393], 1 X d kd p0d k0 a p C2 2[f]( ) = 9 and a worldline formalism [394]. An important question ↔ 4pνa (2π) 2k2p02k0 to resolve in this context is the relation of the dynamics bcd a b c d c d a b of Berry’s phase to that of the chiral anomaly [395–397]. fpfk(1 fp0 )(1 fk0 ) fp0 fk0 (1 fp)(1 fk) × ± ± − ± ± A common goal of these approaches is a consistent frame- ab 2 4 (4) µ µ µ µ (2π) δ (p + k p0 k0 ), (83) work to describe anomalous transport in QCD that can × Mcd − − be matched to an anomalous relativistic hydrodynamic where P is the sum over all particle and antiparticle description at late times [398]. These studies have strong bcd species. The second line represents the phase space loss interdisciplinary connections to chiral transport across and gain terms. ab 2 is the 2 2 scattering ampli- energy scales ranging from Weyl and Dirac semi-metals cd tude squared and|M summed| over spin↔ and color degrees to astrophysical phenomena [399]. of freedom of the external legs, with ν = 2(N 2 1) for g c − gluons and νq = 2Nc for quarks. ab 2 The scattering matrix element cd in Eq. (83) should be calculated using in-medium|M corrected| prop- B. Leading order kinetic theory agators and vertices from the HTL effective La- grangian [195]. At leading order in the coupling constant and for hard p T external legs, the scattering matrix We will briefly recap here the main ingredients of QCD ∼ effective kinetic theory at leading order in the coupling element coincides with the tree level vacuum matrix ele- constant [367]. We will consider the time evolution of the ment; for instance, in the case of two gluon scattering, color and spin/polarization averaged distribution func- gg 2 2 4 st su tu tion fs with effective 2 2 scatterings and 1 2 gg = 8νgNc g 3 2 2 2 , (84) collinear radiation terms.↔ For a transversely homoge-↔ |M | − u − t − s neous and boost invariant system (applicable at early times in central heavy-ion collisions), the phase space dis- where s, t and u are the Mandelstam variables. In- tribution f s(τ, p) f s is a function only of Bjorken time medium corrections become relevant when t, u ≡ p (gT )2 is small but s is large, as is the case for− the− small∼ τ = √t2 z2 and momentum. The resulting Boltzmann angle scattering of hard particles. When the exchanged equation− is gluon or quark is soft, so that q = p0 p T in the t-channel (and likewise in the u-channel),| − | the vacuum z p ∂ s s s collision matrix suffers from a soft Coulomb divergence ∂τ fp = C2 2[f](p) C1 2[f](p) (82) 2 2 2 − τ ∂pz − ↔ − ↔ 1/(q ) . Therefore the problematic scattering |M|matrix∝ elements in this region need to be reevaluated using the non-equilibrium propagators for internal lines, 36 0 p with the massless dispersion relation, p = = p. which regulate the divergence [367]. Consequently, this kinetic theory describes a conformal| | For isotropic distributions and hard p T external system with temperature T as the only dimensionful legs, the soft self-energy (which cuts off the∼ long range scale. The index s refers to different particle species in Coulomb interactions) is proportional to the in-medium the theory such as quarks and gluons in SU(N ) gauge c effective masses of hard gluons and quarks [367]. For theory with N fermion flavors. The second term on the f gluons, it is given by (assuming that f q = f q¯) left hand side is due to the longitudinal gradients in a p p boost invariant expansion [400]. The expansion redshifts Z 3p the distribution in the pz direction, thereby making it 2 2 d g q mg = 2g Ncfp + Nf fp , (85) more anisotropic along the longitudinal direction. Dif- (2π)3p ferent stages of the thermalization process are defined by the competition between the expansion that drives the However, for anisotropic distributions the HTL re- 37 system away from equilibrium and the collision terms summed gluon propagator develops poles at imaginary that isotropize and equilibrate the system. frequency indicating the presence of a soft gauge instability [18, 403]. Formally, this restricts the applicability of kinetic theory to parametrically small anisotropies [367].

36 At leading order, we can neglect the thermal mass correction p 2 2 37 ms ∼ gT to the dispersion relation p = |p| + ms for hard Note that there are no unstable fermionic modes in anisotropic momenta |p| ∼ T on external legs. plasmas [401, 402]. 35

The rich physics of plasma instabilities has been stud- The UV divergence in the diffusion term is canceled by ied extensively [18]. While such instabilities are of funda- the corresponding IR divergence in the large-angle scat- mental importance at early times, remarkably, classical- tering term in Eq. (86). statistical simulations of the non-equilibrium field dy- We can now specify the isotropic screening prescription namics of the Glasma (discussed in Sec.IVC) show that for regulating the elastic collision kernel for anisotropic such instabilities do not play a significant role at late distributions: for a soft gluon exchange in the t-channel times in expanding (3+1) dimensional non-Abelian plas- (likewise for the u-channel), the divergent term is re- 2 2 2 2 mas. Motivated by these findings, phenomenological ap- placed by the IR regulated term t t(q + ξg mg)/q , 5/6 → proaches in kinetic theory simulations for anisotropic dis- where ξg = e /2 is a numerical constant fixed such that tributions use an isotropic screening prescription [379, the new matrix element reproduces the full HTL result 381]. for the drag and momentum diffusion properties of soft gluon scattering [379]. Similarly, one can regulate divergent soft fermion ex- 2. Fokker-Planck limit of elastic scatterings changes to reproduce gluon to quark conversion gg qq¯ at leading order for isotropic distributions [407, →412]. For isotropic distributions, the elastic collision kernel Formally, this regulated collision kernel is accurate for for soft momentum exchange can be rewritten as a drag small couplings and for near-isotropic systems. However, and diffusion process in momentum space [20, 404–409]. in practice numerical simulations for phenomenological First, one needs to separate the full collision kernel into applications are often performed for stronger couplings a diffusion term for soft momentum transfers q < µ and g 1 and anisotropic systems. large-angle scatterings q > µ, where the cutoff scale µ ≈ satisfies gT µ T : 3. Effective collinear one-to-two processes g g g C2 2[f](p) = Cdiff[f](µ) q<µ + C2 2[f](p) q>µ . (86) ↔ | ↔ | In addition to the momentum diffusion of hard par- The physics of the diffusion term is that of hard parti- ticles, soft gluon exchange can also take the particle cles being kicked around by the fluctuating soft gauge slightly off shell and make it kinematically possible for it fields generated by other particles. For an isotropic non- to split into two nearly collinear hard particles. Naively, equilibrium plasma, the expectation value of such gauge such a 2 3 process has an additional vertex rela- field fluctuations can be related to equilibrium fluctua- tive to elastic→ 2 2 scattering making it subleading tions with the help of an effective temperature T (taking in the coupling constant.↔ However both the soft gluon q q¯ ∗ fp = fp) exchange and the perturbed off-shell hard particle have 1/(g2T 2) enhancements from the propagators. These 2 Z 3 ∼ g d p g g q q compensate for the additional vertex insertion and the T 2 3 [Ncfp(1 + fp) + Nf fp(1 fp)]. (87) ∗ ≡ mg (2π) − nearly-collinear emission phase space [375]. For the same reason, multiple soft scatterings N +1 N +2 also have Note that although T = T in equilibrium, T is distinct to be summed over. → ∗ ∗ from the effective temperature defined by the energy den- Physically, this means that the nearly on-shell hard sity and used in Secs.VE andVI. Evaluating the colli- particle lives long enough before splitting to receive sion kernel in the limit of soft momentum transfer and multiple kicks from the plasma that destructively in- isotropic distributions, results in a Fokker-Planck type terfere, leading to the suppression of emissions from collision term very energetic particles. This phenomenon is known as the Landau-Pomeranchuk-Migdal (LPM) effect [413– ∂2f g g i ∂ g g 1 ij p 416]. Collectively these processes are described as an C [f](µ) = ηD(p)ˆp fp(1 + fp) + q , diff ∂pi 2 ∂pi∂pj effective 1 2 matrix element. In Eq. (82) it is denoted 1 2 ↔ (88) by C ↔ [f](p) and has the explicit form,

ij i j 3 Z where ηD is the drag coefficient, q =q ˆLpˆ pˆ + a (2π) X ∞ h 1 ij i j i i C [f](p) = dp dk (90) qˆ δ pˆ pˆ is the diffusion tensor, andp ˆ = p /p is 1 2 2 0 0 2 ↔ 2νap 0 the unit− vector. bc a γ (p; p0, k0)δ(p p0 k0) The transport coefficientsq ˆ andq ˆL can be evaluated bc − − a b c b c a using the resumed HTL propagators, while ηD is con- f [1 f 0 ][1 f 0 ] f 0 f 0 [1 f ] × ppˆ ± p pˆ ± k pˆ − p pˆ k pˆ ± ppˆ strained by the Einstein relation and the requirement b 2γ (p0; p, k0)δ(p0 p k0) that Eq. (88) vanish in equilibrium [404, 407, 410, 411]. − ac − − The leading order result forq ˆ is b a c a c b i f 0 [1 f ][1 f 0 ] f f 0 [1 f 0 ] , × p pˆ ± ppˆ ± k pˆ − ppˆ k pˆ ± p pˆ 2 2 2 g NcT mg µ ∗ where the unit vector pˆ = p/ p defines the splitting di- qˆ(µ) = log 2 . (89) 2π 2mg a | | rection and γbc(p; p0, k0) is the effective collinear splitting 36 rate. whereq ˆ(µ) is as given in Eq. (89). In the opposite limit 2 As required by detailed balance, Eq. (90) describes zpT /mg 1 (but still z 1), the successive scatter- ∗ both particle splitting p p0 + k0 and fusion p + k0 p0 ings by the medium interfere destructively, reducing the processes. Factoring out↔ the kinematic splitting function↔ emission rate to p0 1+z4+(1 z)4 Pg g z = = Nc − for the gluonic process 1/2 → p z(1 z) g z 1 νgαS qˆ(µ)p − γ (p; p0, k0) =P (z) , (96) g gg, this rate is given by gg LPM g g 4 → → (2π) z 2 Z 2 2 g νgg d h 2h Re Fg(h; p, p0, k0) where at next-to-leading-logarithmic order µ solves µ = γgg(p; p0, k0) =Pg g(z) · , 2 γE+π/4p → 2 3 2 2 2√2e − qˆ(µ)pz [422]. 4π (2π) 4 (2π) pp0 k0 (91) Owing to soft gauge field instabilities, collinear radiation in anisotropic plasmas contains unstable modes [423, where the integral has mass dimension 2 and is propor- 424]. In phenomenological applications these unstable tional to the virtuality acquired by the hard particle due modes are neglected and the isotropic approximation in to interactions with the soft gauge field. The complex Eq. (94) is employed instead. two-dimensional function Fg(h; p, p0, k0) (with mass dimension 1) solves the integral equation [367, 375, 417] C. Bottom-up thermalization Z 2 2 Nc d q 2h = iδE(h)Fg(h) + g T ⊥ (q ) (92) 2 ∗ (2π)2 A ⊥ 1. Initial conditions

3Fg(h) Fg(h k0q ) Fg(h p0q ) Fg(h + pq ) , { − − ⊥ − − ⊥ − ⊥ } BMSS [276] spelled out a bottom-up scenario for ther- where the energy diﬀerence between the incoming and malization beginning with the overoccupied Glasma dis- the outgoing states is cussed in SectionsII,III andIV. In this framework, the

2 2 2 2 momentum modes p QS can be interpreted as quasi- mg mg mg h ∼ δE(h; p, p0, k0) + + , (93) particles with a well-defined anisotropic distribution after ≡ 2k 2p − 2p 2pk p 38 2 1 0 0 0 0 time τ QS log αS− . The initial gluon distribution in kinetic simulations≥ of this scenario is parametrized at and h = (p k ) pˆ quantifies the transverse momentum 0 0 Q τ = 1 as [381] in the near collinear× × splitting. S 0 2 2 2 The second term on the r.h.s. of Eq. (92) can be inter- 2 p⊥+ξ0pz g 2A pT c − 3 p 2 preted as a linearized collision integral with loss and gain f = h i e h T ic . (97) p 2 p 2 2 2 g Nc p + pzξ0 terms describing the probability of a particle to scatter ⊥ in and out of transverse momentum h/p. The scattering The normalization A is chosen to reproduce the co- rate (q ) is proportional to the mean square fluctua- moving energy density τ = p dN /d2x dY . In this A ⊥ T c g tion of soft gauge fields; for isotropic distributions it is expression the gluon numberE h densityi at a fixed⊥ initial ra- given by [418], pidity is determined at LO by numerical simulations of the Glasma and the result can be read off Eq. (48). To ob- 1 1 q tain the first principles input for the initial gluon produc- ( ) = 2 2 2 . (94) A ⊥ q − q + 2mg ⊥ ⊥ tion as a function of rapidity, one further needs to solve the JIMWLK equations described previously in Sec.IID. Even with this isotropic approximation, Eq. (92) is highly Likewise, one can determine the average transverse mo- non-trivial. Various numerical schemes have been pro- mentum p 1.8 Q [425]. Finally, the anisotropy posed for solving it [419–421]. h T ic ≈ S parameter ξ0 is varied to quantify our ignorance of the longitudinal momentum distribution. 4. Bethe-Heitler and LPM limits of collinear radiation For the evolution of the overoccupied and highly anisotropic initial state, specified at its initial time by Eq. (97), the typical gluon occupancy and the deviation We will now discuss two limiting cases of the soft gluon 0 from isotropy can be monitored by computing the follow- radiation z = p 1. In the first case, the so-called p ing ratios: Bethe-Heitler (BH) limit, the interference between suc- 3 3 R d p g g 1 R d p 2 g cessive scattering events can be neglected. This corre- pf (2π)3 pfpfp T 2 (2π)3p p fp h i = , P = ⊥ . (98) sponds to the first (decoherence) term in Eq. (92) being R d3p g R d3p 2 g p 3 pf L 3 p f 2 2 h i (2π) p P (2π) p z p much larger than the scattering integral (pzg T /mg 1). In this case, the equation can be solved iteratively.∗ One obtains [408]

38 Plasma instabilities that are operational over shorter time- z 1 νgαS qˆ(µ)p g γgg(p; p0, k0) =Pg g(z) , (95) scales are well described by classical-statistical simulations; see BH → (2π)4 m2 g µ=emg Sec.IVB. 37

0 1.4 1.4 τQS 10 1.4 ≈ 1 τQS 10 1.2 1.2 ≈ 3 1.2 τQS 10 /n /n ≈ )

1 ) 1 1 /n )

0.8 0.8 /dp 0.8 2 g /dp dn/dp E dm 0.6 ( 0.6 0.6 d ( S ( 2 S

0.4 Q 0.4 0.4 Q 0.2 0.2 0.2 0 0 0 012345 012345 012345

p/QS p/QS p/QS

FIG. 19. Momentum differentiated plots of the (a) energy density (b) number density and (c) screening-mass at different stages of bottom-up thermalization. All curves are normalized by the instantaneous number density n and lines correspond to different 0 1 3 times (τQS = 10 , 10 , 10 ).

2. Stage one: collisional broadening 3. Stage two: collinear cascade

The solution of the collisionless Boltzmann equation Once the typical hard parton occupancy becomes (1), 2 O in the boost invariant expansion is a simple rescaling of the diffusion coefficient scales asq ˆ αSng, where we still initial longitudinal momentum that does not change the 2 ∼ have αSngτQS− 1. At this time, the longitudinal mo- typical occupancy but increases the anisotropy quadrat- mentum diffusion∼ rate and the expansion rate are com- ically in time. However in the presence of elastic col- parable, with the result that the longitudinal momentum lisions, gluons scatter into the longitudinal momentum reaches the constant value direction thus broadening the distribution. The longitu- 2 2 dinal momentum diffusion for anisotropic distributions pz αS QS . (102) can be estimated from the Fokker-Planck equation (88): ∼ This ensures that the momentum anisotropy remains z 2 g p ∂ g qˆ ∂ fp constant in the second bottom-up stage. ∂τ z fp = 2 , (99) − τ ∂p 4 ∂pz In this stage, in addition to elastic scatterings, medium induced collinear radiation becomes important, as it where we kept the dominant term on the right hand side. rapidly increases the population of soft gluons. Note that for a highly occupied anisotropic systemq ˆ The soft gluon multiplicity can be estimated using R g 2 ∼ p(fp) , Eq. (99) admits the scaling solution Eq. (73); the Bethe-Heitler formula [Eq. (95)]; integrating over p 2 as discussed in Sec. IV C 2, this solution is singled out in soft momentum mD < p < pz and neglecting loga- h i 3 the classical-statistical simulations. soft αS hard2 rithmic factors, one obtains ng τ 2 ng . The The physical picture is that the longitudinal momen- ∼ mg 2 2 2 2 p 2 screening mass is now dominated by soft isotropic gluons tum diffuses as pz qτˆ , whereq ˆ αSng/(QS pz ) p ∼ ∼ h i (m2 α nsoft/ p2 ). Using the previous expression and the hard gluon number density per rapidity is con- g ∼ S g h zi 2 for the longitudinal momentum, we can show that the stant (αSngτQS− 1). From this, it follows that the longitudinal momentum∼ decreases as soft and hard gluon multiplicities are of the same order at times 2 2 2/3 pz QS(QSτ)− . (100) 5/2 ∼ Q τ α− . (103) S ≥ S This clearly shows that the increase in anisotropy is milder than in the free streaming case. One obtains At this time, the soft gluons have thermalized among 2/3 2/3 themselves, forming a bath with an effective temperature. T / L (τ/τ0) and pf / p (τ/τ0)− , which areP inP agreement∝ with theh scalingi h i behavior∝ of the non- This marks the end of the second stage of bottom-up thermal attractor of Sec. IV C 1. The typical occupancy thermalization. becomes (1) at the time O 3/2 4. Stage three: mini-jet quenching τQ α− . (101) S ≥ S This is the first stage of bottom-up thermalization. As Even though the soft gluons have thermalized, the hard previously discussed, this corresponds to a “quantum gluons still dominate the energy density. They are, how- breaking” time where the classical-statistical approxima- ever, highly diluted ( fp / p 1); the non-equilibrium tion breaks down definitively. After this time, hard glu- modes are now underoccupied,h i h i as opposed to being ove- ons with pT QS are no longer overoccupied, although roccupied in the first bottom-up stage. Although soft they still carry∼ most of the energy and particle number. gluon emission is very efficient in populating the infrared, 38 the successive z 1/2 branching of modes is more effi- dominated by the soft sector, but there is still a notice- cient for energy transfer.∼ Such branching suffers from the able contribution to the energy density from the modes LPM suppression. The hard gluons are finally absorbed with p > QS. by the thermal bath in a “mini-jet” quenching that is formally identical to the jet quenching formalism that is typically applied when describing much harder modes. It is interesting to compare the momentum distribu- The system finally thermalizes when the energy in soft tions in Fig. 19 to the anisotropy and occupancy evo- and hard components becomes comparable. This hap- lution in Fig. 20 (which is a kinetic theory extension of pens at the time the lattice computation in Fig. 12). We mark the times τ Q 1, 10, 103 with a diamond, a circle, and a triangle, S ≈ 1 13/5 respectively, on the λ = 1, ξ0 = 10 simulation trajectory τ = C Q− α− , (104) thermal 1 S S (blue solid line). We observe that typical occupancies

2/5 drop rather quickly below unity and see a slight increase with the thermalization temperature T = C2αS QS. of anisotropy as it happens. However the slope of the Here C1 and C2 are (1) constants [426, 427]. This time anisotropy increase is different than the naive expecta- O 1/10 scale is parametrically αS− longer than when stage tion in the first stage of bottom-up thermalization and is two ends and therefore only cleanly distinguishable at dependent on the choice of initial conditions. asymptotically small values of the coupling. The bottom-up thermalization scenario provides an in- tuitive picture of equilibration at weak coupling. It is re- The anisotropy plateau of the second stage is already reached at τ Q 10, somewhat quicker than the para- markable, given the complexity of the thermalization pro- S ≈ cess in QCD, that this scenario allows one to relate the metric estimates suggest. Finally, because the soft sec- final thermalization time and temperature to the scale tor is more isotropic than that of hard gluons, we ob- for gluon saturation in the nuclear wavefunctions. serve that as the gluon number shifts toward lower momentum (see Fig. 19(b)), the anisotropy starts falling Asymptotic freedom tells us that the coupling constant sharply in Fig. 20. This marks the onset of the third must run with Q , which is the relevant hard scale in S stage of bottom-up thermalization. Although dilute hard the problem. Therefore an interesting consequence of modes still contribute significantly to the energy density, Eq. (104) is that τ log13/5(Q )/Q 0 as thermal S S the balance shifts toward more densely populated soft Q . Thus, contrary∼ to naive expectations,→ the S modes whose occupancy steadily increases as the system bottom-up→ ∞ thermalization scenario predicts that thermal- isotropizes. ization in the asymptotic Regge limit of QCD will occur nearly instantaneously relative to the size of the system. The bottom-up process finally ends when the system isotropizes. In practice, the third stage of bottom-up 5. Numerical realization of bottom-up thermalization equilibration is significantly longer than the second stage, 1/10 in contrast to the αS− difference in the parametric The thermalization time scales in the previous discus- time scales. sion were only parametric estimates. We will now discuss the results of a numerical implementation of the bottom- For an initial distribution with different initial up kinetic evolution from the overoccupied initial phase anisotropy values [ξ = 4, 10 dashed and solid lines in space distribution in Eq. (97) to the Bose-Einstein dis- 0 Fig. 20], the evolution follows a qualitatively similar path. tribution [381]. 2 Although we expect all initial conditions to converge at For ’t Hooft coupling λ = Ncg = 1 and initial thermal equilibrium, it is remarkable that different ini- anisotropy ξ0 = 10, we show in Fig. 19(a-c) the evolu- tializations already merge at rather large values of the tion of the gluon distribution (integrated over the spher- anisotropies T / L 10, when the system is still far ical angle) with different momentum weights. The three away from localP P thermal≈ equilibrium. This precocious panels correspond, respectively, to the distribution of the collapse to a universal curve, independent of the initial gluon energy density d /dp, the number density dn/dp, E 2 conditions, is termed a “hydrodynamic attractor.” This and the screening mass dmg/dp as a function of gluon phenomenon is discussed further in Secs.VE andVID. momentum. To factor out the dilution due to expansion, all of these quantities are normalized by the total gluon number density n. The lines correspond to different times Finally, Fig. 20 also shows kinetic equilibration with an τ Q 1, 10, 103. increasing coupling constant (and decreasing shear vis- S ≈ We see that at early times τ Q 1 10 the hard cosity η/s). For λ 5, corresponding to small values of S ≈ − ≥ p > QS modes dominate both the energy and particle η/s . 2 (and for which the initial occupancy is already number, and even have significant contributions to the below unity), the system starts to isotropize almost im- 3 screening mass. At very late times (τQS 10 ), the mediately and the distinct stages of the bottom-up sce- particle number and the screening mass are≈ completely nario are no longer clearly discernible. 39

FIG. 20. Gluon kinetic theory equilibration in anisotropy- occupancy plane for initial anisotropy ξ0 = 10 and diﬀer- FIG. 21. Time evolution of instantaneous scaling exponents ent values of the coupling constant. Times corresponding to extracted from diﬀerent sets of integral moments of the distri- 0 1 3 τQS = 10 , 10 , 10 are indicated by black symbols. Simula- bution. Horizontal lines indicate possible asymptotic values. tions with the smaller initial anisotropy ξ0 = 4 are shown as From Ref. [428]. dashed curves. Adapted from Ref. [381].

of overoccupied bosons. The same scaling exponents and D. Self-similar evolution in the high-occupancy the scaling function are also reproduced in kinetic the- regime ory simulations [379, 380]. Fermions are never overoccupied and chemical equilibration takes place over longer timescales than the direct energy cascade [412]. 1. Self-similar scaling

When characteristic field occupancies are sufficiently 2. Pre-scaling phenomenon large for the classical-statistical approximation to be valid, but small enough for the perturbative kinetic ex- In Ref. [428] it was found that the far-from-equilibrium pansion to apply, there is an overlapping regime where QGP already exhibits a self-similar behavior before the both approximations to the dynamics of the system are scaling exponents attain their constant values α = 2/3, valid [288, 289, 306]. β = 0 and γ = 1/3. The pre-scaling phenomenon− is As discussed in Secs.IVC andVC, the non- realized through the time dependent rescaling of the dis- equilibrium dynamics of the overoccupied plasma under- tribution function and its arguments (cf. [50]), goes a remarkable simplification in complexity by exhibit- α(τ) ing self-similar evolution. In kinetic theory language, the g prescaling (Qτ) β(τ) γ(τ) fp = fS (Qτ) p , (Qτ) pz , self-similar behavior refers to the situation in which the αS ⊥ particle distributions at different times can be related by (105) rescaling the momentum arguments and the overall nor- where α(τ), β(τ) and γ(τ) are generic time dependent malization; see Eq. (73), where α, β, and γ denote the functions. universal scaling exponents. The relations between the Figure 21 shows the evolution of time dependent scal- exponents are constrained by conservation laws and the ing exponents in QCD kinetic theory at very small cou- Boltzmann equation (81), for which Eq. (73) provides a plings and overoccupied initial conditions [428]. The solution. value of the exponents is calculated from the time de- Longitudinally expanding systems are anisotropic and pendence of various moments of the distribution: subject to soft gauge instabilities. Therefore from a per- Z 3 d p m n g turbative viewpoint it is very surprising that plasma in- nm,n(τ) = 3 pT pz fp . (106) stabilities do not seem to affect the late time evolution of (2π) | | the classical-statistical real time simulations, as shown in Different lines of the same color in Fig. 21 correspond Fig. 12. The self-similar evolution near the non-thermal to integrals with different powers of the momentum. It attractor is consistent with the bottom-up thermaliza- is important to note that the rescaling in Eq. (105) is tion scenario and numerical QCD kinetic theory simula- implicitly assumed to be valid in a certain physically rel- tions [381], which explicitly neglect plasma instabilities. evant momentum range. Therefore a finite set of mo- How to consistently solve the effective kinetic theory in ments of Eq. (106) contains all the physically relevant anisotropic plasmas is an important open question [18]. information in the distribution. As shown in Fig. 21, dif- Finally, as mentioned in Sec. IV C 2, in the case of the ferent extractions rapidly collapse onto each other and a non-expanding isotropic systems, the self-similar direct unique set of scaling exponents emerge that govern the energy cascade plays an important role in equilibration time evolution of all probed moments. 40

10 for the specific shear viscosity η/s can be a factor of 5 smaller than the leading order result at the accessible QGP temperatures T . 1 GeV. It is conceivable that a better reorganization of the perturbative expansion would result in an improved convergence at NLO [408]. 1 Nevertheless, for phenomenological applications in η/s heavy-ion collisions, the strong coupling constant value LO µEQCD αS 0.3 (g 2) is commonly used in leading order LO µMS ≈ ≈ NLO µ calculations. Examples of these include thermal photon EQCD emission [436], heavy quark transport [437], and parton NLO µMS 0.1 energy loss [438]. At this point, it is fair to admit that 0.2 1 10 100 1000 the leading order kinetic theory applications to equilibra- T [GeV] tion processes in the QGP do not provide a controlled expansion at realistic energies and therefore have large FIG. 22. The shear viscosity over entropy ratio as a function theoretical uncertainties. of temperature at leading (LO) and (nearly) next to leading On the other hand, QCD kinetic theory does contain order (NLO) thermal QCD. The bands correspond to the scale the necessary physical processes, such as elastic and in- variation of running coupling prescriptions. Figure taken from elastic scatterings, to describe QCD thermalization at [408]. weak coupling. Therefore in the absence of real time non-perturbative QCD computations, extrapolating the weak coupling results to larger couplings provides a use- The time dependent scaling exponents provide a more ful baseline that can be systematically improved upon. differential picture of how self-similar behavior and infor- As we will later discuss, the dependence on the cou- mation loss emerge near the non-thermal attractor. Here pling constant is better replaced by the value of shear the scaling exponents act as effective degrees of freedom viscosity η/s, a physical property of the QGP. The re- whose slowly varying evolution constitutes a hydrody- laxation to equilibrium is naturally controlled by the namic description of the system around the non-thermal strength of the dissipative processes. Therefore rescaling attractor. In particular, the time dependent exponents weakly coupled kinetic theory dynamics to small values of could be well suited to studying the evolution away from η/s (favored by hydrodynamic modeling of QGP) can be the attractor in equilibrating systems even if the non- fairly compared to heavy-ion phenomenology and other thermal attractor is never fully reached, for instance, at microscopic models. This includes the genuinely strongly larger values of the coupling. For related studies in scalar coupled systems discussed in Sec.VI. An indication that field theory, see also Ref. [429]. lessons learned from QGP equilibration in leading order kinetic theory are more robust than the LO expansion itself. E. Extrapolation to stronger couplings There have been a number of phenomenological applications of kinetic theory to the study of thermalization in QCD. Early notable examples include Refs. [439–441]. Thus far we have discussed a non-equilibrium QCD Numerical implementations of classical kinetic theory in- evolution scenario that is strictly valid only for g 1. cluding elastic gg gg and inelastic gg ggg gluon However the coupling constant is not parametrically scatterings were pioneered↔ in Ref. [442, 443↔]. We will small even at the Z boson mass scale, where α (M 2 ) S Z now focus on the results from the numerical implemen- 0.1179 0.0010 (g = √4πα 1.2) [430]. In the case≈ S tations of quantum kinetic theory, including all of the of finite± temperature perturbation≈ theory, the expansion leading order processes that were discussed in Sec.VB. parameter is αST/mD g – the convergence is therefore very slow∼ [431]. In∼ this section, we will therefore discuss phenomenological extrapolations of the QCD ki- 1. Hydrodynamic attractors in QCD kinetic theory netic theory to “realistic” couplings. The first calculation at next-to-leading order for QGP The universal macroscopic effective theory close to lo- transport properties was performed for heavy quark dif- cal thermal equilibrium is given by fluid dynamics con- fusion and the corrections were found to be large [432]. sisting of the conservation laws and constitutive equa- On the other hand, the NLO contributions to the pho- tions [444] ton emission nearly cancel and the overall contribution is only 20% [433]. Recently computations of the shear µν µν µν µ ∂µT = 0,T = T ( , u ,...). (107) viscosity,∼ quark diffusion and second order transport co- hydro E efficients have been extended to include higher order con- The only surviving information is contained in the macro- tributions (named “almost NLO” in [408, 434]) thanks to scopic fluid variables, the energy density and fluid ve- the breakthrough idea of evaluating HTL correlations on locity uµ; all other information about theE initial condi- the lightcone [435]. In Fig. 22, we see that NLO results tions has been lost. 41

is now called the hydrodynamic attractor. This notion is in fact much richer and its further aspects are discussed in Sec.VID. Figure 23 shows the pressure anisotropy L/ T as a function of rescaled time in an expanding homogeneousP P system for different values of the coupling constant. The system is prepared in an equilibrium state at initial time and then is allowed to undergo a boost invariant expansion which drives the system away from equilibrium. However, as the expansion slows down it relaxes back to isotropy, satisfying L/ T = 1. Note that the kineticP simulationsP for different couplings (which correspond to very different kinetic relaxation times) collapse onto each other even when the pressure anisotropy / is significant. Overall, the kinetic evo- PL PT FIG. 23. Pressure anisotropy evolution in expanding ge- lution is very close to that of an infinitely strongly cou- ometry. Gluon kinetic theory simulations λ = 1,... 10 are pled system. Although neither a weakly coupled kinetic compared to a supersymmetric Yang-Mills holographic model theory, nor an infinitely strongly coupled supersymmet- (λ = ). Note that here Ti is the initial temperature, so at ric Yang-Mills theory is an exact description of QCD in ∞ 4/3 3/2 late times (η/s) Tit 32w ˜ . Adapted from Ref. [445]. ≈ heavy-ion collisions, Fig. 23 gives some indication that in the rescaled time unitsw ˜ the final stages of QCD equilibration could follow a very similar hydrodynamic The surprising phenomenological success of viscous hy- attractor curve. drodynamics in describing many soft hadronic observ- To map the hydrodynamic attractor evolution in di- ables in heavy-ion collisions leads one to consider the mensionless timew ˜ to that in physical units, one needs possibility of whether a fluid dynamic description is ap- to fix the interaction strength by setting the shear vis- plicable to systems with significant deviations from local cosity over entropy ratio η/s and the dimensionful tem- thermal equilibrium. This topic was first investigated in perature scale. Extensive hydrodynamic model com- strongly coupled holographic models, and subsequently parisons to data constrain the shear viscosity to rather in the relaxation time approximation (RTA) kinetic the- small values of 4πη/s 2 close to Tc 155 MeV, al- ory and hydrodynamic models; see the reviews [12, 86] though its value at higher∼ temperatures is≈ not well deter- and Sec.VI. mined [448, 449]. The characteristic temperature scale in In the QCD kinetic theory simulations of boost invari- the hydrodynamic stage is well constrained by the trans- ant expansion of homogeneous plasmas [445–447], it was 3 verse entropy density per rapidity (sτ)hydro (T τ)hydro, observed that the energy-momentum tensor quickly be- which is directly proportional to the produced∼ particle comes a sole function of time measured in units of the multiplicity, and hence can be inferred from the experi- characteristic kinetic relaxation time39 τ η/(sT ), i.e., R ∼ mental measurements [450]. Inverting Eq. (108), we can τT relate the dimensionless timew ˜ in a longitudinally ex- w˜ . (108) panding conformal plasma to Bjorken time τ via ≡ 4πη/s 1/2 3/2 3/2 1/2 In such case the evolution of the energy-momentum τ = κ w˜ (4πη/s) (sτ)hydro− . (109) tensor can be characterized by L/ as a function of 1 P E 3 w˜ [47, 446]. Becausew ˜− is proportional to the Knud- The proportionality coefficient κ = (sτ)hydro/(τT ) be- sen number (the natural expansion parameter for devia- comes a numerical constant in thermal equilibrium, 2 tions from equilibrium) one would expect that for large where κ = νeff4π /90. Because the kinetic simulations w˜ the kinetic theory would agree with the viscous hy- converge toward conventional viscous hydrodynamic pre- 1 16 η/s dictions forw ˜ 1, it was estimated in Ref. [44, 447] drodynamic result L/ = 3 9 τT . Surprisingly, the & simplest viscous constitutiveP E relation− is already satisfied that the hydrodynamic description becomes applicable forw ˜ 1, when viscous correction is comparable to the for times τ 1 fm/c for η/s 0.16 and typical entropy & ≈ equilibrium≈ pressure. Such an effective hydrodynamic de- densities found in central Pb-Pb collisions [44, 447]. This scription of systems substantially away from equilibrium is consistent with the early hydrodynamization picture employed in the modeling of heavy-ion collisions.

39 The effective temperature can be defined as a function of the 2. Entropy production and initial energy density energy density that would play the role of the temperature in equilibrium. In conformal models it is given by the fourth root 2 1/4 of the energy density T = [E/(νeffπ /30)] . For an ideal gas of At even earlier timesw ˜ . 1, kinetic simulations with quarks and gluons, νeff = 47.5 and νeff = 16 for gluons only. very different initial conditions might not have collapsed 42

1.0

hydro 0.9 ) e 3 /

4 0.8 τ ( /

) 0.7 τ ( e

3 0.6 / 4 τ C = 0.87 QCD kinetics 0.5 ∞ C = 0.92 Boltzmann RTA ∞ C = 0.98 YM kinetics ∞ 0.4 C = 1.06 AdS/CFT ∞ free streaming viscous hydro

Energy attractor: 0.3 0.1 1 10 w˜ = τT /(4πη/s) eﬀ FIG. 25. Energy density evolution in a chemically equilibrating quark-gluon plasma. The vertical lines indicate the times FIG. 24. Hydrodynamic attractors for pre-equilibrium evolu- of approximate hydrodynamic, chemical and thermal equilib- tion of energy density for diﬀerent microscopic theories. From riums. From Ref. [43]. [42].

yet onto a single curve [381, 451]. Nevertheless, one may Ref. [42] showed that combining the entropy production employ the hydrodynamic attractor curve, which is reg- from hydrodynamic attractors with initial state energy ular forw ˜ 0, for a macroscopic fluid dynamic de- deposition in the CGC framework gives a good descrip- scription far→ from equilibrium [30]; see also Sec.VIC. tion of the centrality dependence of measured particle In kinetic theory at early times, such an attractor curve multiplicities. In particular, one can extend the original Bjorken estimate [22] of the initial energy density has vanishingly small longitudinal pressure (PL 0) and constant energy density per rapidity ( τ = const).≈ in heavy-ion collisions to much earlier times. For cen- Such initial conditions are typical for kineticE evolution tral Pb-Pb collisions at √sNN = 2.76 TeV one finds 3 in the bottom-up picture discussed in Sec.VC. Figure that (τ0) = 270 GeV/fm at τ0 = 0.1 fm/c, which is E 24 shows the energy density normalized by the equilib- a nearly 1000 times larger energy density than at the 4/3 E 4/3 rium evolution ( τ )hydro/τ for different hydrody- QCD crossover temperature. namic attractorsE obtained from QCD and YM kinetic theory [43, 44, 412, 447], AdS/CFT [29, 30, 47], and Boltzmann RTA [446, 452–455]. All attractors approach 3. Chemical equilibration of QGP the universal viscous hydrodynamic description at late 1 timesw ˜ > 1, while at early times they follow τ − , 40 E ∼ The early quark production from strong gauge fields corresponding to “free-streaming” behavior , which can was discussed in Sec.IVE. However, once the gluon fields be expressed as are no longer overoccupied, chemical equilibration has to 4/3 be described using QCD effective kinetic theory. A study τ (w ˜ 1) 1 4/9 E = C− w˜ . (110) of light quark flavor (up, down and strange) chemical ( τ 4/3) ∞ E hydro equilibration in isotropic and longitudinally expanding Here the dimensionless constant C quantifies the systems were recently presented in [43, 412]. At leading amount of work done. ∞ order, there are two fermion production channels: gluon A directly observable consequence of the equilibration fusion gg qq¯ and splitting g qq¯. It was found that → → process is the particle multiplicity, which is a measure of quark production processes are slower than gluon self- the entropy produced in heavy-ion collisions [456]. For a interactions. Therefore the gluon self-similar energy cas- given hydrodynamic attractor, the final entropy for boost cade seen in non-expanding isotropic systems is over well invariant expansion is proportional to the initial energy before an appreciable number of fermions is produced. and is given by the following simple formula [42] Similarly, gluons maintain an approximate kinetic equilibrium among themselves, while fermions attain a Fermi- 1/3 4 3/4 η 1/3 2/3 Dirac distribution at much later times. (sτ)hydro = C 4π κ ( τ)0 , (111) 3 ∞ s E The longitudinal expansion drives both gluons and fermions from the kinetic equilibrium, ensuring that equilibrium distributions can be approached only at late 40 The presence of scattering terms in Eq. (81) is crucial for the times when the expansion rate slows down. However, the early time anisotropy evolution, but not for the energy density. expansion does not seem to affect fermion production; According to the equations of motion ∂τ (τE) = −PL, and τE ≈ therefore, chemical equilibrium is achieved before ther- const as long as PL/E 1. mal equilibrium. For massless quarks, the quark-gluon 43 plasma satisfies the conformal equation of state = 1 captured by the linearized energy-momentum tensor re- P 3 E µν and the chemical composition of the plasma has little sponse functions Gαβ effect on the total evolution of the energy-momentum tensor. Therefore, hydrodynamization and chemical and Z thermal equilibrium are achieved sequentially [43, 412], µν 2 µν δT (τ , x) = d x0 G (x, x0, τ , τ ) satisfying x hydro αβ hydro ekt ττ αβ T x (τhydro) τhydro < τchem < τtherm . (112) δTx (τekt, x0) ττ . (113) × T x (τekt) Figure 25 shows the total energy density (red solid line), gluon energy density (green dotted line) and quark µν energy density (blue dashed line) as a function of time. Here the Green’s functions Gαβ (x, x0, τekt, τhydro) de- Gluons, which dominate initially, are quickly overtaken scribe the evolution and equilibration of energy- by quarks and the approximate chemical equilibrium en- momentum tensor perturbations from an early time τekt ergy ratios are reached by τ = 1.5 fm/c. This supports an to a later time τhydro. assumption of chemical equilibrium in the lattice equa- Remarkably, the linearized response functions are to tion of state used in hydrodynamic simulations of the a good approximation universal functions of the di- quark-gluon plasma. mensionless timew ˜, which is similar to the hydrody- Finally, an important piece of evidence for the forma- namic attractor describing the background equilibration of a chemically equilibrated QGP in heavy-ion col- tion. This provides a practical tool, the linearized pre- lisions is the enhanced production of hadrons carrying equilibrium propagator KøMPøST, for a pre-equilibrium strange quarks [457]. It is believed that in small col- kinetic description of heavy-ion collisions based on QCD lision systems such as proton-proton collisions, strange kinetic theory-[44, 447]. For the first time, the combi- quarks are not produced thermally in sufficient num- nation of the initial state IP-Glasma model discussed in bers and therefore that strange hadron production is Sec. IIIC3, kinetic equilibration and viscous hydrody- suppressed. Although in the previous kinetic descrip- namics evolution make it possible to describe all the early tion the three light flavors are all taken to be mass- stages of heavy-ion collisions in a theoretically complete less, the chemical equilibration rate can be used to es- setup. Experimental signatures of such setups are cur- timate the necessary life time (and system size) for the rently being investigated [460, 461]. creation of the chemically equilibrated QGP. The results in Ref. [43] showed that the plasma may reach As with the evolution of the background, the equili- chemical equilibrium for particle multiplicities down to bration of linearized perturbations in QCD kinetic the- 2 dNch/dη 10 . Strange hadron production in such high ory shares universal features with other microscopic de- multiplicity∼ proton-proton collisions will be tested in fu- scriptions [462–465]. Thanks to this universal behavior, ture runs of the LHC [458]. “universal pre-flow” is guaranteed to grow linearly with time for small gradients / 1 [382, 447, 466]: ∇E E 4. Equilibration of spatially inhomogeneous systems 1 ~ ~v ∇E τ, (114) ≈ −2 + Thus far we have discussed the equilibration of lon- E PT gitudinally expanding but otherwise homogeneous systems. Realistic heavy-ion collisions create initial condi- where for long wavelength perturbations ~ /( + ) = tions that are not homogeneous in the transverse plane. T const in conformal theories [382]. These∇E responseE P func- Such geometric deformations are strongly believed to be tions have been directly compared in Yang-Mills and the source of the multi-particle correlations that have RTA kinetic theories [467]. been observed experimentally [459]. In the weak coupling picture discussed in Sec.III, the spatial fluctuations are QCD kinetic theory simulations beyond the linearized the result of the uneven color charge distributions in the regime have not yet been accomplished, albeit there ex- colliding nuclei. On the largest scale ( 10 fm) it is deter- ist phenomenological studies of parton transport simula- mined by the overlap of the average nuclear∼ profiles. On tions based on perturbative QCD matrix elements [468]. nucleon scales 1 fm one can resolve event-by-event fluc- To what extent the macroscopic description in terms of tuations of individual∼ colliding nucleons. On yet smaller hydrodynamics can be applied to inhomogeneous systems scales 0.1fm one has stochastic fluctuations of color with non-linear transverse expansion is still an open ques- charges∼ in the internal structure of a nucleon. tion; see Sec. VI E 2 for a discussion of holography. How- Equilibration in kinetic theory, of small transverse per- ever, encouragingly, the results of several works [469– turbations around the homogeneous far-from-equilibrium 472] have demonstrated that for transversely expanding background, has been investigated in several works [44, systems the hydrodynamic attractor remains a good de- 382, 447]. Relevant information on the complicated ki- scription of local equilibration until the evolution time s netic evolution of the particle distribution fp can be becomes comparable to the transverse system size. 44

VI. AB INITIO HOLOGRAPHIC DESCRIPTION at λ one has OF STRONG COUPLING PHENOMENA → ∞ L3 2 N 2 = c (117) A. Holography and heavy-ion collisions GN π

and a particular matter sector. They both follow from SectionsII-V were concerned with the description of relevant string theory considerations [17]. heavy-ion collisions in a weak coupling QCD framework. One should view the Einstein gravity description as Here we will present what currently constitutes the only applicable only when λ . The QFT coupling con- approach capable of describing real time phenomena in stant does not appear in→ any ∞ form in Eq. (116), indicat- genuinely strongly coupled (1+3)-dimensional quantum ing that the coupling constant dependence drops from all field theories in a fully ab initio manner: holography [17, the QFT quantities that one can describe in this way for 34, 35]. λ . When the coupling constant is large but not in- The available description in this case does not make finite,→ ∞ the relevant description becomes Einstein gravity visible use of the gauge field degrees of freedom. Instead, supplemented by higher-curvature terms like the fourth it is based on the notion of a correspondence to higher power of the curvature. The form of these terms fol- dimensional geometries, which arise as solutions of Ein- lows again from string theory considerations, and in con- stein’s equations with a negative cosmological constant trollable situations they should necessarily be treated as and appropriate matter fields. small corrections. Because equations of motion become The guiding principle for our presentation will be uni- generically higher order in derivatives, the uncontrollable versality. We will be interested in phenomena shared extrapolation of the kind that one performs in kinetic across strongly-coupled quantum field theories and seek theory can be done here in only a very limited number in them theoretical lessons and phenomenological impli- of cases [474]. We will discuss these topics in Sec. VI F 2. cations for thermalization in QCD. A prime example of such a quantity is the aforemen- The “vanilla” setting in holography is five-dimensional tioned η/s = 1/(4 π) in all holographic QFTs, as long gravity with a negative cosmological constant, encapsu- as they are described by two-derivative gravity theories. lated by Eq. (116), which provides a consistent dual holo- One purpose of this review is to examine other kinds of graphic description of an infinite class of strongly coupled universalities that exist in the genuine non-equilibrium conformal field theories (CFTs) with a large number of regime. microscopic constituents [475]. Specifically, it describes a class of states in strongly coupled CFTs in which the only local operator acquiring an expectation value is the µν B. Controlled strong coupling regime energy-momentum tensor T . The most comprehensive holographic results on heavy-ion collisions concern this case. The best-known holographic gauge theory is the = 4 super Yang-Mills theory. At the Lagrangian level,N it can A generic five-dimensional metric can always be brought to the form be viewed as the gluon sector of SU(Nc) QCD coupled in a maximally supersymmetric way to four Weyl fermions 2 2 L 2 µ ν and six real scalars, both in the adjoint representation of ds = du + gµν (u, x) dx dx . (118) the gauge group [473]. This theory, as opposed to QCD, u2 − is conformally invariant; the coupling constant does not Here u is an additional direction emerging on the grav- run with the energy and becomes an external parameter ity side interpreted as a scale in a dual QFT. Einstein’s that defines the theory. equations put conditions on acceptable forms of g (u, x). In the planar N limit for asymptotically large µν c The most symmetric solution for gravity with a negative values of the ’t Hooft→ coupling ∞ constant cosmological constant has gµν (u, x) = ηµν , which is the λ 4παSNc , (115) four-dimensional Minkowski metric. This is the empty ≡ → ∞ AdS (anti-de Sitter) solution, which represents in gravi- the degrees of freedom in the = 4 super Yang-Mills 5 N tational language the time development of the vacuum in theory reorganize themselves in such a way that correla- holographic CFTs. The surface u = 0 acts as a boundary tion functions of certain operators, including the energy- of AdS5 and, more generally, gµν (u = 0, x) has the in- momentum tensor in an entire class of interesting states, terpretation of a metric in which the corresponding QFT can be computed using a 5-dimensional Einstein gravity lives. action with a negative cosmological constant The expectation value of the energy-momentum tensor Z 1 5 p 6 arises by looking at the subleading behavior of gµν (u, x) Sgrav = d x detg R 2 2 (116) 16πGN − −L close to the boundary [476, 477]. This is particularly simple for CFTs living in Minkowski space: and supplemented by matter fields. In Eq. (116) R is the Ricci scalar and L is the length scale set by the cosmo- 4π GN 4 logical constant. For the = 4 super Yang-Mills theory gµν (u, x) = ηµν + 3 Tµν (x) u + ... (119) N L h i 45

The second method, in which one solves gravity equations for different initial conditions, allows one to access a larger range of transient behavior. In particular, since we do not know which initial conditions are closest to the physics realized in experiment, one may want to scan as many of these initial conditions as possible to obtain a comprehensive picture. The downside is that in most cases this way of phrasing the problem is very specific to the geometric language of describing strongly coupled QFTs similarly to the one-particle distribution function FIG. 26. Penrose diagrams dual to far-from-equilibrium states being very specific to the weak coupling language. It in strongly coupled QFTs. (a) The system starts in the vac- does not allow for controllable comparisons with other uum with known bulk geometry and is perturbed by a non- frameworks akin to Ref. [445]. This can be somewhat trivial source, which appears as an asymptotic boundary con- ameliorated in holographic collisions in which the ini- dition in gravity. After the source is turned off, the QFT is in tial conditions for gravity originate from superimposing a non-equilibrium state modeled by a time dependent geom- two exact solutions corresponding to individual projec- etry. (b) The sources are always off, but one instead specifies tiles approaching each other. non-trivial initial conditions for the bulk metric. Adapted from Ref. [482]. Thermalization at strong coupling is a process in which one starts with an excited geometry in the bulk that after some time becomes locally very close to a black hole ge- The ellipsis denotes higher order terms in the small-u ometry. This encapsulates the notion of thermalization of expansion that turn out to contain only even powers of expectation values of local operators. Non-local observ- u with the coefficients being polynomials in T µν and ables discussed in Sec. VI F 3 can still show traces of non- its derivatives. One cannot a priori excludeh termsi like equilibrium behavior after local thermalization occurs. exp ( 1/u) that were considered in Ref. [478], but a gen- This should not come as a surprise since the thermaliza- eral understanding− of such terms is lacking. In the fol- tion of non-local observables is necessarily constrained by lowing, we will refer to the interior of AdS spacetimes causality. as “bulk physics” and the QFT physics as “boundary The discussion thus far was quite generic but the ex- physics.” plicit formulas were provided for strongly coupled CFTs. We are interested here in discussing time dependent While QCD is not a CFT, holography does not pose any states in Minkowski spacetime that model the dynamics conceptual problems in studying strongly coupled gauge of heavy-ion collisions. Given Eq. (119), such states can theories with a non-trivial RG, provided that the theory be probed through their expectation value of the energy- remains strongly coupled at all scales. This can be realized by introducing relevant deformations to holographic momentum tensor by solving the equations of motion of R 4 Eq. (116) as an initial value problem. This is achieved CFTs, modifying their Lagrangian by d x J O(x) with using numerical relativity techniques [479–481] and re- the scaling dimension ∆ < 4 of O(x). This triggers a quires one to specify initial conditions, and the solutions non-trivial bulk metric dependence on u providing the are subject to boundary conditions at u = 0. gravitational counterpart of a RG flow. There are two natural ways (with pros and cons) of In holography, the bulk object corresponding to O is a studying the non-equilibrium physics of quantum field scalar field φ appearing in the matter sector that supple- theories using holography: see Fig. 26. The first ap- ments the universal sector in Eq. (116). This scalar field proach circumvents the problem of finding initial con- is non-zero because the J of the QFT translates into its ditions, a key reason for its use in early works on the asymptotic boundary conditions; the latter generates a subject [45, 483]. Moreover, this approach allows one non-trivial profile for φ when solving the bulk equations to compare equilibration across theories by starting with of motion. Of course, the action for the bulk matter fields the same kind of initial state (such as the vacuum or a equips φ with a potential and the form of the potential thermal state) and perturbing it in a defined manner. In determines the physics of the RG flow in the correspond- particular, it underlies a significant body of research on ing QFT (including the information about ∆). We will understanding features of linear response theory in dif- review representative results in Sec. VI F 1. ferent microscopic models [82, 484–486]. As an example, To close, holography provides an ab initio window to Ref. [445] discussed in Sec.VE (see Fig. 23) compared study strongly-coupled QFTs, which include conformal the approach to hydrodynamics across models (including and non-conformal gauge theories. The conceptual prob- holography) using fully non-linear kicks. The drawbacks lem of fully non-perturbative real time evolution of an to perturbing simple states are, first, that the approach entire class of QFTs reduces in this setting to a technical to hydrodynamics is so rapid that it is difficult to disen- challenge of solving a set of coupled partial differential tangle exciting the system from its subsequent relaxation equations in higher number of dimensions, which is well and, second, that the class of states that one obtains in within reach of the existing numerical relativity methods. this way is rather limited. The holographic approach is very general and can be 46

equally well applied to the problem of time evolution of the nuclear medium in heavy-ion collisions, as well as to problems originating in branches of physics [473, 487]. 1.25 Finally, we stress again that holography as a tool for 1. QFT comes with its own limitations illustrated by the 0.75 fact that one needs to work in regimes where the gravity 0.5 description is classical or semi-classical. 0.25

0.5 1. 1.5 C. Early times in Bjorken flow at strong coupling FIG. 27. Evolution of the effective temperature as a func- Bjorken flow [22] without transverse expansion in a tion of time for three different states with nonzero initial en- CFT setting is arguably the best studied example of a ergy density. The gray curves denote the far-from-equilibrium nonlinear non-equilibrium phenomenon in holography.41 regime. The blue dashed area extending indefinitely to the Because of the conservation of the energy-momentum right mark the applicability of viscous hydrodynamic rela- tensor, all the non-trivial information about the dy- tions truncated at the third order in derivatives (122). The ττ red dotted curves denote the series in Eq. (120) extracted namics can be extracted from T (τ). This using the method of Ref. [489]. Adapted from Ref. [479]. parametrization is useful for describingh i ≡ the E early time physics relevant for modeling initial stages of ultrarelativistic heavy-ion collisions. Toward this end, Ref. [489] noticed that combining therefore logically possible42 that there are initial metrics Eq. (119) (expanded to sufficiently high order in u) with defined on other bulk constant time slices that give rise a general Taylor series ansatz for (τ) around τ = 0 does to energies densities of the form other than those dictated not lead to singular bulk metricE coefficients in the limit by Eq. (120). Second, note that in Eq. (120) any number of the lowest order terms can vanish and the energy τ 0 as long at the early time expansion contains only positive→ even powers of proper time: density at early time can behave like τ 2 [491]. E τ 0 ∼ Another point is that there are various reasons≈ why one (τ 0) = + τ 2 + τ 4 + .... (120) E ≈ E0 E2 E4 may not want to start the evolution at τ = 0. The most obvious one is related to creating either non-equilibrium The coefficients in Eq. (120) are not entirely arbitrary, initial states from the vacuum or thermal states, as dis- but they are related one-to-one to the near-boundary ex- cussed in Fig. 26. In these cases, the sources will need pansion of the bulk metric that satisfies the constraints some non-zero time to act [45]. The other reason is more on the initial time slice, as encapsulated by Eqs. (118) conceptual and is related to the observation that while and (119). The early time series (120) turns out to have one should not expect the infinitely strongly coupled ap- a non-zero but finite radius of convergence, which allows proach to be a phenomenologically viable description at one to reliably study the initial dynamics of the system. τ = 0, it may become one from some τ > 0 onward. Note However, as shown in Ref. [489], and as later corrobo- that from the gravity point of view, it is not clear that rated in Ref. [479] using the full numerical solution of all the initial conditions set in the bulk for τ > 0 are bulk Einstein’s equations, the radius of convergence of extendable to τ = 0 and, as a result, one can view them Eq. (120) is much too small for us to see the transition as a priori containing richer behavior. to hydrodynamics. This point is illustrated in Fig. 27 using the effective temperature (see footnote 39). Further- Because of this issue, it is unclear whether all well more, simple analytic continuations of the series (120) behaved initial conditions for numerical relativity sim- based on the Padéapproximants method provide unreli- ulations actually describe genuine states in underlying able extrapolations. QFTs. Unlike Refs. [47, 479, 490], Refs. [30, 469, 492] ini- One lesson therefore is that the only method for ob- tialized their codes at later times with turned off sources. taining T µν in strongly coupled QFTs beyond the early In particular, Ref. [30] found initial conditions at some h i early but non-zero τ such that 1 initially, which is time limit examples is to use numerical relativity. Before E ∼ τ we proceed in that maner, a few more comments related clearly very different from Eq. (120). to Eq. (120) are in order. First, the analysis of Ref. [489] As discussed in Sec.VE, the transition to hydrody- uses regularity of the initial metric on a particular con- namics can be observed in the cleanest way upon in- stant time slice of the bulk geometry, namely, the one troducing the scale-invariant time variablew ˜ defined in dictated by the coordinates chosen in Eq. (118). It is Eq. (108) and using T / L, L/ or any reasonable P P P E

41 Recently devised hyperbolic quenches [488] adopt an eﬀectively (1+1)-dimensional boost invariant geometry of heavy-ion colli- 42 Ref. [490] chose initial surfaces in the bulk as in Fig. 26, with sions in the context of condensed matter physics. results being consistent with Eq. (120). 47

known analytically. The current state of the art was set 10 by Ref. [512] which, improving on the earlier eﬀorts of Ref. [493], numerically computed the lowest 380 terms in the expansion given in Eq. (122). On top of the power law 5 late time (w) expansion come exponentially suppressed terms that represent transient phenomena that are also 0 visible in linear response theory [29, 493, 513, 514]. AAACCHicbVDLSsNAFL2pr1pfVZduBovgqiQiqLuKG5cV7EPaUCbTSTt0MhNmJkIJ/QH3bvUX3Ilb/8I/8DOcpFlo64ELh3Pu5d57gpgzbVz3yymtrK6tb5Q3K1vbO7t71f2DtpaJIrRFJJeqG2BNORO0ZZjhtBsriqOA004wucn8ziNVmklxb6Yx9SM8EixkBBsr9foRNmNNVHo9G1Rrbt3NgZaJV5AaFGgOqt/9oSRJRIUhHGvd89zY+ClWhhFOZ5V+ommMyQSPaM9SgSOq/TQ/eYZOrDJEoVS2hEG5+nsixZHW0yiwnfmJi14m/utlitKhXthvwks/ZSJODBVkvj5MODISZamgIVOUGD61BBPF7AeIjLHCxNjsKjYabzGIZdI+q3vn9au781rjoQipDEdwDKfgwQU04Baa0AICEp7hBV6dJ+fNeXc+5q0lp5g5hD9wPn8AtNaa9Q==

A Fig. 27 illustrates time evolution of the effective temperature T (τ). Hydrodynamics is applicable at a time -5 after which the pressure anisotropy deviates only slightly from Eq. (122). As discussed in detail in Ref. [479], 0.0 0.2 0.4 0.6 0.8 1.0 the precise moment of applicability of hydrodynamics depends on the desired accuracy of the match to Eq. (122) and the order of the truncation. Of course, the latter as- FIG. 28. T µν in a holographic CFT as a function of the pect should be understood in the sense of an asymptotic h i dimensionless clock variable w for 29 different initial states series. (gray curves). Magenta, blue and green curves denote predic- The main message from the studies in Refs. [30, 45, 47, tions of hydrodynamic constitutive relations truncated, re- 469, 479, 490, 492] and related works is that low order hy- spectively, at first, second, and third order [Eq. (122)]. The orange curve is the hydrodynamic attractor [30]. Adapted drodynamic constitutive relations (see Eq. (122)) become from Refs. [30, 47, 479]. applicable at strong coupling after τ = (1/T ). This is the regime where the pressure anisotropyO in the system is sizable, as illustrated in Fig. 28. Since the system is function of this ratio such as still far away from local thermal equilibrium, the word hydrodynamization was coined in [48] to distinguish the T 3 P 3 applicability of viscous hydrodynamics constitutive rela- T L L A = P − P = P − , (121) T tions from local thermalization. The latter phenomenon /3 2 P + 1 L E P occurs at strong coupling for times that can even be 10 times larger than the hydrodynamization time. which was introduced in Refs. [47, 86, 490] as a function The modern perspective on hydrodynamics, viewing of w τT . Note that in the strongly coupled limit of in particular the gradient expansion as a part of a trans- holography≡ 4π η/s = 1, and we will simply denotew ˜ as w. series, was reviewed in detail in Ref. [86]. In the fol- It is well understood by now that at late time A (w) lowing, we will discuss an alternative way of thinking acquires the form of a trans-series [29, 493–495] known about the applicability of hydrodynamics using the con- from the studies of asymptotic expansions in mathemati- cept of hydrodynamic attractors. These objects made cal and quantum physics: see Refs. [496, 497] for reviews. their appearance in Sec.VE and bear a structural sim- The hydrodynamic part is a series in inverse powers of w ilarity to the non-thermal attractors (fixed points) dis- and has a vanishing radius of convergence43. Its first few cussed in Sec.IVC. terms read

2 1 2 2 log 2 2 A (w) = w− + − w− π 3π2 D. Hydrodynamic attractors in holography 2 2 15 2π 45 log 2 + 24 log 2 3 + − − w− + ..., (122) 54π3 Hydrodynamic attractors proposed in Ref. [29], and developed in many works including Refs. [30, 451–454, see Refs. [47, 86, 490, 505–511]. Equation (122) should 469, 494, 498, 500, 512, 515–532] can be viewed as a way be understood as expressing the energy-momentum ten- of approaching the problem of information loss about the sor in terms of hydrodynamic constitutive relations to underlying state from the point of view of observations the third lowest order. The first term carries informa- restricted to the energy-momentum tensor T µν . tion about the first derivative of flow velocity and the h i shear viscosity, while the second term is a contribution Reexamining Fig. 28 through these lenses, we see that from second derivatives of velocity and associated trans- a set of different states considered there follows to a port coefficients. The third term is the last one that is good approximation a single profile of A (w) from a certain value of w onward. This is the notion of attraction between different initial conditions as seen by an effective phase space covered by A at a fixed value of w. While this observation does not call for invoking a trun- 43 The same applies to Gubser [498] and cosmological [499] flows but is not the case for Bjorken flow with fine-tuned transport co- cated gradient expansion, the emerging universality seen efficients [500]. Furthermore, Ref. [501] used the results of [502– in Fig. 28 agrees very well with a hydrodynamic gradient 504] to show that divergence of the hydrodynamic gradient ex- expansion truncated at low order. These observations pansion is a generic feature of linear response theory. lie behind the name hydrodynamic attractor and parallel 48

0.5 L/ℰ AdS/CFT late-time attractor of inflationary cosmology [533]. The previously discussed distance notion leads to the magnitude of velocity of a given state being A 0(w) , and slowly evolving solutions τT (note that Ref. [532| ] instead| defined regions of slow evo- 0.05 0.10 0.50 1 5 10 lution) are those that lead to the flattest form of A (w). In Fig. 28, such a solution given in Ref. [30] using fine τ -dominated -0.5 QNM tuning initial conditions is denoted by an orange curve. Note that this solution at early times has A very close to 3 . This corresponds to free streaming = 0, which 2 PL -1.0 evades the study of initial conditions behind Eq. (120) reported in Ref. [489]. FIG. 29. Hydrodynamization of states whose gravity dual ini- We stress that the notion of slowly evolving solu- tially has support close to the boundary (dashed curves) or tions is a priori independent from the notion of conver- deep in the bulk (solid curves) initialized at different times gence (attraction). However, in full phase space, or at (different colors); see the text for details. Adapted from least a representative projection of it, one can make a Ref. [469]. thermodynamic-like argument, as in Ref. [532], in favor of typical states residing in the slow roll region. One can think of slow evolution as a generalization of the notion the discussion in Sec.VE1. of the gradient expansion that does not involve an expan- We now step back and review this phenomenon from sion with individual terms badly behaving at very early a broader perspective advocated recently in Ref. [532]. times, namely, as inverse powers of w in Eq. (122). To proceed, we will utilize the aforementioned notion Finally, the approach to the hydrodynamic attractor at of phase space introduced in this context in Ref. [520]. strong coupling and mechanisms that govern it were ex- Specifically, one should think of A as a particularly clean amined in Ref. [469] by looking at results of simulations scale invariant way of representing information about with different initialization times. This is depicted in T µν and w as a useful way of parametrizing time evo- Fig. 29. The idea behind it, building on earlier results in lution,h i adjusted to the fact that transient phenomena in Refs. [453, 525], is that information loss can be driven by conformal theories occur over time scales set by the en- at least two distinct mechanisms. The first one involves ergy density. exponentially suppressed corrections to Eq. (122), which Of course, knowing A at a given value of w does not stem from linear response theory physics. The charac- allow one to predict its value later, since the true mi- teristic feature of them is that their decay rates do not croscopic variable is the bulk metric. A larger chunk of depend on w. The second mechanisms driving the infor- information is provided by considering A and some of mation loss is expansion, which for the comoving velocity µ µ 1 its derivatives with respect to w (or and its derivatives u ∂µ ∂τ gives µu = . What one therefore expects ≡ ∇ τ with respect to τ). Such sets of variablesE form the no- is that information loss predominantly driven by the ex- tion of an effective phase space. In fact, there is a limit pansion is going to be faster at earlier times (smaller w) to how large such phase space needs to be: the numeri- and slower at later times. Indeed, such a feature was cal solutions of Einstein’s equations displayed in Fig. 28 seen in Ref. [469] for hydrodynamic models and for the typically require one to specify a few functions on several kinetic theory for early initialization times. However, in dozen grid points. holography this does not seem to be the case and the One can then assign a metric to an effective phase approach to the hydrodynamic attractor takes roughly a space, i.e., the distance between points representing fixed amount of time regardless of the chosen initializa- classes of solutions here, and track how such a distance tion time (see Fig. 29), which is consistent with it being changes as time evolves. The loss of information is ex- governed by transients. pected to make a set of solutions reduce its volume in the effective phase space. For example, in Fig. 28 one introduces the notion of proximity between two solutions E. Holographic collisions A (w) A (w) . With respect to this notion, various | 1 − 2 | solutions from the chosen set eventually collapse to ap- In CFTs, Bjorken flow in the absence of transverse ex- proximately a point in A at a fixed value of w. It should pansion has a high degree of symmetry that allows for be clear that the hydrodynamic attractor at a given value comprehensive studies of hydrodynamization and associ- of w is not a notion relevant to all states. It needs to be ated phenomena. In particular, the numerical approach regarded as a statement about properties of some class pursued in Refs. [45, 480, 490] fully determines the evo- of states initialized prior to that. lution of T µν as a function of proper time τ upon spec- Furthermore, assigning a distance measure to phase ifying oneh positivei number (initial energy density ) and space allows one to define the notion of slow evolution. a single function of the AdS direction u, see Eq.E (119). This topic was introduced in Ref. [29] under the name As a result, it was possible to comprehensively scan over slow roll approximation, which originates from the field initial states in search of universal behavior. 49

If one relaxes these symmetry assumptions and allows for dynamics in the transverse plane [464, 534], the space of initial conditions becomes too large to allow for a comprehensive analysis. Therefore, one wants to have another guiding principle to arrive at interesting configurations for modeling non-equilibrium evolution of T µν in holographic heavy-ion collisions. The key ideah is toi study holographic collisions of localized lumps of matter [24, 46, 491, 535–539]. The localized objects (shockwaves) in question move at the speed of light and are characterized by the following non-zero components of T µν , h i T 00 = T 33 = T 03 = µ (x )h(x0 x3) , (123) h i h i ±h i ± ⊥ ∓ where x0 is the lab-frame time, x3 is the direction along which the object is moving (specified by in the argument of h), µ (x ) 0 is an arbitrary∓ function spec- ± ⊥ ≥ ifying the transverse profile and h(x0 x3) 0 is an- µν ∓ ≥ FIG. 30. T resulting from a collision of thin planar shocks other arbitrary function specifying the longitudinal pro- with % d =h 0.08i (124). Top panel: lab-frame energy density file [538]. While a single projectile defined by Eq. (123) as a function of time x0 and longitudinal position x3. Be- is exact, the superposition of two projectiles approaching tween the remnants and the central rapidity region, there are each other and overlapping in the transverse plane leads small regions of negative energy density. Bottom-left panel: at mid-rapidity, the transverse and longitudinal pressure af- to a non-trivial collisional process. 00 2 Such collisions should not be regarded as literal mod- ter the collision are consistent with T τ in Eq. (120). Bottom-right panel: the color encodingh i denotes ∼ deviations els of the early stages of heavy-ion collisions, since from constitutive relations and points to the applicability of the projectiles do not originate from QCD. (See, how- hydrodynamics. The post-collision T µν does not have a rest ever, [540–542].) Instead, one should treat holographic frame in the gray region [543]. Adaptedh i from Ref. [24]. shockwave collisions as illustrating possible far-from- equilibrium phenomena accessible in a fully ab initio way at strong coupling that goes well beyond the previously discussed Bjorken flow geometry. projectiles is not boost invariant even when they are in- 1. Planar shocks finitely thin. The extent to which this is the case was explored in [24] and, quite remarkably, the results fit well [545] with complex deformations of the purely boost The simplest settings to consider are collisions of pla- invariant flow introduced in [546]. nar shockwaves: objects defined by Eq. (123) with µ constant. Following Ref. [24], one can consider a Gaus-± As it turns out, the features of the collision change as a sian longitudinal profile for h of the form, function of γ. First, the collision of “low-γ” thick shock-

2 (x0∓x3)2 waves proceed such that the two blobs of matter first 0 3 Nc 4 h(x x ) = % e− 2d2 , (124) merge and their subsequent evolution is approximated ∓ 2π2 well by viscous hydrodynamics. This is referred to [24] and recognize that, in heavy-ion collisions, the dimen- as to the Landau scenario [547, 548]. As seen in Fig. 30, sionless product of the amplitude % (not to be confused the “high-γ” regime of thin shocks leads to a rich set with the previously discussed charge density) and the of transient physics before hydrodynamics becomes ap- 1/2 width d decreases as γ− as the total center-of-mass plicable. Another important phenomenon, discussed in energy of the collision (√s = 2γMion) increases. [535, 549, 550], is the notion of longitudinal coherence. Within this analogy, high energy collisions correspond This notion applies to the “centre-of-mass” frame of high to collisions of very thin shockwaves44. The collisions of energy collisions and states that the longitudinal struc- projectiles defined by Eq. (123) do not lead to longitu- ture of projectiles does not leave an imprint on the tran- dinal boost invariance since the initial state of the two sient form of the energy-momentum tensor in the post- collision region provided that it is sufficiently localized. Finally, despite the differences between thin and thick shocks’ collisions at transient times after the remnants 44 The problem of colliding planar projectiles in Eq. (123) with dissolve, which take a much longer time than shown in h(x0 ∓ x3) ∼ δ(x0 ∓ x3) was originally posed in [544] and ad- Fig. 30, the structure of the late time hydrodynamic flow dressed in an early time expansion akin to Eq. (120) in [491]. is very similar in the two cases [536]. 50

t = 2 t =0 t =2 t =4 F. Other aspects of thermalization at strong coupling

1. Non-conformal strongly-coupled QFTs

energy density All the strong coupling results reviewed thus far have x collision axis t = 1.125 t =0 t =1.125 t =2.25 concerned well deﬁned QFTs without a scale. As reviewed in Sec.VIB, in holography there are no conceptual obstacles to breaking conformal symmetry. How-

one of the transverse directions ever, considering QFTs with non-trivial renormalization Energy energy flux group flows does make gravitational calculations more in- x volved due to the presence of field(s) in addition to grav- z ity that one needs to solve for and due to the more in- FIG. 31. Energy density in holographic heavy-ion collisions volved near-boundary analysis that generalizes Eq. (119). with transverse dynamics. Top panels: off-central collision All in all, the number of results on this front relevant for with modestMomentum elliptic flow. Bottom panels: proton-nucleus col- thermalization in QCD is significantly lower than in the lision as modeled with a shockwave with a small Gaussian ex- conformal case but still allows one to draw conclusions. z tent in the transverse plane (left projectile) and a planar shock Broadly speaking, there are two approaches to this (right projectile). The smaller projectile punches out a hole in problem. The first is top-down and studies renormaliza- the larger projectile and excites matter at mid-rapidity, lead- tion group flows originating from turning on a relevant ing to substantial radial flow. Adapted from [537] and [538]. deformation in a known holographic CFT. The prime example is the so-called = 2∗ gauge theory arising as a deformation of = 4 superN Yang-Mills theory by adding masses to halfN of its fields [551]. The advantage of this approach is that one makes sure that one is studying well 2. Transverse dynamics in holography defined features of a strongly coupled QFT. The draw- back is that such well understood examples are scarce and might have rigid features that do not exist in QCD. Studies of hydrodynamization in the presence of trans- The other class are so-called bottom-up models that verse expansion in [537–539] still define the state-of-the- couple AdS gravity to a bulk scalar field or fields whose art in numerical applied holography. Figure 31 illustrates Lagrangian is chosen by insisting that it reproduce some the profile of the energy density in such collisions. The desired feature of QCD. One such approach was intro- main lesson from these works is the early applicability duced in [552, 553] using the QCD β-function as a guide- of viscous hydrodynamics not only for very large longi- line; another model [554] uses as a benchmark reproduc- tudinal gradients of the energy-momentum tensor (as for ing the QCD equation of state at vanishing baryon den- Bjorken flow and planar shocks) but also in the pres- sity. ence of large transverse gradients generating transverse Furthermore, one can also introduce confinement by expansion. making the geometry end smoothly in the bulk [555]. One can think of it as the manifestation of a mass gap, From the perspective of these strong coupling results, with no excitations below the lowest bound state energy. the applicability of hydrodynamics in pA and even pp The breaking of conformal symmetry introduces an collisions [539] is as natural as the applicability of hy- additional scale in the problem of thermalization and drodynamics in Bjorken flow and can be explained in changes hydrodynamization times, although in none of terms of fast decaying contributions to the trans-series the setups explored to date by an order of magnitude or for T µν . Further, these works corroborate studies in more with respect to the strong coupling CFT predic- [534h] byi providing successful tests of the early time ration of 1/T [556, 557]. This also indicates that thew ˜ dial expansion model proposed in [466]. Toward this end, defined∼ in Eq. (108) plays a less prominent role in non- Ref. [537] found very small elliptic flow despite off-central conformal QFTs than it does in strongly coupled CFTs. collision and confirmed that near mid-rapidity the en- Furthermore, the hydrodynamic gradient expansion ergy flux grew linearly with proper time, as predicted in acquires new transport terms, most notably the bulk vis- Ref. [466]. cosity ζ. Hydrodynamization and on a much later time scale isotropization still do occur, but there are now two As discussed in Sec. VE4, such “universal flow” at more emergent time scales related to i) the applicability small wavenumber is also reproduced by weak coupling of the equation of state and ii) the expectation value of kinetic theory. It would be interesting to see whether the operator breaking conformal symmetry reaching its the full transverse response functions of the energy- thermal value. The relation between these scales depends momentum tensor in strong coupling agrees with those on the details of the model [558–560]. discussed in Sec.VE4 in the context of kinetic theory. Finally, confinement represented holographically as the 51 appearance of an infrared wall leads to the new physical least in some cases, as a model of QFT at a finite value effect in which excitations of the bulk geometry and mat- of the “coupling constant.” ter fields bounce back and forth as in a cavity [561, 562]. In the context of the planar shockwave collisions dis- Such an effect was not present in the studies reviewed cussed in Sec.VIE, perturbative calculations in λGB pre- earlier and has not yet been explored in the context of dict less stopping and more energy deposited close to the expanding plasmas. lightcone [569, 570]. There also appears to be a correlation between the shear viscosity and hydrodynamization times, as encapsulated by Eq. (108). 2. Away from the strong coupling regime Furthermore, linear response calculations performed exactly in λGB reveal that the singularity structure of real time correlators in equilibrium can change drasti- Another important direction studied in the context of cally as the coupling is varied [485]. In particular, the thermalization in strongly coupled gauge theories con- results seem to mimic features expected from a kinetic cerns corrections from finite values of the coupling con- theory, such as the appearance of branch cuts [484, 486], stant. In the context of the = 4 super Yang-Mills rather than single pole singularities known in strongly theory, the leading correctionN in the inverse power of coupled QFTs [82]. the ’t Hooft coupling constant behaves as λ 3/2; on the − The situation at a nonlinear level is more complicated. gravity side, it arises at least in part due to a partic- While the equations of motion are second order, the coef- ular expression quartic in the curvature [563]. Such a ficients in front of the highest derivative terms are com- higher curvature gravity action when treated exactly is plicated and can vanish in regions of spacetime. This ill-behaved due to the Ostrogradsky instability [474]. It signals a breakdown of the initial value problem. Over- is, however, not meant to be considered as such, since it coming this obstacle is currently an active topic of re- is just an effective field theory truncated at a fixed order search in the relativity community [571–574]. in the derivative expansion. Finally, we mention a more phenomenological set of Treating these higher curvature terms as small contri- hybrid approaches [575–578] in which gravity is used to butions to the Einstein’s equations with negative cosmo- model the IR of a QFT and a weak coupling framework logical constant allows one to derive the leading order is put to work to represent the UV. Both frameworks are corrections to various holographic predictions at λ . coupled to each other and predictions rely on a subtle For example, they increase the shear viscosity of→ the ∞ interplay between the two combined models. Such a set- = 4 super Yang-Mills theory from η/s = 1/(4π) at ting bears structural similarity to [552, 553], as discussed λN [13] to η/s = 1/(4π) 1 + 15 ζ(3) λ 3/2 for − in Sec. VI E 2. However it uses the gravitational descrip- large→ but∞ finite λ [564, 565]. × tion only where it can be trusted, which is the regime The previously discussed quartic term is the first where the coupling constant is large. higher order term appearing for the = 4 super Yang-Mills theory, but remember thatN the Einstein- Hilbert action with negative cosmological constant de- 3. Non-local correlators scribes infinitely many strongly coupled CFTs. For some of these [566], the leading correction to Eq. (116) is quadratic in curvature and can be written as the so-called All the quantities we have discussed at strong cou- Gauss-Bonnet term pling concerned one-point functions of gauge invariant operators. Because of the underlying large-Nc hierarchy, λ the problem of finding connected two- and higher-point δSGB = GB L2 R2 4R Rab + R Rabcd . (125) grav 2 − ab abcd functions correlation functions decouples from the problem of finding the one-point functions discussed thus far. This contribution has λGB 1 in top-down settings Such correlation functions can be thought of as correla- | | and the sign of λGB can be either positive or negative. tion functions of the bulk free (for two-point functions) or As a result, there are bona fide holographic CFTs for weakly interacting (for higher-point functions) quantum which the ratio of shear viscosity to entropy density is fields45 living on top of gravitational backgrounds when slightly lower than 1/(4π)[566, 567]. This important the insertion points of the bulk correlators are taken to result showed that the celebrated value of 1/(4π) is not the boundary [579]. In the following, we will focus on the lower bound in nature as originally conjectured in two-point functions. Ref. [15], although the existence of another lower bound Since we are talking about time dependent setups and, cannot be excluded. hence, Lorentzian correlators, the distinction between Furthermore, the combined gravity action of Eqs. (116) and (125) leads to, at least superficially, second order equations of motion. While it is known that microscopi- cally this does not correspond to a well behaved QFT out- 45 They should not be confused with the underlying strongly cou- side the regime λGB 1 [568], in the spirit of bottom- pled QFT for which both the classical bulk background and free up models discussed| | in Sec. VI F 1 one can treat it, at bulk quantum fields are effective descriptions. 52

Wightman, retarded, or other correlators is appropri- decades, greatly advanced our understanding of decon- ate [80, 81, 580, 581]. Toward this end, the retarded fined QCD matter. Successful multi-observable data-to- correlator depends only on the gravitational background model comparisons have provided ample evidence that a and captures the response of the strongly coupled QFT new phase of matter is created with the thermodynamic to sources. However, the Wightman correlator depends properties predicted by lattice QCD [448, 457, 590–594]. on both the constructed gravitational background and While thermodynamic features of QCD can also possi- the state of the bulk quantum field. Therefore, its cal- bly be extracted from neutron star physics, with a spec- culation is challenging in time dependent processes and, tacular recent example being the gravitational radiation unless one creates a non-equilibrium state using sources pattern of neutron star mergers [595], heavy-ion colli- exciting the vacuum or a thermal state [582–585], one has sions are likely the only place in the Universe where the to deal with an additional freedom of initial conditions non-equilibrium many-body properties of QCD can be to scan. explored. It should perhaps not come as a surprise that to date We will not discuss here signatures of high parton den- there have been no studies of such correlators in an ex- sity matter in the hadron wavefunctions that were dis- panding plasma. Noteworthy works in this area were cussed elsewhere [596]. Uncovering definitive evidence for Refs. [582–585], which studied equilibration of scalar and systematic study of gluon saturation is a major goal operator two-point functions under a spatially uniform of the Electron-Ion Collider (EIC) [186, 187]. We note quench. that diffractive and exclusive signatures of gluon satura- Many researchers use a proxy for correlators being a tion at the EIC are especially promising [597, 598]. bulk geodesic spanned between the insertion points ap- Our focus here will be on quark-gluon matter formed propriate for operators of large scaling dimension in the after the collision. In the high parton density framework Euclidean signature. However, in Lorentzian signature of the CGC EFT, the Glasma matter at the earliest times this is an uncontrollable approximation [584, 586, 587]. is most sensitive to the physics of gluon saturation. In- On the other hand, the comparison between Wightman deed, if the contributions of the initial state can be iso- functions calculated according to the correct microscopic lated from that of the final state, heavy-ion collisions prescription and the geodesic proxy led to qualitatively could present definitive evidence for gluon saturation. similar results [584, 585]. However, as we later discuss, a clean separation of ini- If one takes this as an indication of the geodesic proxy tial and final state effects in the complex spacetime evo- as capturing the relevant physics, then one lesson follow- lution of the heavy-ion collision is challenging [599]. Nev- ing from such studies is that the symmetrized correla- ertheless, data from both light and heavy-ion collisions tor with small spacelike separation between its insertion at RHIC and the LHC can help constrain key features of points thermalizes sooner than the one with larger sepa- gluon saturation, with an example being the energy and ration [588, 589]. This is also natural from the point of nuclear dependence of the saturation scale QS. view of causality. Furthermore, Ref. [585] observed a relation between the equilibration time scale of the spatially Fourier trans- A. Electromagnetic and hard probes formed Wightman function and the equilibration time scale of 1/T governing hydrodynamization at strong cou- Since the Glasma matter is likely to be far off- pling and discussed in Sec.VIC. This study was done for equilibrium at the earliest instants of the heavy-ion colli- a scalar operator, which does not exhibit a hydrodynamic sion, its features can be extracted most directly in probes tail. that are the least sensitive to the later stages of the col- It is natural to conjecture that the energy-momentum lision. The primary candidates here are electromagnetic tensor or a U(1) current Wightman function would take probes of the medium such as photons and dileptons that, longer to equilibrate due to the presence of hydrodynamic once emitted, do not interact with the medium. modes, but such studies have not been yet performed. The problem here is that photons and di-leptons are Finally, as noted in Ref. [585], we should stress that the produced continuously throughout the spacetime evolu- aforementioned momentum space features of equilibration of the quark-gluon matter and from the subsequent tion do not easily translate to the real-space properties. hadronic phase as well. Current models of heavy-ion col- This is so because sharp features in the correlator do not lisions, which include photon yields from the pre-hydro necessarily reside at small distances. kinetic theory phase tend to under predict the produced photon yields [461, 600]; for an alternative mechanism, see [601]. VII. SIGNATURES OF NON-EQUILIBRIUM Photons emitted from highly occupied Glasma have QCD been suggested as an additional source of radiation [602]. While phenomenological model comparisons show a sig- The experimental heavy-ion collision programs at BNL nificant Glasma contribution [603], the theoretical mod- and CERN, combined with advances in theory and em- eling of photon rates currently carries sizable uncertainty. pirically motivated models have, over the last couple of Besides photons and di-leptons, inclusive yields of high 53

momentum strongly interacting final states are also sensi- forward to disentangling initial state physics of CGCs tive to gluon saturation and to early time dynamics in the and the Glasma at early times from late time dynamics heavy-ion collision. These include hadrons at high trans- is to look at the evolution of two-particle correlations verse momenta, jets, and heavy quarkonia. Gluon satu- with their rapidity separation [621] Another approach ration influences the production rates for these processes is to study the long range correlations of particles with and rescattering in the Glasma influences their dynam- large transverse momenta that do not follow hydrody- ics. These effects are most pronounced for p QS. We namically [622, 623]. discussed heavy quark pair production in the⊥ ∼ Glasma in SectionIV. The diffusion coefficient of these heavy quarks was computed recently in this framework and scales as C. Bulk observables 3 QS [604]. Heavy quark diffusion in Glasma-like environ- ments and their subsequent evolution were also explored We previously discussed limiting fragmentation of recently in several works [605–607]. A non-trivial prob- hadron distributions and its potential to distinguish ini- lem is distinguishing this early-time evolution of heavy tial and final state effects in hadron-hadron collisions [25]. quarks from their late time evolution [608–610]. Similar 46 We will now discuss other bulk observables in high energy considerations also hold for the propagation of jets in nucleus-nucleus, hadron-nucleus, and hadron-hadron col- the Glasma [613–616]. lisions that can help constrain the properties of saturated Higher point correlations of hard probes, add signif- gluons and their early-time evolution. In the smaller sys- icant sensitivity to the dynamics of quark-gluon mat- tems, even if the system hydrodynamizes quickly, the ter off-equilibrium. An example is the potential of large shape fluctuations of partons will provide insight two-particle Hanbury-Brown–Twiss (HBT) photon in- into multi-parton correlations in the initial state [256]; terferometry to study early time dynamics [617]. Such understanding these from first principles is a challenging measurements are sensitive to the large longitudinal- problem [624] that may also require the EIC to resolve. transverse anisotropies that are not reflected in photon A number of works have explored applications of holo- yields. However experimental measurements of soft pho- graphic ideas to the study of bulk observables in heavy- ton correlations are very challenging experimentally and ion collisions. A universal prediction of holography is high statistics would be needed to disentangle the signal. that of hydrodynamization being distinct from local thermalization. A specific phenomenological investigation B. Long-range rapidity correlations implementing this idea used holographic boost invariant dynamics with transverse expansion as a successful model of preflow [464]. Another development was discussed in Long-range rapidity correlations are an important tool Ref. [542], which treated the planar shockwave collisions in disentangling initial and final state effects in hadron- discussed in Sec. VI E 1 as an explicit model of initial nucleus and nucleus-nucleus collisions. This is because state physics. While this study recovered qualitative fea- causality dictates that the latest time that a correla- tures of soft particle spectra, the rapidity distribution of tion can be induced between two particles A and B that produced particles is too narrow relative to the exper- freeze-out is given by imental data. It would be very interesting to explore more complicated holographic models of heavy-ion colli- yA yB τ = τfreeze out exp | − | . (126) sions and constrain them with experimental data. − − 2 In a thermalizing system, the loss of information of the initial conditions manifests itself as the production Thus two particles that are long-range in rapidity y A of entropy. Therefore, if the system locally thermalizes y 1 would be correlated at very early times| in− B and its flow is nearly isentropic, the measured number the| collision [230]. A particular example is the so-called of particles probes the entropy produced during the non- “ridge” effect, reviewed in [253], which correlates two par- equilibrium evolution of quark-gluon matter. The CGC ticles not only in rapidity but also in relative azimuthal framework accounts for the increase of particle multiplic- angle [618]. A recent summary of the physics of initial ity with increasing collision energy with the growth of the state correlations was given in Ref. [619]. saturation scale Q [625]. Recent calculations of entropy However, if hydrodynamic flow also sets in early, this S production in the equilibration processes using hydrody- ridge could be a final state effect [620] due to the under- namic attractors provide a quantitative relation between lying boost-invariance of the hydrodynamic fluid. A way the energy deposition in the CGC picture and the final particle numbers [42]. On the other hand, the energy of the observed particles depends on the work done during the entire expansion 46 The final stage of “bottom up” thermalization corresponds to and therefore has different dependencies on the dynam- the “jet quenching” of partons of momentum ∼ QS that are quenched to the thermal medium; this framework also explains ics of the pre-equilibrium stage. Comparing these two ro- key features of the quenching of very high momentum jets in the bust experimental measurements (energies and multiplic- QGP [611, 612]. ities) already casts doubts on the complete equilibration 54 of QGP in peripheral nucleus-nucleus collisions [42, 470]. next will bring many opportunities to exploit the signa- Many of the experimental signatures of QGP tures that we have articulated here, and likely several (strangeness enhancement, jet suppression, flow harmon- novel ones, of the properties of QCD off-equilibrium. ics, etc.) show a smooth dependence on system size from central to peripheral nucleus-nucleus, proton-nucleus, and proton-proton collisions. As the system size shrinks, VIII. INTERDISCIPLINARY CONNECTIONS so also does its lifetime, corresponding to an increase in the relative importance of non-equilibrium QCD process Understanding the thermalization process in QCD as- increases. sociated with heavy-ion collisions addresses some of the Equilibration studies in large systems already put a most fundamental questions in quantum dynamics, with lower bound below which the system will not reach hy- exciting interdisciplinary connections made to very dif- drodynamization or chemical equilibrium [43, 44]. There- ferent many-body systems. The transient “fireball” ex- fore, explaining observed signals of collectivity (or the panding in vacuum explores far-from-equilibrium condi- absence thereof) in small collisions systems requires tions at early times, followed by a series of character- a proper treatment of non-equilibrium QCD dynam- istic stages that are finally expected to lead to a fluid- ics. Some recent examples of work in this direction in- like behavior governing the approach to local thermal clude studies of flow harmonics [259, 472], parton energy equilibrium. Very similar questions of equilibration and loss [626] and heavy-quark evolution [605]. Furthermore, the emergence of collective behavior from the underlying as discussed in Sec. VI E 2, hydrodynamization without unitary quantum dynamics are relevant for diverse appli- equilibration of small systems is very natural in hologra- cations ranging from high-energy and condensed matter phy. physics to practical quantum technology. For reviews in Also noteworthy is recent phenomenological work [627] the context of condensed matter physics, see [631–633]. quantifying the role of non-equilibrium dynamics in the Several non-equilibrium phenomena were first pro- Chiral Magnetic Effect, which we discussed in SectionIV. posed in the context of QCD matter in extreme condi- A topic that demands further investigation is the ori- tions and then explored and experimentally probed in al- gin of the very large vorticities measured in off-central ternative quantum many-body systems. For instance, the heavy-ion collisions, as extracted from measurements of phenomenon of prethermalization [273] with the rapid es- the polarization of Λ-baryons [628]. The vorticities are tablishment of an effective equation of state during the introduced on macroscopic scales of the order of the sys- early stages of heavy-ion collisions [274, 634] has been ex- tem size; how these propagate efficiently down to the plored for early Universe inflaton dynamics [635] and con- microscopic scales of Λ is not yet understood. densed matter systems [636–638], and experimentally discovered in ultracold quantum gases on an atom chip [639]. In turn, aspects of entanglement represent one of the D. Future prospects major overarching schemes in contemporary physics of quantum-many body systems, and gravity in and out of A recent recommendation from the European Strat- equilibrium, while investigations about its relevance to egy for Particle Physics report emphasized that the main the thermalization process in QCD are relatively recent. physics goal of future experiments with heavy-ion and There are many excellent topical reviews on entangle- proton beams at the LHC will be a detailed, experiment and we refer the reader to Refs. [640–643]. We mentally tested dynamical understanding of how out-of- discuss some aspects of entanglement in our context in equilibrium evolution occurs and equilibrium properties more detail later. arise in a non-Abelian quantum field theory [458, 629]. To capture the thermalization dynamics in QCD re- The scheduled runs 3 and 4 of the LHC will mark a lated to heavy-ion collisions, detailed comparisons take decade of high-statistics data across system sizes at the into account the fact that the coupling of non-Abelian highest achievable collision energies. gauge theories is not a constant but changes with char- In the United States, continued operation of RHIC acteristic energy or momentum scale in a particular way. will provide further insight into several of the signatures While strong at low scales, the coupling becomes weak at that we have discussed. In particular, with the antic- sufficiently high energies because of the phenomenon of ipated commissioning of the sPHENIX detector [630], asymptotic freedom [644, 645]. Even in the high-energy hard probes of QCD off-equilibrium will be studied in limit, where the gauge coupling is weak, one faces a a dynamical range that is complementary to that of the strongly correlated system because a plasma of gluons LHC. with high occupancy [f(QS) 1/αS(QS)] is expected to Looking further into the future, the EIC project has form; see Sec.III. Such a transient∼ over-occupation lead- received Critical Decision Zero (CD0) approval from the ing to strong correlations even for weakly coupled sys- U.S. Department of Energy. The EIC will explore with tems can be found in a variety of physical applications high precision the landscape of hadron structure at high that are far from equilibrium. Examples include the pre- energies [186, 187]. heating scenario for the very early stages of our Universe One may therefore anticipate that this decade and the after a period of strongly accelerated expansion called 55 inflation [646] and the relaxation dynamics in table-top gases at temperatures that differ by 20 orders of mag- setups with ultracold quantum gases following a sudden nitude! Strong interactions also play a central role in change in external control parameters such as magnetic holographic approaches, a concept that is addressed in fields [54]. Sec.VI, and there are concrete proposals on how to re- The very high level of control in experiments with alize holographically systems resembling unitary Fermi synthetic quantum systems, such as ultracold quantum gases starting with Refs. [675, 676]. A comprehensive gases, enables dedicated quantum simulations. These review of common aspects of QCD, unitary Fermi gases systems provide very flexible testbeds, which can realize and holography was provided in Ref. [677]. a wide range of Hamiltonians with variable interactions and degrees of freedom based on atomic, molecular and optical physics engineering [647]. Since these setups can B. Highly occupied systems I: Preheating in the be well isolated from the environment, they offer the pos- early Universe sibility of studying fundamental aspects such as the thermalization process from the underlying unitary quantum The dilution of matter and radiation during the infla- evolution. tionary period of the early Universe leads to an extreme While digital quantum simulations based on a Trot- condition that may be well characterized by a pure state terized time evolution on a universal quantum computer with vacuum-like energy density carried by a time depen- are challenging to scale up, present large scale analog dent coherent inflaton field with large amplitude [646]. quantum simulators using ultracold quantum gases al- A wide class of post-inflationary models with weak cou- ready explore the many-body limit described by quantum plings exhibit the subsequent decay of the inflaton field field theory [54, 55, 647–663]. In principle, with quan- amplitude via non-equilibrium instabilities [315, 316]. tum simulators non-universal aspects of the dynamics of Detailed mechanisms for the origin of an instability and gauge theories can be studied. This was first achieved for the scattering processes are different than in QCD with Abelian gauge theory with digital quantum simulations, strong color fields. such as those using trapped ions [664] or superconducting However, the rapid growth of fluctuations from the in- qubits [665]. flaton decay leads to a non-linear time evolution that An interesting possibility to consider is the applica- follows along lines similar to those outlined in Sec.IV for tion of a hybrid quantum-classical framework to real time QCD. For instance, for scalar fields with weak quartic in- problems. This has been discussed in a “single particle” teraction λ 1, a corresponding overoccupation 1/λ digital strategy for scattering problems whereby higher up to a characteristic momentum scale is achieved∼ after loop quantum contributions can be simulated digitally the instability. Likewise, at this stage the prethermal- and the background gauge field treated in principle on ization [273, 274] of characteristic properties, including a quantum simulator [666, 667]. It is also important to an effective equation of state, is observed in these scalar note that scalable analog systems for the quantum simu- models [635]. lations of gauge theories using ultracold atoms have been Moreover, a self-similar attractor solution is ap- reported [668, 669]. We anticipate significant progress in proached subsequently, as discussed in Sec. IV C 3. Com- all of these approaches to quantum computation of real pared to the longitudinally expanding QCD plasma, a time problems in the decade ahead. major difference stems from the isotropic expansion of the Universe. Some aspects of isotropic expansion can be lifted for the inflaton field dynamics by introducing A. Strong interactions: Unitary Fermi gas suitably rescaled conformal time and field amplitudes, such that the dynamics is essentially that of Minkowski A paradigmatic example of the interdisciplinary cross- spacetime without expansion [314]. In fact if compared fertilization among the different physical applications is to QCD dynamics without expansion, then characteristic the work on collective motion of a unitary Fermi gas. dynamical properties such as the values of scaling expo- Near unitarity, the s-wave scattering length, which char- nents in the attractor regime agree with what is found for acterizes the two-body interaction strength, becomes self-interacting scalar field dynamics with quartic inter- very large and the effective scale invariance of the interactions in the absence of spontaneous symmetry break- action at unitarity can lead to universal behavior [670], ing [319]. which can also be accessed out of equilibrium [671]. Many This concerns the gauge theory’s direct energy cascade similarities for dynamical properties, such as a low ratio toward the perturbative high-momentum regime [310, of shear viscosity to entropy density, have been discussed 317, 318], as well as the inverse particle cascade toward in this context in comparison to QCD. See the discussion low momenta in the non-perturbative regime associated in Sec.V. with non-thermal fixed points [325]. In turn, scalar fields We noted that heavy-ion experiments indicate that the with longitudinal expansion seem to exhibit several uni- hot quark-gluon plasma may be described as the most versal features shared with QCD dynamics in the tran- perfect fluid realized in nature [2,4, 672–674]. The only sient scaling regime [313]. In particular, the inverse cas- serious experimental competitors are ultracold quantum cade essentially follows the behavior of the correspond- 56

a b 1 1 0 conditions. After an initial non-equilibrium instability 1 0 50 0 10 2 3 regime, all data in the self-similar scaling regime are seen 40 -1 2 2 30 1 to collapse to a single curve after rescaling with time us- (k,t) θ

20 0 f ‒ α atoms/µm ) ing universal scaling exponents. While this example con- 10 -1 ref 0 1 3 cerns infrared scaling, bi-directional scaling including a

0 Initial condition self-similar evolution toward higher momenta with sub- -1 20 µm 0 1 2 3 Fourier transform transversal spin α sequent thermalization was experimentally analyzed in β 1 0.7 0 Ref. [56]. 10 −1 10 Rescaled amplitude (t/t amplitude Rescaled 0.5 −1 2 10

−2 exponents Scaling 10 0.3 3

−3 10 −2 −1 10 10 0.0025 0.005 0.01 0.02 β k (1/µm) Rescaled momentum (t/tref) k (1/µm) D. Highly occupied systems III: Classicalization FIG. 32. (a) Absorption images of different magnetic hyper- and unitarization of gravitational amplitudes fine states of a spin-1 Bose gas with the extracted transversal spin (solid lines) for three different far-from-equilibrium initial conditions. (b) All initial conditions lead to the same univer- An intriguing idea is that of black holes as long lived sal scaling behavior, such that all data points collapse onto states of highly occupied gravitons (f 1) that satisfy a single curve after rescaling with time using the universal the condition α f = 1 [678]. Here α = L2 /R2 , exponents α and β. From Ref. [54]. grav grav P S where LP is the Planck length and RS denotes the Schwarzchild radius. A dynamical picture of the formation of such a black hole state is in 2 N scattering ing non-expanding system because of the strong Bose of gravitons at trans-Planckian energies.→ In the Regge enhancement of rates at low momenta [313]; see also limit, as first discussed in Ref. [679] and subsequently in Sec. IV D 2. Ref. [680], the scattering is dominated by the formation of N 2 soft quanta. The argument of Dvali and collaborators− was that the copious production of soft gravi- C. Highly occupied systems II: Bose gases far from equilibrium tons leads to perturbative unitarization of the scattering cross-section precisely when αgravf = 1. Although the inflaton dynamics is described by a rel- This “classicalization of amplitudes” was shown explic- ativistic field theory, the self-similar scaling behavior at itly [681] using the tree level Kawai-Lewellen-Tye (KLT) sufficiently low momenta below the screening mass scale relations [682] that express N-point tree level gravity is predicted to exhibit universal properties of a non- amplitudes in terms of sums of products of Yang-Mills relativistic system [334]. The non-equilibrium infrared N-point tree amplitudes. These results are in remark- dynamics for scalars starting from overoccupation has able agreement with computations in Lipatov’s EFT ap- been theoretically studied in great detail [26, 52, 320, proach [683]. 333, 335–344]. However important aspects of this far- The ideas of the classicalization and unitarization of from-equilibrium dynamics can be probed experimentally 2 N gravitational amplitudes are remarkably similar using Bose gases in an optical trap. For the example of to→ the discussion of the CGC EFT in Secs. sec:hadrons an interacting, non-relativistic Bose gas of density n in andIII. The BFKL results on 2 N gluon scattering are three spatial dimensions, this concerns the dilute regime likewise reproduced in the semi-classical→ CGC EFT. A (√na3 1), with a characteristic inverse coherence path forward is to employ so-called “double copy” meth- √ length given by the momentum scale Q = 16πan. Here ods that exploit a color-kinematics duality between grav- Q plays a similar role as the saturation scale for gluons in ity and QCD amplitudes [684]. Such a correspondence √ 3 the gauge theory case, and the diluteness na provides was prefigured in the high energy limit in Ref. [679] and the dimensionless coupling parameter. An overoccupied further discussed more recently [685, 686]. Bose gas then features large occupancies 1/√na3 for modes with momenta of the order of Q [334∼]. Of particular interest in our context is the “classical Universal scaling far from equilibrium associated with double copy” between classical Yang-Mills equations and non-thermal fixed points has been experimentally discov- classical gravity [687, 688]. This points to a concrete ered using different cold atom systems [54, 55]. For in- correspondence between collisions of the classical gluon stance, in Refs. [54, 662] the non-equilibrium dynamics of shock waves producing the Glasma and that of gravita- magnetic hyperfine excitations of a spin-1 Bose gas was tional shock waves that produce black holes [689]. It studied in an elongated trap, following a sudden change would also be interesting to understand whether this in the applied magnetic field as an external control pa- correspondence shares universal features at the unitar- rameter. Figure 32 exemplifies the scaling dynamics of ity limit with that of the holographic gravitational shock the measured transversal spin for three different initial waves discussed in Sec.VI. 57

E. Anomalous currents in non-equilibrium QED: cussed in Refs. [708, 709]. Recently an entanglement en- Condensed matter systems and strong laser fields tropy measure devised for proton-proton collisions at the LHC was argued to be consistent with the data; the latter Strong color fields as well as strong electromagnetic is at variance with expectations from Monte-Carlo sim- fields are an essential ingredient for the understanding of ulations [710]. In the same vein, Ref. [711] explored the the early stages of the plasma’s space-time evolution in behavior of the entanglement entropy in a holographic off-central heavy-ion collisions. Strong gauge fields lead model of heavy-ion collisions discussed in Sec.VIE and to a wealth of intriguing phenomena related to quantum found it can serve as an order parameter distinguishing anomalies, such as the chiral magnetic effect [349, 350] between the Landau (full stopping) and Bjorken (trans- described in Sec.IV. As discussed, there are strong con- parency) scenarios. nections between the transport properties of anomalous The notion of entanglement plays a key role in ten- currents in hot QCD and in strongly correlated con- sor network methods that represent quantum-many body densed matter systems, in particular Dirac and Weyl wave functions and density matrices of physical interest semimetals with applied fields [66]. yet with low enough entanglement to allow for their effi- Here we note that the similar questions could also be cient manipulation on classical computers; see Ref. [712] addressed in future strong laser field experiments that for a review. Such methods are robust in describing will be able to explore QED dynamics in extreme con- ground states and low-lying excited states in 1+1 dimen- ditions [690]. For instance, for QED field strengths ex- sions [713, 714], and considerable progress has been made ceeding the Schwinger limit for pair production, a highly in the past few years with using them for condensed- absorptive medium with quantum anomaly-induced dy- matter physics applications in 2+1 dimensions [715–719]. namical refractive properties related to the chiral mag- In the context of this review, we highlight a number of netic effect was predicted [363]. recent developments in applying tensor networks to QCD and heavy-ion collision motivated problems in (1+1)- dimensional settings ranging from the applications to F. Thermalization and entanglement gauge theories reviewed in Ref. [720] to non-equilibrium processes in interacting QFTs on a lattice [721–724]. In While the time evolution of isolated quantum systems the last cases, the aforementioned growth of entangle- is unitary, relevant observables in non-equilibrium quan- ment with time is a bottleneck preventing simulations tum field theory can approach thermal equilibrium values from reaching late times. at sufficiently late times, without the need for any coarse- Finally, entanglement entropy in holography arises as graining or reference to a reduced density operator. a Bekenstein-Hawking entropy of a special class of sur- Thermalization in quantum field theory has been demon- faces [725–728]. This discovery has led to new insight strated for scalar quantum field theories in various spatial into quantum gravity by bringing quantum information dimensions [304, 691–693] and with fermions [694, 695]; tools to the mix. An impressive result in this direction see Ref. [293] for an introductory review47. In gauge is the quantitative understanding of the time evolution theories at strong coupling, thermalization from unitary of the entropy of Hawking radiation from an evaporating dynamics was observed using holographic approaches, as black hole [63–65, 729, 730]. The cited works point to a discussed in Sec.VI. new mechanism toward resolving Hawking’s information It has been analyzed in detail how, in particular, lo- paradox [61, 62]. From the point of view of this review, cally defined quantities of isolated quantum many-body they can be thought of as including finite-Nc effects in systems can exhibit thermal features [697–699]. In such holographic studies of a class of thermalization processes time-dependent processes, entanglement entropy of spa- at very late times. tial subregions (the von Neumann entropy of spatially reduced density matrices) was seen to reach the value predicted by thermal states after exhibiting a period of IX. SUMMARY AND OUTLOOK growth; see, e.g., Refs. [700–705]. Understanding why and how this happens has been an active sub-field of re- In 1974, T.D. Lee suggested that it would be inter- search in lattice systems, quantum field theory and holog- esting to explore new phenomena by distributing a high raphy. amount of energy or high nuclear density over relatively Ref. [706, 707] applied similar considerations to a large volume [731]. We are beginning to come to grips + model of e e− collisions and pursued the idea of viewing with the richness of many-body QCD dynamics 46 years entanglement as a source of an apparent thermal behav- later owing to experimental programs in nucleus-nucleus ior seen in multiparticle production in such events, as dis- collisions in the decades since, culminating in the discovery of the quark-gluon plasma at RHIC and the LHC. As demonstrated at these colliders, the non-Abelian QGP is a nearly perfect fluid showing little resistance to pressure 47 For thermalization studies in classical-statistical field theories for gradients. given regularization, see Ref. [696]. This conclusion is a consequence of the remarkable 58 and apparently unreasonable success of relativistic vis- RHIC and the LHC will greatly enhance these prospects cous hydrodynamics in the description of the heavy-ion through both novel measurements and larger datasets data from RHIC and the LHC. However the quantitative than are currently available. The EIC will provide in- phenomenological success of hydrodynamical models also formation complementary to those of the heavy-ion ex- owes a great deal to our improved understanding of the periments to further tease out and make more precise our initial conditions for hydrodynamic evolution, in particu- understanding of the initial state. Further progress will lar, in the modeling of event-by-event fluctuations in the also depend on theoretical developments in the weak and nuclear geometry, as well as a deepening understanding strong coupling frameworks and the convergence between of how the quark-gluon matter is released in the heavy- the two when extrapolated to the realistic couplings of ion collisions and thermalizes to form the QGP. the heavy-ion experiments. With regard to the latter, comparisons of the hydrody- Computations of the properties of saturated gluons namical models to data require thermalization to occur in the CGC EFT are now at next-to-leading-order and very rapidly on time scales on the order of 3 yoctoseconds: next-to-leading log accuracy for a few processes. We ex- approximately a tenth of the lifetime of the nuclear col- pect this trend to continue, which will allow for very lision. These very short lifetimes and the nearly perfect precise extractions of the saturation scale in DIS and fluidity of the subsequent flow of the QGP suggest that proton-nucleus collisions. A more conceptual challeng- the non-equilibrium matter formed is very strongly cor- ing problem is to understand the large fluctuations in related. The quest to understand ab initio the structure the large x initial conditions that may generate very of strongly correlated QCD matter in nuclear wavefunc- anisotropic shape distributions of small x partons. As tions at high energies, and how this matter is released, we noted briefly, such studies may benefit from the uni- decoheres, and thermalizes, has motivated a large body versality between the non-linear equations that describe of work over the last couple of decades, from the incep- high energy QCD evolution and those that describe re- tion of the RHIC program to the present. action diffusion processes in statistical mechanics. Strongly correlated QCD matter can arise either in In the description of the Glasma, a straightforward but weak coupling when the occupancies of the constituents technically challenging problem is to extend several of the are very large or in strong coupling. Further, since the computations in fixed box geometries to the more realis- coupling runs towards strong coupling as the system tic longitudinally expanding case. A more difficult chal- evolves, both weak and strong couplings may be realized lenge is to implement fully quantum contributions be- in the fluid. In this review, we summarized the theoret- yond the classical-statistical approximation. While there ical ideas and techniques in both strong and weak cou- is considerable insight gained from ongoing studies of pling frameworks that address the thermalization process scalar field theories in this regard, further progress will in heavy-ion collisions. require additional conceptual breakthroughs. A notewor- We emphasized the emergence of attractors in both the thy feature of the overoccupied Glasma is the emergence weak coupling EFT and holographic approaches that may of infrared structures that may have non-trivial topo- be universal across a wide range of energy scales. We also logical features [732]. This may be universal to other noted concomitantly the very concrete interdisciplinary many-body systems, leading to novel potential synergies connections of strongly correlated QCD (and QCD-like) in addition to those discussed in this review. matter off-equilibrium to dynamical features of phenom- Recent numerical simulations using QCD effective ki- ena ranging from pre-heating in inflationary cosmology netic theory have painted a detailed picture of the dif- to pair-production in laser induced strong QED fields to ferent equilibration stages in longitudinally expanding, to non-equilibrium dynamics in ultracold atomic gases. albeit homogeneous, QCD matter. However the kinetic In particular, we discussed an intriguing universality description of inhomogeneous systems with rapid radial in the non-thermal attractor discovered in simulations expansion needs further development. This is especially of overoccupied expanding Glasma to that discovered important for studies of collisions of light nuclei or in in identically prepared simulations of the self-interacting proton-nucleus collisions, where tantalizing signals of col- scalar fields that model the ultracold systems. Remark- lective behavior have been seen. It will be interesting ably, cold atom experiments have discovered such a non- within this framework to understand whether a unified thermal attractor, albeit with a different geometry than many-body description emerges that smoothly interpo- that of a heavy-ion collision. This opens up the excit- lates from a few parton scatterings in the smallest col- ing prospect of extending the program underway of the lision systems to the emergent fluid-like behavior in the “tabletop engineering” of ultracold atom systems as ana- largest systems. log quantum simulators of the ground state properties of On the more formal side, computations of various gauge theories to uncover far-from-equilibrium properties transport properties of the QGP beyond leading order of non-Abelian gauge theories. have higher order corrections that are large for all but We also discussed the signatures for QCD matter off- extremely small values of the coupling constant. Finite equilibrium and the challenges of disentangling these temperature resummation techniques may help improve from contributions at later stages of the heavy-ion col- the convergence of the perturbative expansion. A po- lision. Ongoing and near-term experiments at both tential path forward is to combine a non-perturbative 59 description of the infrared sector with kinetic theory in Löhrer,Mark Mace, David Mateos, Aleksandr Mikheev, the UV. Guy Moore, Alfred Mueller, Niklas Mueller, Swagato A key part of our review was devoted to developments Mukherjee, Larry McLerran, Rob Myers, Yasushi Nara, in holographic approaches to off-equilibrium dynamics in Markus Oberthaler, Robert Ott, Jean-Fran¸coisPaquet, QCD like theories. An important discovery is that the Monica Pate, Asier PiñeiroOrioli, Jan M. Pawlowski, hydrodynamic gradient expansion is an asymptotic se- Robi Peschanski, Rob Pisarski, Maximilian Prüfer,Ana ries, which allows one to view the applicability of hydro- Raclariu, Klaus Reygers, Paul Romatschke, Alexan- dynamics through the emergent universal behavior of a der Rothkopf, Kaushik Roy, Björn Schenke, Sören hydrodynamic attractor. Schlichting, JörgSchmiedmayer, Julien Serreau, Dénes An open problem is the existence of hydrodynamic at- Sexty, Linda Shen, Vladimir Skokov, Michal Spalin- tractors for flows with transverse expansion and/or bro- ski, Daniel Spitz, Andy Strominger, Viktor Svensson, ken conformal symmetry. It would be very interesting Naoto Tanji, Derek Teaney, Robin Törnkvist,Prithwish to make a clear-cut statement as to what extent these Tribedy, Wilke van der Schee, Benjamin Wallisch, Qun phenomena appear in a tracktable manner outside ide- Wang, Christof Wetterich, Paul Wiesemeyer, Przemek alizations of the geometry of ultrarelativistic heavy-ion Witaszczyk, Larry Yaffe, Yi Yin and Torsten Zache. collisions or highly-symmetric cosmologies. Another im- We would like to thank Paul Romatschke for corre- portant future direction is to address collisions in holo- spondence about previous work [30] and for providing graphic models that incorporate confinement following us with the numerical data needed for Fig. 28, Björn recent promising work in this direction. Not least, it Schenke for the plot in Fig.4, which was adapted from would be interesting to reconsider expanding plasma se- Ref. [174] and Wilke van der Schee for the plot in Fig. 29, tups and, more broadly, thermalization at strong cou- which was adapted from Ref. [469]. pling in the context of the Gauss-Bonnet gravity dis- We also thank Kirill Boguslavski, Jacopo Ghighlieri, cussed in Section VI F 2. First steps in this direction Michal Spalinski, BjörnSchenke, Viktor Svensson, and relied on treating the Gauss-Bonnet term as a small cor- Wilke van der Schee for reading the manuscript and of- rection. Going beyond this regime, which is challenging fering their suggestions. from many perspectives, can reveal genuinely new effects The work of JB and AM is part of and supported by the in holographic setups like non-thermal fixed points dis- Deutsche Forschungsgemeinschaft [(DFG) German Re- cussed in SectionIVC. Finally, an important open ques- search Foundation] Collaborative Research Center “SFB tion in holography is to understand whether long-range 1225 (ISOQUANT).” JB’s work is also partly supported “ridge-like” correlations can naturally arise at strong cou- by the DFG under Germany’s Excellence Strategy EXC pling and whether they can survive until late time. 2181/1 - 390900948 (the Heidelberg STRUCTURES Ex- cellence Cluster), by DFG Grant No. BE 2795/4-1, and by the Bundesministerium fürBildung und Forschung [ ACKNOWLEDGMENTS (BMBF) German Federal Ministry of Education and Re- search] Grant No. 05P18VHFCA. We have all benefited greatly from the combined wis- M.P.H and the Gravity, Quantum Fields and Infor- dom on this topic of our collaborators and colleagues mation group at AEI are generously supported by the over the years. We thank, in particular, Gert Aarts, Alexander von Humboldt Foundation and the Federal Peter Arnold, Rudolf Baier, Adam Ball, Guillaume Ministry for Education and Research through the Sofja Beuf, Jean-Paul Blaizot, Kirill Boguslavski, Szabolcs Kovalevskaja Award. Borsányi, Alex Buchel, Jorge Casalderrey-Solana, Paul RV is supported by the U.S. Department of Energy, Chesler, Jian Deng, Adrian Dumitru, Kevin Dusling, Office of Science, Office of Nuclear Physics, under Con- Gia Dvali, Thomas Epelbaum, Sebastian Erne, Stefan tract No. DE- SC0012704, and within the framework Floerchinger, Wojciech Florkowski, Charles Gale, Os- of the Beam Energy Scan Theory (BEST) DOE Topi- car Garcia-Montero, Thomas Gasenzer, Daniil Gelfand, cal Collaboration. He also acknowledges the Humboldt Fran¸coisGelis, Jacopo Ghighlieri, Giuliano Giacalone, Foundation for its generous support through a Humboldt Philipp Hauke, Florian Hebenstreit, Edmond Iancu, Ja- Prize, ITP Heidelberg for their kind hospitality, and the mal Jalilian-Marian, Romuald Janik, Fred Jendrzejew- DFG Collaborative Research Centre “SFB 1225 (ISO- ski, Sangyong Jeon, Valentin Kasper, Dmitri Kharzeev, QUANT)” for supporting his research collaboration with Alex Krasnitz, Aleksi Kurkela, Tuomas Lappi, Nicole Heidelberg University.

[1] I. Arsene et al. (BRAHMS), Nucl. Phys. A757, 1 (2007), arXiv:nucl-ex/0610037 [nucl-ex]. (2005), arXiv:nucl-ex/0410020 [nucl-ex]. [4] J. Adams et al. (STAR), Nucl. Phys. A 757, 102 (2005), [2] K. Adcox et al. (PHENIX), Nucl. Phys. A 757, 184 arXiv:nucl-ex/0501009. (2005), arXiv:nucl-ex/0410003. [5] B. M¨uller,J. Schukraft, and B. Wyslouch, Ann. Rev. [3] B. Alver et al. (PHOBOS), Phys. Rev. Lett. 98, 242302 Nucl. Part. Sci. 62, 361 (2012), arXiv:1202.3233 [hep- 60

ex]. arXiv:hep-ph/0611157 [hep-ph]. [6] G. Roland, K. Safarik, and P. Steinberg, Prog. Part. [34] S. Gubser, I. R. Klebanov, and A. M. Polyakov, Phys. Nucl. Phys. 77, 70 (2014). Lett. B 428, 105 (1998), arXiv:hep-th/9802109. [7] P. Foka and M. g. A. Janik, Rev. Phys. 1, 172 (2016), [35] E. Witten, Adv. Theor. Math. Phys. 2, 253 (1998), arXiv:1702.07231 [hep-ex]. arXiv:hep-th/9802150. [8] P. Foka and M. g. A. Janik, Rev. Phys. 1, 154 (2016), [36] L. V. Gribov, E. M. Levin, and M. G. Ryskin, Phys. arXiv:1702.07233 [hep-ex]. Rept. 100, 1 (1983). [9] J. Adam et al. (ALICE), Phys. Lett. B 754, 235 (2016), [37] F. Gelis, E. Iancu, J. Jalilian-Marian, and R. Venu- arXiv:1509.07324 [nucl-ex]. gopalan, Ann. Rev. Nucl. Part. Sci. 60, 463 (2010), [10] A. Bazavov et al. (HotQCD), Phys. Lett. B795, 15 arXiv:1002.0333 [hep-ph]. (2019), arXiv:1812.08235 [hep-lat]. [38] Y. V. Kovchegov and E. Levin, Camb. Monogr. Part. [11] E. V. Shuryak, Phys. Rept. 61, 71 (1980). Phys. Nucl. Phys. Cosmol. 33, 1 (2012). [12] P. Romatschke and U. Romatschke, Relativistic Fluid [39] J. Jalilian-Marian, A. Kovner, and H. Weigert, Phys. Dynamics In and Out of Equilibrium, Cambridge Mono- Rev. D59, 014015 (1998), arXiv:hep-ph/9709432 [hep- graphs on Mathematical Physics (Cambridge University ph]. Press, 2019) arXiv:1712.05815 [nucl-th]. [40] E. Iancu, A. Leonidov, and L. D. McLerran, Nucl. Phys. [13] G. Policastro, D. T. Son, and A. O. Starinets, Phys. A692, 583 (2001), arXiv:hep-ph/0011241 [hep-ph]. Rev. Lett. 87, 081601 (2001), arXiv:hep-th/0104066. [41] P. Romatschke and R. Venugopalan, Phys. Rev. Lett. [14] A. Buchel and J. T. Liu, Phys. Rev. Lett. 93, 090602 96, 062302 (2006), arXiv:hep-ph/0510121 [hep-ph]. (2004), arXiv:hep-th/0311175. [42] G. Giacalone, A. Mazeliauskas, and S. Schlichting, [15] P. Kovtun, D. T. Son, and A. O. Starinets, Phys. Rev. Phys. Rev. Lett. 123, 262301 (2019), arXiv:1908.02866 Lett. 94, 111601 (2005), arXiv:hep-th/0405231 [hep-th]. [hep-ph]. [16] N. Iqbal and H. Liu, Phys. Rev. D 79, 025023 (2009), [43] A. Kurkela and A. Mazeliauskas, Phys. Rev. Lett. 122, arXiv:0809.3808 [hep-th]. 142301 (2019), arXiv:1811.03040 [hep-ph]. [17] J. M. Maldacena, Int.J.Theor.Phys. 38 1113-1133 and [44] A. Kurkela, A. Mazeliauskas, J.-F. Paquet, S. Schlicht- also Adv.Theor.Math.Phys. 2 231-252 (1999 and 1998), ing, and D. Teaney, Phys. Rev. Lett. 122, 122302 arXiv:hep-th/9711200. (2019), arXiv:1805.01604 [hep-ph]. [18] S. Mrówczyński,B. Schenke, and M. Strickland, Phys. [45] P. M. Chesler and L. G. Yaffe, Phys. Rev. D82, 026006 Rept. 682, 1 (2017), arXiv:1603.08946 [hep-ph]. (2010), arXiv:0906.4426 [hep-th]. [19] W. Busza, K. Rajagopal, and W. van der Schee, Ann. [46] P. M. Chesler and L. G. Yaffe, Phys. Rev. Lett. 106, Rev. Nucl. Part. Sci. 68, 339 (2018), arXiv:1802.04801 021601 (2011), arXiv:1011.3562 [hep-th]. [hep-ph]. [47] M. P. Heller, R. A. Janik, and P. Witaszczyk, Phys. [20] S. Schlichting and D. Teaney, Ann. Rev. Nucl. Part. Sci. Rev. Lett. 108, 201602 (2012), arXiv:1103.3452 [hep- 69, 447 (2019), arXiv:1908.02113 [nucl-th]. th]. [21] L. Van Hove and S. Pokorski, Nucl. Phys. B86, 243 [48] J. Casalderrey-Solana, H. Liu, D. Mateos, K. Rajagopal, (1975). and U. A. Wiedemann, Gauge/String Duality, Hot QCD [22] J. D. Bjorken, Phys. Rev. D27, 140 (1983). and Heavy Ion Collisions (Cambridge University Press, [23] F. Gelis, A. M. Stasto, and R. Venugopalan, Eur. Phys. 2014) arXiv:1101.0618 [hep-th]. J. C48, 489 (2006), arXiv:hep-ph/0605087 [hep-ph]. [49] D. E. Kharzeev, L. D. McLerran, and H. J. Warringa, [24] J. Casalderrey-Solana, M. P. Heller, D. Mateos, and Nucl. Phys. A803, 227 (2008), arXiv:0711.0950 [hep- W. van der Schee, Phys. Rev. Lett. 111, 181601 (2013), ph]. arXiv:1305.4919 [hep-th]. [50] R. Micha and I. I. Tkachev, Phys. Rev. D70, 043538 [25] K. J. Gon¸calves, A. V. Giannini, D. D. Chinellato, (2004), arXiv:hep-ph/0403101 [hep-ph]. and G. Torrieri, Phys. Rev. C100, 054901 (2019), [51] V. Zakharov, V. L’vov, and G. Falkovich, Kolmogorov arXiv:1906.00947 [nucl-th]. Spectra of Turbulence I: Wave Turbulence, Springer Se- [26] J. Berges, A. Rothkopf, and J. Schmidt, Phys. Rev. ries in Nonlinear Dynamics (Springer Berlin Heidelberg, Lett. 101, 041603 (2008), arXiv:0803.0131 [hep-ph]. 2012). [27] J. Berges, K. Boguslavski, S. Schlichting, and [52] J. Berges, K. Boguslavski, S. Schlichting, and R. Venugopalan, Phys. Rev. D89, 074011 (2014), R. Venugopalan, Phys. Rev. Lett. 114, 061601 (2015), arXiv:1303.5650 [hep-ph]. arXiv:1408.1670 [hep-ph]. [28] J. Berges, K. Boguslavski, S. Schlichting, and [53] P. Hohenberg and B. Halperin, Rev. Mod. Phys. 49, 435 R. Venugopalan, Phys. Rev. D89, 114007 (2014), (1977). arXiv:1311.3005 [hep-ph]. [54] M. Prüfer,P. Kunkel, H. Strobel, S. Lannig, D. Lin- [29] M. P. Heller and M. Spaliński, Phys. Rev. Lett. 115, nemann, C.-M. Schmied, J. Berges, T. Gasenzer, 072501 (2015), arXiv:1503.07514 [hep-th]. and M. K. Oberthaler, Nature 563, 217 (2018), [30] P. Romatschke, Phys. Rev. Lett. 120, 012301 (2018), arXiv:1805.11881 [cond-mat.quant-gas]. arXiv:1704.08699 [hep-th]. [55] S. Erne, R. Bücker, T. Gasenzer, J. Berges, [31] J. Schukraft, A. Timmins, and S. A. Voloshin, Phys. and J. Schmiedmayer, Nature 563, 225 (2018), Lett. B719, 394 (2013), arXiv:1208.4563 [nucl-ex]. arXiv:1805.12310 [cond-mat.quant-gas]. [32] T. Lappi and L. McLerran, Nucl. Phys. A772, 200 [56] J. A. P. Glidden, C. Eigen, L. H. Dogra, T. A. (2006), arXiv:hep-ph/0602189 [hep-ph]. Hilker, R. P. Smith, and Z. Hadzibabic, (2020), [33] F. Gelis and R. Venugopalan, Theoretical physics. Pro- arXiv:2006.01118 [cond-mat.quant-gas]. ceedings, 46th Cracow School, Zakopane, Poland, May [57] J. Eisert, M. Friesdorf, and C. Gogolin, Nature Phys. 27-June 5, 2006, Acta Phys. Polon. B37, 3253 (2006), 11, 124 (2015), arXiv:1408.5148 [quant-ph]. 61

[58] N. Lashkari, D. Stanford, M. Hastings, T. Osborne, and (1976). P. Hayden, JHEP 04, 022 (2013), arXiv:1111.6580 [hep- [88] F. Wilczek, Particles and nuclei. Proceedings, 15th th]. International Conference, PANIC ’99, Uppsala, Swe- [59] J. Maldacena, S. H. Shenker, and D. Stanford, JHEP den, June 10-16, 1999, Nucl. Phys. A663, 3 (2000), 08, 106 (2016), arXiv:1503.01409 [hep-th]. arXiv:hep-ph/9907340 [hep-ph]. [60] S. Hawking, Nature 248, 30 (1974). [89] H.-W. Lin et al., Prog. Part. Nucl. Phys. 100, 107 [61] S. Hawking, Phys. Rev. D 14, 2460 (1976). (2018), arXiv:1711.07916 [hep-ph]. [62] D. N. Page, Phys. Rev. Lett. 71, 3743 (1993), arXiv:hep- [90] W. Detmold, R. G. Edwards, J. J. Dudek, M. En- th/9306083. gelhardt, H.-W. Lin, S. Meinel, K. Orginos, and [63] G. Penington, (2019), arXiv:1905.08255 [hep-th]. P. Shanahan (USQCD), Eur. Phys. J. A55, 193 (2019), [64] A. Almheiri, N. Engelhardt, D. Marolf, and H. Max- arXiv:1904.09512 [hep-lat]. field, JHEP 12, 063 (2019), arXiv:1905.08762 [hep-th]. [91] A. Alexandru, G. Basar, P. F. Bedaque, and [65] A. Almheiri, R. Mahajan, J. Maldacena, and Y. Zhao, G. W. Ridgway, Phys. Rev. D95, 114501 (2017), JHEP 03, 149 (2020), arXiv:1908.10996 [hep-th]. arXiv:1704.06404 [hep-lat]. [66] Q. Li and D. E. Kharzeev, Proceedings, 25th Interna- [92] J. Preskill, Proceedings, 36th International Symposium tional Conference on Ultra-Relativistic Nucleus-Nucleus on Lattice Field Theory (Lattice 2018): East Lans- Collisions (Quark Matter 2015): Kobe, Japan, Septem- ing, MI, United States, July 22-28, 2018, PoS LAT- ber 27-October 3, 2015, Nucl. Phys. A956, 107 (2016). TICE2018, 024 (2018), arXiv:1811.10085 [hep-lat]. [67] Y. Akamatsu, D. Teaney, F. Yan, and Y. Yin, Phys. [93] J. C. Collins, D. E. Soper, and G. F. Sterman, Adv. Rev. C100, 044901 (2019), arXiv:1811.05081 [nucl-th]. Ser. Direct. High Energy Phys. 5, 1 (1989), arXiv:hep- [68] A. Bzdak, S. Esumi, V. Koch, J. Liao, M. Stephanov, ph/0409313 [hep-ph]. and N. Xu, Phys. Rept. 853, 1 (2020), arXiv:1906.00936 [94] J. D. Bjorken and E. A. Paschos, Phys. Rev. 185, 1975 [nucl-th]. (1969). [69] M. Bluhm et al., (2020), arXiv:2001.08831 [nucl-th]. [95] I. Abt et al. (H1), Nucl. Phys. B407, 515 (1993). [70] Y. Akamatsu, A. Mazeliauskas, and D. Teaney, Phys. [96] T. Ahmed et al. (H1), Nucl. Phys. B439, 471 (1995), Rev. C95, 014909 (2017), arXiv:1606.07742 [nucl-th]. arXiv:hep-ex/9503001 [hep-ex]. [71] X. An, G. Basar, M. Stephanov, and H.-U. Yee, Phys. [97] M. Derrick et al. (ZEUS), Phys. Lett. B316, 412 (1993). Rev. C100, 024910 (2019), arXiv:1902.09517 [hep-th]. [98] M. Derrick et al. (ZEUS), Z. Phys. C65, 379 (1995). [72] J. Polchinski and M. J. Strassler, JHEP 05, 012 (2003), [99] A. D. Martin, W. J. Stirling, and R. G. Roberts, Phys. arXiv:hep-th/0209211. Rev. D50, 6734 (1994), arXiv:hep-ph/9406315 [hep-ph]. [73] Y. Hatta, E. Iancu, and A. Mueller, JHEP 01, 063 [100] H. L. Lai, J. Botts, J. Huston, J. G. Morfin, J. F. Owens, (2008), arXiv:0710.5297 [hep-th]. J.-w. Qiu, W. K. Tung, and H. Weerts, Phys. Rev. D51, [74] E. Shuryak and I. Zahed, Annals Phys. 396, 1 (2018), 4763 (1995), arXiv:hep-ph/9410404 [hep-ph]. arXiv:1707.01885 [hep-ph]. [101] V. N. Gribov and L. N. Lipatov, Sov. J. Nucl. Phys. 15, [75] C. Herzog, A. Karch, P. Kovtun, C. Kozcaz, and 438 (1972), [Yad. Fiz.15,781(1972)]. L. Yaffe, JHEP 07, 013 (2006), arXiv:hep-th/0605158. [102] L. N. Lipatov, Sov. J. Nucl. Phys. 20, 94 (1975), [Yad. [76] S. S. Gubser, Phys. Rev. D 74, 126005 (2006), Fiz.20,181(1974)]. arXiv:hep-th/0605182. [103] G. Altarelli and G. Parisi, Nucl. Phys. B126, 298 [77] P. M. Chesler, K. Jensen, and A. Karch, Phys. Rev. D (1977). 79, 025021 (2009), arXiv:0804.3110 [hep-th]. [104] Y. L. Dokshitzer, Sov. Phys. JETP 46, 641 (1977), [Zh. [78] P. M. Chesler, K. Jensen, A. Karch, and L. G. Yaffe, Eksp. Teor. Fiz.73,1216(1977)]. Phys. Rev. D 79, 125015 (2009), arXiv:0810.1985 [hep- [105] N. Nakanishi and K. Yamawaki, Nucl. Phys. B122, 15 th]. (1977). [79] P. M. Chesler, M. Lekaveckas, and K. Rajagopal, JHEP [106] J. Collins, (2018), arXiv:1801.03960 [hep-ph]. 10, 013 (2013), arXiv:1306.0564 [hep-ph]. [107] A. L. Fitzpatrick, J. Kaplan, E. Katz, L. G. Vitale, and [80] D. T. Son and A. O. Starinets, JHEP 09, 042 (2002), M. T. Walters, JHEP 08, 120 (2018), arXiv:1803.10793 arXiv:hep-th/0205051. [hep-th]. [81] C. Herzog and D. Son, JHEP 03, 046 (2003), arXiv:hep- [108] S. J. Brodsky, H.-C. Pauli, and S. S. Pinsky, Phys. th/0212072. Rept. 301, 299 (1998), arXiv:hep-ph/9705477 [hep-ph]. [82] P. K. Kovtun and A. O. Starinets, Phys. Rev. D72, [109] A. H. Mueller and J.-w. Qiu, Nucl. Phys. B268, 427 086009 (2005), arXiv:hep-th/0506184 [hep-th]. (1986). [83] O. DeWolfe, S. S. Gubser, C. Rosen, and D. Teaney, [110] E. Iancu and R. Venugopalan, in Quark-gluon plasma Prog. Part. Nucl. Phys. 75, 86 (2014), arXiv:1304.7794 4 , edited by R. C. Hwa and X.-N. Wang (2003) pp. [hep-th]. 249–3363, arXiv:hep-ph/0303204 [hep-ph]. [84] P. M. Chesler and W. van der Schee, Int. J. Mod. Phys. [111] L. D. McLerran and R. Venugopalan, Phys. Rev. D49, E 24, 1530011 (2015), arXiv:1501.04952 [nucl-th]. 2233 (1994), arXiv:hep-ph/9309289 [hep-ph]. [85] M. P. Heller, Acta Phys. Polon. B 47, 2581 (2016), [112] L. D. McLerran and R. Venugopalan, Phys. Rev. D49, arXiv:1610.02023 [hep-th]. 3352 (1994), arXiv:hep-ph/9311205 [hep-ph]. [86] W. Florkowski, M. P. Heller, and M. Spalinski, Rept. [113] L. D. McLerran and R. Venugopalan, Phys. Rev. D50, Prog. Phys. 81, 046001 (2018), arXiv:1707.02282 [hep- 2225 (1994), arXiv:hep-ph/9402335 [hep-ph]. ph]. [114] S. Jeon and R. Venugopalan, Phys. Rev. D70, 105012 [87] J. D. Bjorken, Intnl. Summer Inst. in Theoretical (2004), arXiv:hep-ph/0406169 [hep-ph]. Physics on Current Induced Reactions Hamburg, Ger- [115] J. Jalilian-Marian, S. Jeon, and R. Venugopalan, Phys. many, September 15-26, 1975, Lect. Notes Phys. 56, 93 Rev. D63, 036004 (2001), arXiv:hep-ph/0003070 [hep- 62

ph]. [143] H. Fujii, Nucl. Phys. A709, 236 (2002), arXiv:nucl- [116] Y. V. Kovchegov, Phys. Rev. D60, 034008 (1999), th/0205066 [nucl-th]. arXiv:hep-ph/9901281 [hep-ph]. [144] J. P. Blaizot, F. Gelis, and R. Venugopalan, Nucl. Phys. [117] S. Jeon and R. Venugopalan, Phys. Rev. D71, 125003 A743, 57 (2004), arXiv:hep-ph/0402257 [hep-ph]. (2005), arXiv:hep-ph/0503219 [hep-ph]. [145] F. Dominguez, C. Marquet, B.-W. Xiao, and F. Yuan, [118] A. Dumitru and E. Petreska, Nucl. Phys. A879, 59 Phys. Rev. D83, 105005 (2011), arXiv:1101.0715 [hep- (2012), arXiv:1112.4760 [hep-ph]. ph]. [119] J. Jalilian-Marian, A. Kovner, L. D. McLerran, and [146] K. Dusling, M. Mace, and R. Venugopalan, Phys. Rev. H. Weigert, Phys. Rev. D55, 5414 (1997), arXiv:hep- D97, 016014 (2018), arXiv:1706.06260 [hep-ph]. ph/9606337 [hep-ph]. [147] K. Fukushima and Y. Hidaka, JHEP 11, 114 (2017), [120] J. D. Jackson, Classical Electrodynamics (Wiley, 1998). arXiv:1708.03051 [hep-ph]. [121] T. Lappi, Eur. Phys. J. C55, 285 (2008), [148] E. Iancu, K. Itakura, and L. McLerran, Nucl. Phys. arXiv:0711.3039 [hep-ph]. A708, 327 (2002), arXiv:hep-ph/0203137 [hep-ph]. [122] M. Pate, A.-M. Raclariu, and A. Strominger, Phys. [149] A. M. Stasto, K. J. Golec-Biernat, and J. Kwiecinski, Rev. Lett. 119, 261602 (2017), arXiv:1707.08016 [hep- Phys. Rev. Lett. 86, 596 (2001), arXiv:hep-ph/0007192 th]. [hep-ph]. [123] A. Ball, M. Pate, A.-M. Raclariu, A. Strominger, [150] L. Frankfurt, V. Guzey, and M. Strikman, Phys. Rept. and R. Venugopalan, Annals Phys. 407, 15 (2019), 512, 255 (2012), arXiv:1106.2091 [hep-ph]. arXiv:1805.12224 [hep-ph]. [151] K. Dusling, F. Gelis, T. Lappi, and R. Venugopalan, [124] A. Strominger and A. Zhiboedov, JHEP 01, 086 (2016), Nucl. Phys. A836, 159 (2010), arXiv:0911.2720 [hep- arXiv:1411.5745 [hep-th]. ph]. [125] A. Strominger, (2017), arXiv:1703.05448 [hep-th]. [152] S. Munier and R. B. Peschanski, Phys. Rev. Lett. 91, [126] L. Bieri and D. Garfinkle, Class. Quant. Grav. 30, 232001 (2003), arXiv:hep-ph/0309177 [hep-ph]. 195009 (2013), arXiv:1307.5098 [gr-qc]. [153] A. H. Mueller and S. Munier, Phys. Rev. D98, 034021 [127] D. Kapec, M. Perry, A.-M. Raclariu, and (2018), arXiv:1805.02847 [hep-ph]. A. Strominger, Phys. Rev. D96, 085002 (2017), [154] A. H. Mueller and S. Munier, Phys. Rev. Lett. 121, arXiv:1705.04311 [hep-th]. 082001 (2018), arXiv:1805.09417 [hep-ph]. [128] J. Jalilian-Marian, A. Kovner, A. Leonidov, and [155] A. H. Mueller and D. N. Triantafyllopoulos, Nucl. Phys. H. Weigert, Phys. Rev. D59, 014014 (1998), arXiv:hep- B640, 331 (2002), arXiv:hep-ph/0205167 [hep-ph]. ph/9706377 [hep-ph]. [156] G. Beuf, (2010), arXiv:1008.0498 [hep-ph]. [129] H. Weigert, Nucl. Phys. A703, 823 (2002), arXiv:hep- [157] D. N. Triantafyllopoulos, QCD, low x physics, satu- ph/0004044 [hep-ph]. ration and diffraction. Proceedings, School, Copanello, [130] I. Balitsky, Nucl. Phys. B463, 99 (1996), arXiv:hep- Italy, July 1-14, 2007, Acta Phys. Polon. B39, 2287 ph/9509348 [hep-ph]. (2008), arXiv:0804.1918 [hep-ph]. [131] E. A. Kuraev, L. N. Lipatov, and V. S. Fadin, [158] I. Balitsky and G. A. Chirilli, Phys. Rev. D88, 111501 Sov. Phys. JETP 45, 199 (1977), [Zh. Eksp. Teor. (2013), arXiv:1309.7644 [hep-ph]. Fiz.72,377(1977)]. [159] A. Kovner, M. Lublinsky, and Y. Mulian, Phys. Rev. [132] I. I. Balitsky and L. N. Lipatov, Sov. J. Nucl. Phys. 28, D89, 061704 (2014), arXiv:1310.0378 [hep-ph]. 822 (1978), [Yad. Fiz.28,1597(1978)]. [160] A. Kovner, M. Lublinsky, and Y. Mulian, JHEP 08, [133] E. Iancu and L. D. McLerran, Phys. Lett. B510, 145 114 (2014), arXiv:1405.0418 [hep-ph]. (2001), arXiv:hep-ph/0103032 [hep-ph]. [161] I. Balitsky and A. V. Grabovsky, JHEP 01, 009 (2015), [134] M. E. Peskin and D. V. Schroeder, An Introduction to arXiv:1405.0443 [hep-ph]. quantum field theory (Addison-Wesley, Reading, USA, [162] S. Caron-Huot, JHEP 03, 036 (2018), arXiv:1501.03754 1995). [hep-ph]. [135] L. D. McLerran and R. Venugopalan, Phys. Rev. D59, [163] I. Balitsky and G. A. Chirilli, Phys. Rev. D77, 014019 094002 (1999), arXiv:hep-ph/9809427 [hep-ph]. (2008), arXiv:0710.4330 [hep-ph]. [136] R. Venugopalan, Strong interactions at low and high- [164] S. Caron-Huot and M. Herranen, JHEP 02, 058 (2018), energies. Proceedings, 39th Cracow School of Theoretical arXiv:1604.07417 [hep-ph]. Physics, Zakopane, Poland, May 29-June 8, 1999, Acta [165] G. P. Salam, JHEP 07, 019 (1998), arXiv:hep- Phys. Polon. B30, 3731 (1999), arXiv:hep-ph/9911371 ph/9806482 [hep-ph]. [hep-ph]. [166] M. Ciafaloni, D. Colferai, and G. P. Salam, Phys. Rev. [137] J. D. Bjorken, J. B. Kogut, and D. E. Soper, Phys. D60, 114036 (1999), arXiv:hep-ph/9905566 [hep-ph]. Rev. D3, 1382 (1971). [167] E. Iancu, J. D. Madrigal, A. H. Mueller, G. Soyez, and [138] A. H. Mueller, Nucl. Phys. B335, 115 (1990). D. N. Triantafyllopoulos, Phys. Lett. B744, 293 (2015), [139] J. Bartels, K. J. Golec-Biernat, and H. Kowalski, Phys. arXiv:1502.05642 [hep-ph]. Rev. D66, 014001 (2002), arXiv:hep-ph/0203258 [hep- [168] B. Ducloué, E. Iancu, A. H. Mueller, G. Soyez, ph]. and D. N. Triantafyllopoulos, JHEP 04, 081 (2019), [140] H. Kowalski and D. Teaney, Phys. Rev. D68, 114005 arXiv:1902.06637 [hep-ph]. (2003), arXiv:hep-ph/0304189 [hep-ph]. [169] J. L. Albacete, N. Armesto, A. Kovner, C. A. Salgado, [141] H. Kowalski, T. Lappi, and R. Venugopalan, Phys. Rev. and U. A. Wiedemann, Phys. Rev. Lett. 92, 082001 Lett. 100, 022303 (2008), arXiv:0705.3047 [hep-ph]. (2004), arXiv:hep-ph/0307179 [hep-ph]. [142] A. H. Rezaeian, M. Siddikov, M. Van de Klundert, [170] J. L. Albacete, N. Armesto, J. G. Milhano, C. A. Sal- and R. Venugopalan, Phys. Rev. D87, 034002 (2013), gado, and U. A. Wiedemann, Phys. Rev. D71, 014003 arXiv:1212.2974 [hep-ph]. (2005), arXiv:hep-ph/0408216 [hep-ph]. 63

[171] J. L. Albacete and Y. V. Kovchegov, Phys. Rev. D75, tional Conference on Hard and Electromagnetic Probes 125021 (2007), arXiv:0704.0612 [hep-ph]. of High-Energy Nuclear Collisions (Hard Probes 2006): [172] T. Lappi and H. Mäntysaari, Phys. Rev. D93, 094004 Asilomar, USA, June 9-16, 2006, Nucl. Phys. A782, (2016), arXiv:1601.06598 [hep-ph]. 297 (2007), [Nucl. Phys.A785,146(2007)], arXiv:hep- [173] B. Ducloué, E. Iancu, G. Soyez, and D. N. Tri- ph/0608117 [hep-ph]. antafyllopoulos, Phys. Lett. B803, 135305 (2020), [201] F. Gelis and R. Venugopalan, Nucl. Phys. A779, 177 arXiv:1912.09196 [hep-ph]. (2006), arXiv:hep-ph/0605246 [hep-ph]. [174] A. Dumitru, J. Jalilian-Marian, T. Lappi, B. Schenke, [202] F. Gelis, T. Lappi, and R. Venugopalan, Hadron and R. Venugopalan, Phys. Lett. B706, 219 (2011), physics. Proceedings, 10th International Workshop, Flo- arXiv:1108.4764 [hep-ph]. rianopolis, Brazil, April 26-31, 2007, Int. J. Mod. Phys. [175] J.-P. Blaizot, E. Iancu, and H. Weigert, Nucl. Phys. E16, 2595 (2007), arXiv:0708.0047 [hep-ph]. A713, 441 (2003), arXiv:hep-ph/0206279 [hep-ph]. [203] Y. V. Kovchegov and D. H. Rischke, Phys. Rev. C56, [176] K. Rummukainen and H. Weigert, Nucl. Phys. A739, 1084 (1997), arXiv:hep-ph/9704201 [hep-ph]. 183 (2004), arXiv:hep-ph/0309306 [hep-ph]. [204] A. Kovner, L. D. McLerran, and H. Weigert, Phys. Rev. [177] A. Kovner and M. Lublinsky, Phys. Rev. D83, 034017 D52, 6231 (1995), arXiv:hep-ph/9502289 [hep-ph]. (2011), arXiv:1012.3398 [hep-ph]. [205] A. Kovner, L. D. McLerran, and H. Weigert, Phys. Rev. [178] K. Dusling, M. Mace, and R. Venugopalan, Phys. Rev. D52, 3809 (1995), arXiv:hep-ph/9505320 [hep-ph]. Lett. 120, 042002 (2018), arXiv:1705.00745 [hep-ph]. [206] M. Gyulassy and L. D. McLerran, Phys. Rev. C56, 2219 [179] T. Lappi and A. Ramnath, Phys. Rev. D100, 054003 (1997), arXiv:nucl-th/9704034 [nucl-th]. (2019), arXiv:1904.00782 [hep-ph]. [207] J. F. Gunion and G. Bertsch, Phys. Rev. D25, 746 [180] T. Lappi and H. Mäntysaari, Eur. Phys. J. C73, 2307 (1982). (2013), arXiv:1212.4825 [hep-ph]. [208] D. Kharzeev, Y. V. Kovchegov, and K. Tuchin, Phys. [181] J. A. M. Vermaseren, A. Vogt, and S. Moch, Nucl. Rev. D68, 094013 (2003), arXiv:hep-ph/0307037 [hep- Phys. B724, 3 (2005), arXiv:hep-ph/0504242 [hep-ph]. ph]. [182] I. Balitsky and G. A. Chirilli, Phys. Rev. D87, 014013 [209] J. P. Blaizot, F. Gelis, and R. Venugopalan, Nucl. Phys. (2013), arXiv:1207.3844 [hep-ph]. A743, 13 (2004), arXiv:hep-ph/0402256 [hep-ph]. [183] R. Boussarie, A. V. Grabovsky, D. Yu. Ivanov, L. Szy- [210] J. Blaizot and A. H. Mueller, Nucl. Phys. B 289, 847 manowski, and S. Wallon, Phys. Rev. Lett. 119, 072002 (1987). (2017), arXiv:1612.08026 [hep-ph]. [211] Y. V. Kovchegov and A. H. Mueller, Nucl. Phys. B529, [184] K. Roy and R. Venugopalan, (2019), arXiv:1911.04519 451 (1998), arXiv:hep-ph/9802440 [hep-ph]. [hep-ph]. [212] A. Dumitru and L. D. McLerran, Nucl. Phys. A700, [185] K. Roy and R. Venugopalan, Phys. Rev. D101, 034028 492 (2002), arXiv:hep-ph/0105268 [hep-ph]. (2020), arXiv:1911.04530 [hep-ph]. [213] A. Krasnitz and R. Venugopalan, in Proceedings, 3rd [186] A. Accardi et al., Eur. Phys. J. A52, 268 (2016), International Conference on Physics and astrophysics arXiv:1212.1701 [nucl-ex]. of quark-gluon plasma (ICPA-QGP ’97): Jaipur, In- [187] E. C. Aschenauer, S. Fazio, J. H. Lee, H. Mantysaari, dia, March 17-21, 1997 (1998) pp. 560–569, arXiv:hep- B. S. Page, B. Schenke, T. Ullrich, R. Venugopalan, ph/9706329 [hep-ph]. and P. Zurita, Rept. Prog. Phys. 82, 024301 (2019), [214] A. Krasnitz and R. Venugopalan, Nucl. Phys. B557, arXiv:1708.01527 [nucl-ex]. 237 (1999), arXiv:hep-ph/9809433 [hep-ph]. [188] S. S. Gubser, Phys. Rev. D84, 085024 (2011), [215] A. Krasnitz and R. Venugopalan, Phys. Rev. Lett. 84, arXiv:1102.4040 [hep-th]. 4309 (2000), arXiv:hep-ph/9909203 [hep-ph]. [189] A. Kovner and U. A. Wiedemann, Phys. Lett. B551, [216] A. Krasnitz and R. Venugopalan, Phys. Rev. Lett. 86, 311 (2003), arXiv:hep-ph/0207335 [hep-ph]. 1717 (2001), arXiv:hep-ph/0007108 [hep-ph]. [190] F. Gelis and N. Tanji, Prog. Part. Nucl. Phys. 87, 1 [217] A. Krasnitz, Y. Nara, and R. Venugopalan, Phys. Rev. (2016), arXiv:1510.05451 [hep-ph]. Lett. 87, 192302 (2001), arXiv:hep-ph/0108092 [hep- [191] M. K. Parikh and F. Wilczek, Phys. Rev. Lett. 85, 5042 ph]. (2000), arXiv:hep-th/9907001 [hep-th]. [218] T. Lappi, Phys. Rev. C67, 054903 (2003), arXiv:hep- [192] S. Jeon, Annals Phys. 340, 119 (2014), arXiv:1308.0263 ph/0303076 [hep-ph]. [hep-th]. [219] A. Krasnitz, Y. Nara, and R. Venugopalan, Nucl. Phys. [193] B. Wu and Y. V. Kovchegov, JHEP 03, 158 (2018), A727, 427 (2003), arXiv:hep-ph/0305112 [hep-ph]. arXiv:1709.02866 [hep-ph]. [220] J. B. Kogut and L. Susskind, Phys. Rev. D11, 395 [194] A. V. Leonidov and A. A. Radovskaya, Eur. Phys. J. (1975). C79, 55 (2019), arXiv:1809.06812 [hep-ph]. [221] A. Krasnitz, Y. Nara, and R. Venugopalan, Phys. Lett. [195] J. Ghiglieri, A. Kurkela, M. Strickland, and A. Vuori- B554, 21 (2003), arXiv:hep-ph/0204361 [hep-ph]. nen, (2020), arXiv:2002.10188 [hep-ph]. [222] A. Krasnitz, Y. Nara, and R. Venugopalan, Nucl. Phys. [196] J. S. Schwinger, J. Math. Phys. 2, 407 (1961). A717, 268 (2003), arXiv:hep-ph/0209269 [hep-ph]. [197] L. V. Keldysh, Zh. Eksp. Teor. Fiz. 47, 1515 (1964), [223] B. Schenke, P. Tribedy, and R. Venugopalan, Phys. [Sov. Phys. JETP20,1018(1965)]. Rev. Lett. 108, 252301 (2012), arXiv:1202.6646 [nucl- [198] F. Gelis and R. Venugopalan, Nucl. Phys. A776, 135 th]. (2006), arXiv:hep-ph/0601209 [hep-ph]. [224] A. Dumitru, T. Lappi, and Y. Nara, Phys. Lett. B734, [199] V. A. Abramovsky, V. N. Gribov, and O. V. 7 (2014), arXiv:1401.4124 [hep-ph]. Kancheli, Yad. Fiz. 18, 595 (1973), [Sov. J. Nucl. [225] K. Boguslavski, A. Kurkela, T. Lappi, and J. Peuron, Phys.18,308(1974)]. Proceedings, 25th Cracow Epiphany Conference on Ad- [200] F. Gelis and R. Venugopalan, Proceedings, 2nd Interna- vances in Heavy Ion Physics (Epiphany 2019): Cracow, 64

Poland, January 8-11, 2019, Acta Phys. Polon. B50, [253] K. Dusling, W. Li, and B. Schenke, Int. J. Mod. Phys. 1105 (2019), arXiv:1903.11942 [hep-ph]. E25, 1630002 (2016), arXiv:1509.07939 [nucl-ex]. [226] T. Lappi and J. Peuron, Phys. Rev. D97, 034017 (2018), [254] H. Mäntysaari and B. Schenke, Phys. Rev. Lett. 117, arXiv:1712.02194 [hep-lat]. 052301 (2016), arXiv:1603.04349 [hep-ph]. [227] T. Lappi, Phys. Lett. B643, 11 (2006), arXiv:hep- [255] H. Mäntysaari and B. Schenke, Phys. Rev. D94, 034042 ph/0606207 [hep-ph]. (2016), arXiv:1607.01711 [hep-ph]. [228] D. Kharzeev, A. Krasnitz, and R. Venugopalan, Phys. [256] H. Mäntysaari, (2020), arXiv:2001.10705 [hep-ph]. Lett. B545, 298 (2002), arXiv:hep-ph/0109253 [hep- [257] T. Lappi, B. Schenke, S. Schlichting, and R. Venu- ph]. gopalan, JHEP 01, 061 (2016), arXiv:1509.03499 [hep- [229] G. Chen, R. J. Fries, J. I. Kapusta, and Y. Li, Phys. ph]. Rev. C92, 064912 (2015), arXiv:1507.03524 [nucl-th]. [258] B. Schenke, S. Schlichting, P. Tribedy, and R. Venu- [230] A. Dumitru, F. Gelis, L. McLerran, and R. Venu- gopalan, Phys. Rev. Lett. 117, 162301 (2016), gopalan, Nucl. Phys. A810, 91 (2008), arXiv:0804.3858 arXiv:1607.02496 [hep-ph]. [hep-ph]. [259] B. Schenke, C. Shen, and P. Tribedy, Phys. Lett. B803, [231] G. S. Bali, Phys. Rept. 343, 1 (2001), arXiv:hep- 135322 (2020), arXiv:1908.06212 [nucl-th]. ph/0001312 [hep-ph]. [260] F. Gelis, T. Lappi, and R. Venugopalan, Phys. Rev. [232] B. Andersson, G. Gustafson, and C. Peterson, Z. Phys. D78, 054019 (2008), arXiv:0804.2630 [hep-ph]. C1, 105 (1979). [261] F. Gelis, T. Lappi, and R. Venugopalan, Phys. Rev. [233] X. Artru, Phys. Rept. 97, 147 (1983). D78, 054020 (2008), arXiv:0807.1306 [hep-ph]. [234] B. Andersson, G. Gustafson, G. Ingelman, and T. Sjos- [262] K. Dusling, F. Gelis, and R. Venugopalan, Nucl. Phys. trand, Phys. Rept. 97, 31 (1983). A872, 161 (2011), arXiv:1106.3927 [nucl-th]. [235] A. H. Mueller, Nucl. Phys. B572, 227 (2000), [263] K. Fukushima, F. Gelis, and L. McLerran, Nucl. Phys. arXiv:hep-ph/9906322 [hep-ph]. A786, 107 (2007), arXiv:hep-ph/0610416 [hep-ph]. [236] T. Lappi, S. Srednyak, and R. Venugopalan, JHEP 01, [264] T. Epelbaum and F. Gelis, Phys. Rev. D88, 085015 066 (2010), arXiv:0911.2068 [hep-ph]. (2013), arXiv:1307.1765 [hep-ph]. [237] F. Gelis, T. Lappi, and L. McLerran, Nucl. Phys. A828, [265] P. Romatschke and R. Venugopalan, Phys. Rev. D74, 149 (2009), arXiv:0905.3234 [hep-ph]. 045011 (2006), arXiv:hep-ph/0605045 [hep-ph]. [238] B. Schenke, P. Tribedy, and R. Venugopalan, Phys. [266] S. Mrówczyński, Phys. Lett. B314, 118 (1993). Rev. C 86, 034908 (2012), arXiv:1206.6805 [hep-ph]. [267] A. Rebhan, P. Romatschke, and M. Strickland, Phys. [239] B. Alver and G. Roland, Phys. Rev. C81, 054905 Rev. Lett. 94, 102303 (2005), arXiv:hep-ph/0412016 (2010), [Erratum: Phys. Rev.C82,039903(2010)], [hep-ph]. arXiv:1003.0194 [nucl-th]. [268] M. Attems, A. Rebhan, and M. Strickland, Phys. Rev. [240] B. H. Alver, C. Gombeaud, M. Luzum, and J.-Y. Olli- D87, 025010 (2013), arXiv:1207.5795 [hep-ph]. trault, Phys. Rev. C82, 034913 (2010), arXiv:1007.5469 [269] P. B. Arnold, J. Lenaghan, and G. D. Moore, JHEP [nucl-th]. 08, 002 (2003), arXiv:hep-ph/0307325 [hep-ph]. [241] B. Schenke, P. Tribedy, and R. Venugopalan, Phys. [270] G. Aarts and J. Berges, Phys. Rev. Lett. 88, 041603 Rev. C89, 024901 (2014), arXiv:1311.3636 [hep-ph]. (2002), arXiv:hep-ph/0107129 [hep-ph]. [242] B. Schenke, P. Tribedy, and R. Venugopalan, Phys. [271] K. Fukushima and F. Gelis, Nucl. Phys. A874, 108 Rev. C89, 064908 (2014), arXiv:1403.2232 [nucl-th]. (2012), arXiv:1106.1396 [hep-ph]. [243] F. D. Aaron et al. (H1, ZEUS), JHEP 01, 109 (2010), [272] J. Berges and S. Schlichting, Phys. Rev. D87, 014026 arXiv:0911.0884 [hep-ex]. (2013), arXiv:1209.0817 [hep-ph]. [244] H. Abramowicz et al. (H1, ZEUS), Eur. Phys. J. C73, [273] J. Berges, S. Borsanyi, and C. Wetterich, Phys. Rev. 2311 (2013), arXiv:1211.1182 [hep-ex]. Lett. 93, 142002 (2004), arXiv:hep-ph/0403234. [245] B. Schenke, S. Jeon, and C. Gale, Phys. Rev. Lett. 106, [274] P. B. Arnold, J. Lenaghan, G. D. Moore, and L. G. 042301 (2011), arXiv:1009.3244 [hep-ph]. Yaffe, Phys. Rev. Lett. 94, 072302 (2005), arXiv:nucl- [246] C. Gale, S. Jeon, B. Schenke, P. Tribedy, and th/0409068. R. Venugopalan, Phys. Rev. Lett. 110, 012302 (2013), [275] K. Dusling, T. Epelbaum, F. Gelis, and R. Venu- arXiv:1209.6330 [nucl-th]. gopalan, Nucl. Phys. A850, 69 (2011), arXiv:1009.4363 [247] S. Ryu, J. F. Paquet, C. Shen, G. S. Denicol, B. Schenke, [hep-ph]. S. Jeon, and C. Gale, Phys. Rev. Lett. 115, 132301 [276] R. Baier, A. H. Mueller, D. Schiff, and D. T. Son, Phys. (2015), arXiv:1502.01675 [nucl-th]. Lett. B502, 51 (2001), arXiv:hep-ph/0009237 [hep-ph]. [248] B. Schenke and S. Schlichting, Phys. Rev. C94, 044907 [277] D. Bodeker, JHEP 10, 092 (2005), arXiv:hep- (2016), arXiv:1605.07158 [hep-ph]. ph/0508223 [hep-ph]. [249] D. Müller, Simulations of the Glasma in 3+1D, Ph.D. [278] D. Son, (1996), arXiv:hep-ph/9601377. thesis, TU Vienna (2019), arXiv:1904.04267 [hep-ph]. [279] S. Khlebnikov and I. Tkachev, Phys. Rev. Lett. 77, 219 [250] S. McDonald, S. Jeon, and C. Gale, in 28th Interna- (1996), arXiv:hep-ph/9603378. tional Conference on Ultrarelativistic Nucleus-Nucleus [280] G. Aarts and J. Smit, Nucl. Phys. B 555, 355 (1999), Collisions (Quark Matter 2019) Wuhan, China, Novem- arXiv:hep-ph/9812413. ber 4-9, 2019 (2020) arXiv:2001.08636 [nucl-th]. [281] S. Borsanyi and M. Hindmarsh, Phys. Rev. D 79, [251] A. Bzdak, B. Schenke, P. Tribedy, and R. Venugopalan, 065010 (2009), arXiv:0809.4711 [hep-ph]. Phys. Rev. C87, 064906 (2013), arXiv:1304.3403 [nucl- [282] P. M. Saffin and A. Tranberg, JHEP 07, 066 (2011), th]. arXiv:1105.5546 [hep-ph]. [252] B. Schenke and R. Venugopalan, Phys. Rev. Lett. 113, [283] J. Berges, D. Gelfand, and J. Pruschke, Phys. Rev. 102301 (2014), arXiv:1405.3605 [nucl-th]. Lett. 107, 061301 (2011), arXiv:1012.4632 [hep-ph]. 65

[284] V. Kasper, F. Hebenstreit, and J. Berges, Phys. Rev. [312] J.-P. Blaizot, F. Gelis, J.-F. Liao, L. McLerran, D 90, 025016 (2014), arXiv:1403.4849 [hep-ph]. and R. Venugopalan, Nucl. Phys. A873, 68 (2012), [285] F. Gelis, K. Kajantie, and T. Lappi, Phys. Rev. Lett. arXiv:1107.5296 [hep-ph]. 96, 032304 (2006), arXiv:hep-ph/0508229. [313] J. Berges, K. Boguslavski, S. Schlichting, and [286] D. Gelfand, F. Hebenstreit, and J. Berges, Phys. Rev. R. Venugopalan, Phys. Rev. D92, 096006 (2015), D 93, 085001 (2016), arXiv:1601.03576 [hep-ph]. arXiv:1508.03073 [hep-ph]. [287] N. Tanji and J. Berges, Phys. Rev. D 97, 034013 (2018), [314] R. Micha and I. I. Tkachev, Phys. Rev. Lett. 90, 121301 arXiv:1711.03445 [hep-ph]. (2003), arXiv:hep-ph/0210202. [288] A. H. Mueller and D. T. Son, Phys. Lett. B582, 279 [315] J. H. Traschen and R. H. Brandenberger, Phys. Rev. D (2004), arXiv:hep-ph/0212198 [hep-ph]. 42, 2491 (1990). [289] S. Jeon, Phys. Rev. C72, 014907 (2005), arXiv:hep- [316] L. Kofman, A. D. Linde, and A. A. Starobinsky, Phys. ph/0412121 [hep-ph]. Rev. Lett. 73, 3195 (1994), arXiv:hep-th/9405187. [290] G. Baym, Phys. Rev. 127, 1391 (1962). [317] S. Schlichting, Phys. Rev. D86, 065008 (2012), [291] J. M. Cornwall, R. Jackiw, and E. Tomboulis, Phys. arXiv:1207.1450 [hep-ph]. Rev. D10, 2428 (1974). [318] A. Kurkela and G. D. Moore, Phys. Rev. D86, 056008 [292] E. Calzetta and B. Hu, Phys. Rev. D 37, 2878 (1988). (2012), arXiv:1207.1663 [hep-ph]. [293] J. Berges, Proceedings, 9th Hadron Physics and 7th Rel- [319] J. Berges and B. Wallisch, Phys. Rev. D 95, 036016 ativistic Aspects of Nuclear Physics (HADRON-RANP (2017), arXiv:1607.02160 [hep-ph]. 2004): A Joint Meeting on QCD and QGP: Rio de [320] L. Shen and J. Berges, Phys. Rev. D101, 056009 (2020), Janeiro, Brazil, March 28-April 3, 2004, AIP Conf. arXiv:1912.07565 [hep-ph]. Proc. 739, 3 (2004), arXiv:hep-ph/0409233 [hep-ph]. [321] T. Lappi and J. Peuron, Phys. Rev. D95, 014025 (2017), [294] J. Berges, Phys. Rev. D70, 105010 (2004), arXiv:hep- arXiv:1610.03711 [hep-ph]. ph/0401172 [hep-ph]. [322] M. Mace, S. Schlichting, and R. Venugopalan, Phys. [295] Y. Hatta and A. Nishiyama, Nucl. Phys. A873, 47 Rev. D93, 074036 (2016), arXiv:1601.07342 [hep-ph]. (2012), arXiv:1108.0818 [hep-ph]. [323] K. Boguslavski, A. Kurkela, T. Lappi, and J. Peuron, [296] A. Arrizabalaga and J. Smit, Phys. Rev. D66, 065014 Phys. Rev. D98, 014006 (2018), arXiv:1804.01966 [hep- (2002), arXiv:hep-ph/0207044 [hep-ph]. ph]. [297] J.-P. Blaizot and E. Iancu, Phys. Rept. 359, 355 (2002), [324] J. Berges, M. Mace, and S. Schlichting, Phys. Rev. Lett. arXiv:hep-ph/0101103 [hep-ph]. 118, 192005 (2017), arXiv:1703.00697 [hep-th]. [298] J. Berges, B. Schenke, S. Schlichting, and R. Venu- [325] J. Berges, K. Boguslavski, M. Mace, and J. M. gopalan, Proceedings, 24th International Conference Pawlowski, (2019), arXiv:1909.06147 [hep-ph]. on Ultra-Relativistic Nucleus-Nucleus Collisions (Quark [326] I. Montvay and G. Munster, Quantum fields on a lat- Matter 2014): Darmstadt, Germany, May 19-24, 2014, tice, Cambridge Monographs on Mathematical Physics Nucl. Phys. A931, 348 (2014), arXiv:1409.1638 [hep- (Cambridge University Press, 1997). ph]. [327] G. D. Moore and M. Tassler, JHEP 02, 105 (2011), [299] K. Fukushima, Phys. Rev. C76, 021902 (2007), [Er- arXiv:1011.1167 [hep-ph]. ratum: Phys. Rev.C77,029901(2007)], arXiv:0711.2634 [328] T. Gasenzer, L. McLerran, J. M. Pawlowski, [hep-ph]. and D. Sexty, Nucl. Phys. A930, 163 (2014), [300] J. Berges and J. Serreau, Phys. Rev. Lett. 91, 111601 arXiv:1307.5301 [hep-ph]. (2003), arXiv:hep-ph/0208070 [hep-ph]. [329] C. Ford, U. G. Mitreuter, T. Tok, A. Wipf, and J. M. [301] J. Berges, K. Boguslavski, and S. Schlichting, Phys. Pawlowski, Annals Phys. 269, 26 (1998), arXiv:hep- Rev. D85, 076005 (2012), arXiv:1201.3582 [hep-ph]. th/9802191 [hep-th]. [302] J. Berges, S. Scheffler, and D. Sexty, Phys. Rev. D77, [330] U. G. Mitreuter, J. M. Pawlowski, and A. Wipf, Nucl. 034504 (2008), arXiv:0712.3514 [hep-ph]. Phys. B514, 381 (1998), arXiv:hep-th/9611105 [hep- [303] J. Berges, D. Gelfand, S. Scheffler, and D. Sexty, Phys. th]. Lett. B677, 210 (2009), arXiv:0812.3859 [hep-ph]. [331] J. Braun, H. Gies, and J. M. Pawlowski, Phys. Lett. [304] J. Berges, Nucl. Phys. A699, 847 (2002), arXiv:hep- B684, 262 (2010), arXiv:0708.2413 [hep-th]. ph/0105311 [hep-ph]. [332] L. Fister and J. M. Pawlowski, Phys. Rev. D88, 045010 [305] G. Aarts, D. Ahrensmeier, R. Baier, J. Berges, and (2013), arXiv:1301.4163 [hep-ph]. J. Serreau, Phys. Rev. D66, 045008 (2002), arXiv:hep- [333] J. Berges and D. Sexty, Phys. Rev. Lett. 108, 161601 ph/0201308 [hep-ph]. (2012), arXiv:1201.0687 [hep-ph]. [306] G. Aarts and J. Smit, Nucl. Phys. B511, 451 (1998), [334] A. PiñeiroOrioli, K. Boguslavski, and J. Berges, Phys. arXiv:hep-ph/9707342 [hep-ph]. Rev. D92, 025041 (2015), arXiv:1503.02498 [hep-ph]. [307] J. Berges, D. Gelfand, and D. Sexty, Phys. Rev. D 89, [335] I. Chantesana, A. Piñeiro Orioli, and T. Gasen- 025001 (2014), arXiv:1308.2180 [hep-ph]. zer, Phys. Rev. A99, 043620 (2019), arXiv:1801.09490 [308] T. Epelbaum, F. Gelis, and B. Wu, Phys. Rev. D 90, [cond-mat.quant-gas]. 065029 (2014), arXiv:1402.0115 [hep-ph]. [336] C. Scheppach, J. Berges, and T. Gasenzer, Phys. Rev. [309] J. Berges, K. Boguslavski, S. Schlichting, and R. Venu- A81, 033611 (2010), arXiv:0912.4183 [cond-mat.quant- gopalan, JHEP 05, 054 (2014), arXiv:1312.5216 [hep- gas]. ph]. [337] J. Berges and D. Sexty, Phys. Rev. D83, 085004 (2011), [310] A. Kurkela and G. D. Moore, JHEP 12, 044 (2011), arXiv:1012.5944 [hep-ph]. arXiv:1107.5050 [hep-ph]. [338] B. Nowak, D. Sexty, and T. Gasenzer, Phys. Rev. B84, [311] A. Kurkela and G. D. Moore, JHEP 11, 120 (2011), 020506 (2011), arXiv:1012.4437 [cond-mat.quant-gas]. arXiv:1108.4684 [hep-ph]. [339] B. Nowak, J. Schole, D. Sexty, and T. Gasenzer, 66

Phys. Rev. A85, 043627 (2012), arXiv:1111.6127 [cond- [364] M. Mace, N. Mueller, S. Schlichting, and S. Sharma, mat.quant-gas]. (2019), arXiv:1910.01654 [hep-ph]. [340] G. D. Moore, Phys. Rev. D93, 065043 (2016), [365] P. Buividovich and M. Ulybyshev, Phys. Rev. D 94, arXiv:1511.00697 [hep-ph]. 025009 (2016), arXiv:1509.02076 [hep-th]. [341] R. Walz, K. Boguslavski, and J. Berges, Phys. Rev. [366] M. Carrington and E. Kovalchuk, Phys. Rev. D 80, D97, 116011 (2018), arXiv:1710.11146 [hep-ph]. 085013 (2009), arXiv:0906.1140 [hep-ph]. [342] J. Deng, S. Schlichting, R. Venugopalan, and Q. Wang, [367] P. B. Arnold, G. D. Moore, and L. G. Yaffe, JHEP 01, Phys. Rev. A97, 053606 (2018), arXiv:1801.06260 [hep- 030 (2003), arXiv:hep-ph/0209353 [hep-ph]. th]. [368] E. Braaten and R. D. Pisarski, Nucl. Phys. B337, 569 [343] A. PiñeiroOrioli and J. Berges, Phys. Rev. Lett. 122, (1990). 150401 (2019), arXiv:1810.12392 [cond-mat.quant-gas]. [369] S. Jeon, Phys. Rev. D52, 3591 (1995), arXiv:hep- [344] K. Boguslavski and A. Piñeiro Orioli, (2019), ph/9409250 [hep-ph]. arXiv:1911.04506 [hep-ph]. [370] S. Jeon and L. G. Yaffe, Phys. Rev. D53, 5799 (1996), [345] J. Berges and G. Hoffmeister, Nucl. Phys. B813, 383 arXiv:hep-ph/9512263 [hep-ph]. (2009), arXiv:0809.5208 [hep-th]. [371] E. A. Calzetta, B. L. Hu, and S. A. Ramsey, Phys. Rev. [346] J. Berges and D. Mesterhazy, Physics at all scales: D61, 125013 (2000), arXiv:hep-ph/9910334 [hep-ph]. The Renormalization Group. Proceedings, 49. Inter- [372] G. Aarts and J. M. Martinez Resco, JHEP 03, 074 nationale Universitätswochen fürTheoretische Physik, (2005), arXiv:hep-ph/0503161 [hep-ph]. Winter School: Schladming, Austria, February 26- [373] J.-S. Gagnon and S. Jeon, Phys. Rev. D76, 105019 March 5, 2011, Nucl. Phys. Proc. Suppl. 228, 37 (2012), (2007), arXiv:0708.1631 [hep-ph]. arXiv:1204.1489 [hep-ph]. [374] P. Aurenche, F. Gelis, R. Kobes, and H. Zaraket, Phys. [347] L. Corell, A. K. Cyrol, M. Heller, and J. M. Pawlowski, Rev. D 58, 085003 (1998), arXiv:hep-ph/9804224. (2019), arXiv:1910.09369 [hep-th]. [375] P. B. Arnold, G. D. Moore, and L. G. Yaffe, JHEP 11, [348] R. Chatterjee, L. Bhattacharya, and D. K. Srivastava, 057 (2001), arXiv:hep-ph/0109064 [hep-ph]. Lect. Notes Phys. 785, 219 (2010), arXiv:0901.3610 [376] P. B. Arnold, G. D. Moore, and L. G. Yaffe, JHEP 05, [nucl-th]. 051 (2003), arXiv:hep-ph/0302165 [hep-ph]. [349] D. E. Kharzeev, J. Liao, S. A. Voloshin, and G. Wang, [377] P. B. Arnold, C. Dogan, and G. D. Moore, Phys. Rev. Prog. Part. Nucl. Phys. 88, 1 (2016), arXiv:1511.04050 D74, 085021 (2006), arXiv:hep-ph/0608012 [hep-ph]. [hep-ph]. [378] M. A. York and G. D. Moore, Phys. Rev. D79, 054011 [350] V. Koch, S. Schlichting, V. Skokov, P. Sorensen, (2009), arXiv:0811.0729 [hep-ph]. J. Thomas, S. Voloshin, G. Wang, and H.-U. Yee, Chin. [379] M. C. Abraao York, A. Kurkela, E. Lu, and Phys. C 41, 072001 (2017), arXiv:1608.00982 [nucl-th]. G. D. Moore, Phys. Rev. D89, 074036 (2014), [351] N. Tanji and R. Venugopalan, Phys. Rev. D95, 094009 arXiv:1401.3751 [hep-ph]. (2017), arXiv:1703.01372 [hep-ph]. [380] A. Kurkela and E. Lu, Phys. Rev. Lett. 113, 182301 [352] A. Lehmann and A. Rothkopf, in 37th International (2014), arXiv:1405.6318 [hep-ph]. Symposium on Lattice Field Theory (Lattice 2019) [381] A. Kurkela and Y. Zhu, Phys. Rev. Lett. 115, 182301 Wuhan, Hubei, China, June 16-22, 2019 (2020) (2015), arXiv:1506.06647 [hep-ph]. arXiv:2003.02509 [hep-lat]. [382] L. Keegan, A. Kurkela, A. Mazeliauskas, and [353] K. Fukushima, D. E. Kharzeev, and H. J. Warringa, D. Teaney, JHEP 08, 171 (2016), arXiv:1605.04287 Phys. Rev. D 78, 074033 (2008), arXiv:0808.3382 [hep- [hep-ph]. ph]. [383] F. A. Berezin and M. S. Marinov, Annals Phys. 104, [354] V. Skokov, A. Yu. Illarionov, and V. Toneev, Int. J. 336 (1977). Mod. Phys. A24, 5925 (2009), arXiv:0907.1396 [nucl- [384] N. Mueller and R. Venugopalan, Phys. Rev. D99, th]. 056003 (2019), arXiv:1901.10492 [hep-th]. [355] E. Shuryak and I. Zahed, Phys. Rev. D67, 014006 [385] D. T. Son and N. Yamamoto, Phys. Rev. D87, 085016 (2003), arXiv:hep-ph/0206022 [hep-ph]. (2013), arXiv:1210.8158 [hep-th]. [356] N. Müller, S. Schlichting, and S. Sharma, Phys. Rev. [386] M. A. Stephanov and Y. Yin, Phys. Rev. Lett. 109, Lett. 117, 142301 (2016), arXiv:1606.00342 [hep-ph]. 162001 (2012), arXiv:1207.0747 [hep-th]. [357] M. Mace, N. Mueller, S. Schlichting, and S. Sharma, [387] J.-Y. Chen, D. T. Son, and M. A. Stephanov, Phys. Phys. Rev. D 95, 036023 (2017), arXiv:1612.02477 [hep- Rev. Lett. 115, 021601 (2015), arXiv:1502.06966 [hep- lat]. th]. [358] N. Tanji, N. Mueller, and J. Berges, Phys. Rev. D 93, [388] N. Weickgenannt, X.-L. Sheng, E. Speranza, Q. Wang, 074507 (2016), arXiv:1603.03331 [hep-ph]. and D. H. Rischke, Phys. Rev. D100, 056018 (2019), [359] N. Tanji, Phys. Rev. D 98, 014025 (2018), arXiv:1902.06513 [hep-ph]. arXiv:1805.00775 [hep-ph]. [389] K. Hattori, Y. Hidaka, and D.-L. Yang, Phys. Rev. [360] T. Zache, N. Mueller, J. Schneider, F. Jendrzejewski, D100, 096011 (2019), arXiv:1903.01653 [hep-ph]. J. Berges, and P. Hauke, Phys. Rev. Lett. 122, 050403 [390] J.-H. Gao and Z.-T. Liang, Phys. Rev. D100, 056021 (2019), arXiv:1808.07885 [quant-ph]. (2019), arXiv:1902.06510 [hep-ph]. [361] D. E. Kharzeev and Y. Kikuchi, (2020), [391] D.-L. Yang, K. Hattori, and Y. Hidaka, (2020), arXiv:2001.00698 [hep-ph]. arXiv:2002.02612 [hep-ph]. [362] R. Ott, T. V. Zache, N. Mueller, and J. Berges, (2019), [392] X.-L. Sheng, Q. Wang, and X.-G. Huang, (2020), arXiv:1903.11109 [cond-mat.quant-gas]. arXiv:2005.00204 [hep-ph]. [363] N. Mueller, F. Hebenstreit, and J. Berges, Phys. Rev. [393] S. Carignano, C. Manuel, and J. M. Torres-Rincon, Lett. 117, 061601 (2016), arXiv:1605.01413 [hep-ph]. (2019), arXiv:1908.00561 [hep-ph]. 67

[394] N. Mueller and R. Venugopalan, Phys. Rev. D96, Collisions (2020) arXiv:2001.10046 [hep-ph]. 016023 (2017), arXiv:1702.01233 [hep-ph]. [425] T. Lappi, Phys. Lett. B703, 325 (2011), [395] K. Fujikawa, Phys. Rev. D73, 025017 (2006), arXiv:1105.5511 [hep-ph]. arXiv:hep-th/0511142 [hep-th]. [426] R. Baier, A. H. Mueller, D. Schiff, and D. Son, Phys. [396] N. Mueller and R. Venugopalan, Phys. Rev. D97, Lett. B 539, 46 (2002), arXiv:hep-ph/0204211. 051901 (2018), arXiv:1701.03331 [hep-ph]. [427] R. Baier, A. Mueller, D. Schiff, and D. Son, (2011), [397] H.-U. Yee and P. Yi, Phys. Rev. D101, 045007 (2020), arXiv:1103.1259 [nucl-th]. arXiv:1909.12409 [hep-th]. [428] A. Mazeliauskas and J. Berges, Phys. Rev. Lett. 122, [398] G. Inghirami, M. Mace, Y. Hirono, L. Del Zanna, D. E. 122301 (2019), arXiv:1810.10554 [hep-ph]. Kharzeev, and M. Bleicher, Eur. Phys. J. C80, 293 [429] C.-M. Schmied, A. N. Mikheev, and T. Gasenzer, Phys. (2020), arXiv:1908.07605 [hep-ph]. Rev. Lett. 122, 170404 (2019), arXiv:1807.07514 [cond- [399] K. Landsteiner, Proceedings, 56th Cracow School of mat.quant-gas]. Theoretical Physics : A Panorama of Holography: Za- [430] M. Tanabashi et al. (Particle Data Group), Phys. Rev. kopane, Poland, May 24-June 1, 2016, Acta Phys. D98, 030001 (2018). Polon. B47, 2617 (2016), arXiv:1610.04413 [hep-th]. [431] J. Blaizot, E. Iancu, and A. Rebhan, Phys. Rev. D 68, [400] G. Baym, Phys. Lett. 138B, 18 (1984). 025011 (2003), arXiv:hep-ph/0303045. [401] S. Mrówczyński, Phys. Rev. D65, 117501 (2002), [432] S. Caron-Huot and G. D. Moore, JHEP 02, 081 (2008), arXiv:hep-ph/0112100 [hep-ph]. arXiv:0801.2173 [hep-ph]. [402] B. Schenke and M. Strickland, Phys. Rev. D74, 065004 [433] J. Ghiglieri, J. Hong, A. Kurkela, E. Lu, G. D. Moore, (2006), arXiv:hep-ph/0606160 [hep-ph]. and D. Teaney, JHEP 05, 010 (2013), arXiv:1302.5970 [403] S. Mrówczyński, Phys. Lett. B393, 26 (1997), [hep-ph]. arXiv:hep-ph/9606442 [hep-ph]. [434] J. Ghiglieri, G. D. Moore, and D. Teaney, Phys. Rev. [404] G. D. Moore and D. Teaney, Phys. Rev. C71, 064904 Lett. 121, 052302 (2018), arXiv:1805.02663 [hep-ph]. (2005), arXiv:hep-ph/0412346 [hep-ph]. [435] S. Caron-Huot, Phys. Rev. D79, 065039 (2009), [405] J. Hong and D. Teaney, Phys. Rev. C82, 044908 (2010), arXiv:0811.1603 [hep-ph]. arXiv:1003.0699 [nucl-th]. [436] J.-F. Paquet, C. Shen, G. S. Denicol, M. Luzum, [406] J. Ghiglieri and D. Teaney, Int. J. Mod. Phys. E24, B. Schenke, S. Jeon, and C. Gale, Phys. Rev. C93, 1530013 (2015), [,271(2016)], arXiv:1502.03730 [hep- 044906 (2016), arXiv:1509.06738 [hep-ph]. ph]. [437] X. Yao, W. Ke, Y. Xu, S. A. Bass, and B. Müller, [407] J. Ghiglieri, G. D. Moore, and D. Teaney, JHEP 03, (2020), arXiv:2004.06746 [hep-ph]. 095 (2016), arXiv:1509.07773 [hep-ph]. [438] K. M. Burke et al. (JET), Phys. Rev. C90, 014909 [408] J. Ghiglieri, G. D. Moore, and D. Teaney, JHEP 03, (2014), arXiv:1312.5003 [nucl-th]. 179 (2018), arXiv:1802.09535 [hep-ph]. [439] R. C. Hwa and K. Kajantie, Phys. Rev. Lett. 56, 696 [409] J.-P. Blaizot, B. Wu, and L. Yan, Nucl. Phys. A930, (1986). 139 (2014), arXiv:1402.5049 [hep-ph]. [440] T. Biro, E. van Doorn, B. Muller, M. Thoma, and [410] P. B. Arnold, Phys. Rev. E61, 6099 (2000), arXiv:hep- X. Wang, Phys. Rev. C 48, 1275 (1993), arXiv:nucl- ph/9912209 [hep-ph]. th/9303004. [411] P. B. Arnold, Phys. Rev. E61, 6091 (2000), arXiv:hep- [441] K. Geiger and B. Muller, Nucl. Phys. B 369, 600 (1992). ph/9912208 [hep-ph]. [442] Z. Xu and C. Greiner, Phys. Rev. C71, 064901 (2005), [412] A. Kurkela and A. Mazeliauskas, Phys. Rev. D99, arXiv:hep-ph/0406278 [hep-ph]. 054018 (2019), arXiv:1811.03068 [hep-ph]. [443] A. El, Z. Xu, and C. Greiner, Nucl. Phys. A 806, 287 [413] L. D. Landau and I. Pomeranchuk, Dokl. Akad. Nauk (2008), arXiv:0712.3734 [hep-ph]. Ser. Fiz. 92, 735 (1953). [444] L. D. Landau and E. M. Lifshits, Fluid mechanics, [414] L. D. Landau and I. Pomeranchuk, Dokl. Akad. Nauk Course of theoretical physics, Vol. 6 (London, Perga- Ser. Fiz. 92, 535 (1953). mon Press; Reading, Mass., Addison-Wesley Pub. Co., [415] A. B. Migdal, Phys. Rev. 103, 1811 (1956). 1959., 1959). [416] A. B. Migdal, Dokl. Akad. Nauk Ser. Fiz. 105, 77 [445] L. Keegan, A. Kurkela, P. Romatschke, W. van der (1955). Schee, and Y. Zhu, JHEP 04, 031 (2016), [417] P. B. Arnold, G. D. Moore, and L. G. Yaffe, JHEP 06, arXiv:1512.05347 [hep-th]. 030 (2002), arXiv:hep-ph/0204343 [hep-ph]. [446] M. P. Heller, A. Kurkela, M. Spaliński, and V. Svens- [418] P. Aurenche, F. Gelis, and H. Zaraket, JHEP 05, 043 son, Phys. Rev. D97, 091503 (2018), arXiv:1609.04803 (2002), arXiv:hep-ph/0204146 [hep-ph]. [nucl-th]. [419] P. Aurenche, F. Gelis, G. D. Moore, and H. Zaraket, [447] A. Kurkela, A. Mazeliauskas, J.-F. Paquet, S. Schlicht- JHEP 12, 006 (2002), arXiv:hep-ph/0211036 [hep-ph]. ing, and D. Teaney, Phys. Rev. C99, 034910 (2019), [420] J. Ghiglieri and G. D. Moore, JHEP 12, 029 (2014), arXiv:1805.00961 [hep-ph]. arXiv:1410.4203 [hep-ph]. [448] J. E. Bernhard, J. S. Moreland, and S. A. Bass, Nature [421] I. Ghisoiu and M. Laine, JHEP 10, 083 (2014), Phys. 15, 1113 (2019). arXiv:1407.7955 [hep-ph]. [449] D. Devetak, A. Dubla, S. Floerchinger, E. Grossi, [422] P. B. Arnold and C. Dogan, Phys. Rev. D78, 065008 S. Masciocchi, A. Mazeliauskas, and I. Selyuzhenkov, (2008), arXiv:0804.3359 [hep-ph]. (2019), arXiv:1909.10485 [hep-ph]. [423] S. Hauksson, S. Jeon, and C. Gale, Phys. Rev. C97, [450] P. Hanus, A. Mazeliauskas, and K. Reygers, Phys. Rev. 014901 (2018), arXiv:1709.03598 [nucl-th]. C 100, 064903 (2019), arXiv:1908.02792 [hep-ph]. [424] S. Hauksson, S. Jeon, and C. Gale, in 28th Interna- [451] D. Almaalol, A. Kurkela, and M. Strickland, (2020), tional Conference on Ultrarelativistic Nucleus-Nucleus arXiv:2004.05195 [hep-ph]. 68

[452] M. Strickland, J. Noronha, and G. Denicol, Phys. Rev. Rev. D85, 126002 (2012), arXiv:1203.0755 [hep-th]. D 97, 036020 (2018), arXiv:1709.06644 [nucl-th]. [480] P. M. Chesler and L. G. Yaffe, JHEP 07, 086 (2014), [453] J.-P. Blaizot and L. Yan, Phys. Lett. B 780, 283 (2018), arXiv:1309.1439 [hep-th]. arXiv:1712.03856 [nucl-th]. [481] H. Liu and J. Sonner, (2018), arXiv:1810.02367 [hep- [454] M. Strickland, JHEP 12, 128 (2018), arXiv:1809.01200 th]. [nucl-th]. [482] M. P. Heller, D. Mateos, W. van der Schee, and [455] A. Behtash, S. Kamata, M. Martinez, and H. Shi, Phys. D. Trancanelli, Phys. Rev. Lett. 108, 191601 (2012), Rev. D 99, 116012 (2019), arXiv:1901.08632 [hep-th]. arXiv:1202.0981 [hep-th]. [456] B. Müllerand A. Schafer, Int. J. Mod. Phys. E20, 2235 [483] P. M. Chesler and L. G. Yaffe, Phys. Rev. Lett. 102, (2011), arXiv:1110.2378 [hep-ph]. 211601 (2009), arXiv:0812.2053 [hep-th]. [457] A. Andronic, P. Braun-Munzinger, K. Redlich, and [484] P. Romatschke, Eur. Phys. J. C76, 352 (2016), J. Stachel, Nature 561, 321 (2018), arXiv:1710.09425 arXiv:1512.02641 [hep-th]. [nucl-th]. [485] S. Grozdanov, N. Kaplis, and A. O. Starinets, JHEP [458] Z. Citron et al., CERN Yellow Rep. Monogr. 7, 1159 07, 151 (2016), arXiv:1605.02173 [hep-th]. (2019), arXiv:1812.06772 [hep-ph]. [486] A. Kurkela and U. A. Wiedemann, Eur. Phys. J. C79, [459] U. Heinz and R. Snellings, Ann. Rev. Nucl. Part. Sci. 776 (2019), arXiv:1712.04376 [hep-ph]. 63, 123 (2013), arXiv:1301.2826 [nucl-th]. [487] S. A. Hartnoll, A. Lucas, and S. Sachdev, (2016), [460] B. Schenke, C. Shen, and D. Teaney, (2020), arXiv:1612.07324 [hep-th]. arXiv:2004.00690 [nucl-th]. [488] P. Mitra, M. Ippoliti, R. Bhatt, S. Sondhi, and K. Agar- [461] C. Gale, J.-F. Paquet, B. Schenke, and C. Shen, in 28th wal, Phys. Rev. B 99, 104308 (2019), arXiv:1809.01681 International Conference on Ultrarelativistic Nucleus- [cond-mat.stat-mech]. Nucleus Collisions (2020) arXiv:2002.05191 [hep-ph]. [489] G. Beuf, M. P. Heller, R. A. Janik, and R. Peschanski, [462] W. Broniowski, W. Florkowski, M. Chojnacki, JHEP 10, 043 (2009), arXiv:0906.4423 [hep-th]. and A. Kisiel, Phys. Rev. C80, 034902 (2009), [490] J. Jankowski, G. Plewa, and M. Spalinski, JHEP 12, arXiv:0812.3393 [nucl-th]. 105 (2014), arXiv:1411.1969 [hep-th]. [463] J. Liu, C. Shen, and U. Heinz, Phys. Rev. C91, 064906 [491] D. Grumiller and P. Romatschke, JHEP 08, 027 (2008), (2015), [Erratum: Phys. Rev.C92,no.4,049904(2015)], arXiv:0803.3226 [hep-th]. arXiv:1504.02160 [nucl-th]. [492] B. Wu and P. Romatschke, Int. J. Mod. Phys. C22, [464] W. van der Schee, P. Romatschke, and S. Pratt, Phys. 1317 (2011), arXiv:1108.3715 [hep-th]. Rev. Lett. 111, 222302 (2013), arXiv:1307.2539 [nucl- [493] M. P. Heller, R. A. Janik, and P. Witaszczyk, Phys. th]. Rev. Lett. 110, 211602 (2013), arXiv:1302.0697 [hep- [465] P. Romatschke, Eur. Phys. J. C75, 305 (2015), th]. arXiv:1502.04745 [nucl-th]. [494] I. Aniceto and M. Spaliński, Phys. Rev. D93, 085008 [466] J. Vredevoogd and S. Pratt, Phys. Rev. C 79, 044915 (2016), arXiv:1511.06358 [hep-th]. (2009), arXiv:0810.4325 [nucl-th]. [495] I. Aniceto, B. Meiring, J. Jankowski, and M. Spaliński, [467] S. Kamata, M. Martinez, P. Plaschke, S. Ochsenfeld, JHEP 02, 073 (2019), arXiv:1810.07130 [hep-th]. and S. Schlichting, (2020), arXiv:2004.06751 [hep-ph]. [496] D. Dorigoni, Annals Phys. 409, 167914 (2019), [468] M. Greif, C. Greiner, B. Schenke, S. Schlicht- arXiv:1411.3585 [hep-th]. ing, and Z. Xu, Phys. Rev. D 96, 091504 (2017), [497] I. Aniceto, G. Basar, and R. Schiappa, Phys. Rept. arXiv:1708.02076 [hep-ph]. 809, 1 (2019), arXiv:1802.10441 [hep-th]. [469] A. Kurkela, W. van der Schee, U. A. Wiedemann, [498] G. S. Denicol and J. Noronha, Phys. Rev. D 99, 116004 and B. Wu, Phys. Rev. Lett. 124, 102301 (2020), (2019), arXiv:1804.04771 [nucl-th]. arXiv:1907.08101 [hep-ph]. [499] A. Buchel, M. P. Heller, and J. Noronha, Phys. Rev. [470] A. Kurkela, U. A. Wiedemann, and B. Wu, Eur. Phys. D94, 106011 (2016), arXiv:1603.05344 [hep-th]. J. C 79, 965 (2019), arXiv:1905.05139 [hep-ph]. [500] G. S. Denicol and J. Noronha, (2019), arXiv:1908.09957 [471] A. Kurkela, U. A. Wiedemann, and B. Wu, Phys. Lett. [nucl-th]. B 783, 274 (2018), arXiv:1803.02072 [hep-ph]. [501] M. P. Heller, A. Serantes, M. Spaliński,V. Svensson, [472] A. Kurkela, U. A. Wiedemann, and B. Wu, Eur. Phys. and B. Withers, (2020), arXiv:2007.05524 [hep-th]. J. C 79, 759 (2019), arXiv:1805.04081 [hep-ph]. [502] B. Withers, JHEP 06, 059 (2018), arXiv:1803.08058 [473] M. Ammon and J. Erdmenger, Gauge/gravity duality [hep-th]. (Cambridge University Press, Cambridge, 2015). [503] S. Grozdanov, P. K. Kovtun, A. O. Starinets, and [474] R. P. Woodard, Scholarpedia 10, 32243 (2015), P. Tadić, Phys. Rev. Lett. 122, 251601 (2019), arXiv:1506.02210 [hep-th]. arXiv:1904.01018 [hep-th]. [475] S. Bhattacharyya, R. Loganayagam, I. Mandal, S. Min- [504] S. Grozdanov, P. K. Kovtun, A. O. Starinets, and walla, and A. Sharma, JHEP 12, 116 (2008), P. Tadić, JHEP 11, 097 (2019), arXiv:1904.12862 [hep- arXiv:0809.4272 [hep-th]. th]. [476] V. Balasubramanian and P. Kraus, Commun. Math. [505] S. Nakamura and S.-J. Sin, JHEP 09, 020 (2006), Phys. 208, 413 (1999), arXiv:hep-th/9902121. arXiv:hep-th/0607123. [477] S. de Haro, S. N. Solodukhin, and K. Skenderis, [506] R. A. Janik, Phys. Rev. Lett. 98, 022302 (2007), Commun. Math. Phys. 217, 595 (2001), arXiv:hep- arXiv:hep-th/0610144. th/0002230. [507] M. P. Heller and R. A. Janik, Phys. Rev. D 76, 025027 [478] M. P. Heller, D. Mateos, W. van der Schee, and M. Tri- (2007), arXiv:hep-th/0703243. ana, JHEP 09, 026 (2013), arXiv:1304.5172 [hep-th]. [508] M. P. Heller, P. Surowka, R. Loganayagam, M. Spalin- [479] M. P. Heller, R. A. Janik, and P. Witaszczyk, Phys. ski, and S. E. Vazquez, Phys. Rev. Lett. 102, 041601 69

(2009), arXiv:0805.3774 [hep-th]. [537] P. M. Chesler and L. G. Yaffe, JHEP 10, 070 (2015), [509] S. Kinoshita, S. Mukohyama, S. Nakamura, and arXiv:1501.04644 [hep-th]. K.-y. Oda, Prog. Theor. Phys. 121, 121 (2009), [538] P. M. Chesler, Phys. Rev. Lett. 115, 241602 (2015), arXiv:0807.3797 [hep-th]. arXiv:1506.02209 [hep-th]. [510] S. Kinoshita, S. Mukohyama, S. Nakamura, and [539] P. M. Chesler, JHEP 03, 146 (2016), arXiv:1601.01583 K.-y. Oda, Phys. Rev. Lett. 102, 031601 (2009), [hep-th]. arXiv:0901.4834 [hep-th]. [540] S. S. Gubser, S. S. Pufu, and A. Yarom, Phys. Rev. D [511] I. Booth, M. P. Heller, and M. Spalinski, Phys. Rev. D 78, 066014 (2008), arXiv:0805.1551 [hep-th]. 80, 126013 (2009), arXiv:0910.0748 [hep-th]. [541] S. Lin and E. Shuryak, Phys. Rev. D 79, 124015 (2009), [512] J. Casalderrey-Solana, N. I. Gushterov, and B. Meiring, arXiv:0902.1508 [hep-th]. JHEP 04, 042 (2018), arXiv:1712.02772 [hep-th]. [542] W. van der Schee and B. Schenke, Phys. Rev. C 92, [513] R. A. Janik and R. B. Peschanski, Phys. Rev. D 74, 064907 (2015), arXiv:1507.08195 [nucl-th]. 046007 (2006), arXiv:hep-th/0606149. [543] P. Arnold, P. Romatschke, and W. van der Schee, JHEP [514] M. P. Heller and V. Svensson, Phys. Rev. D 98, 054016 10, 110 (2014), arXiv:1408.2518 [hep-th]. (2018), arXiv:1802.08225 [nucl-th]. [544] R. A. Janik and R. B. Peschanski, Phys. Rev. D 73, [515] G. Basar and G. V. Dunne, Phys. Rev. D 92, 125011 045013 (2006), arXiv:hep-th/0512162. (2015), arXiv:1509.05046 [hep-th]. [545] S. S. Gubser and W. van der Schee, JHEP 01, 028 [516] M. Spaliński, Phys. Lett. B776, 468 (2018), (2015), arXiv:1410.7408 [hep-th]. arXiv:1708.01921 [hep-th]. [546] S. S. Gubser, Phys. Rev. C 87, 014909 (2013), [517] P. Romatschke, JHEP 12, 079 (2017), arXiv:1710.03234 arXiv:1210.4181 [hep-th]. [hep-th]. [547] L. D. Landau, Izv. Akad. Nauk Ser. Fiz. 17, 51 (1953). [518] G. S. Denicol and J. Noronha, Phys. Rev. D 97, 056021 [548] S. Belenk’ij and L. Landau, Nuovo Cim. Suppl. 3, 15 (2018), arXiv:1711.01657 [nucl-th]. (1956). [519] W. Florkowski, E. Maksymiuk, and R. Ryblewski, [549] S. Waeber, A. Rabenstein, A. Schäfer, and L. G. Yaffe, Phys. Rev. C 97, 024915 (2018), arXiv:1710.07095 [hep- JHEP 08, 005 (2019), arXiv:1906.05086 [hep-th]. ph]. [550] B. Müller, A. Rabenstein, A. Schäfer, S. Waeber, [520] A. Behtash, C. Cruz-Camacho, and M. Martinez, Phys. and L. G. Yaffe, Phys. Rev. D 101, 076008 (2020), Rev. D 97, 044041 (2018), arXiv:1711.01745 [hep-th]. arXiv:2001.07161 [hep-ph]. [521] D. Almaalol and M. Strickland, Phys. Rev. C 97, 044911 [551] A. Buchel, S. Deakin, P. Kerner, and J. T. Liu, Nucl. (2018), arXiv:1801.10173 [hep-ph]. Phys. B 784, 72 (2007), arXiv:hep-th/0701142. [522] A. Behtash, C. Cruz-Camacho, S. Kamata, and [552] U. Gursoy and E. Kiritsis, JHEP 02, 032 (2008), M. Martinez, Phys. Lett. B 797, 134914 (2019), arXiv:0707.1324 [hep-th]. arXiv:1805.07881 [hep-th]. [553] U. Gursoy, E. Kiritsis, and F. Nitti, JHEP 02, 019 [523] M. Spaliński, Phys. Lett. B 784, 21 (2018), (2008), arXiv:0707.1349 [hep-th]. arXiv:1805.11689 [hep-th]. [554] S. S. Gubser, A. Nellore, S. S. Pufu, and F. D. Rocha, [524] M. Strickland and U. Tantary, JHEP 10, 069 (2019), Phys. Rev. Lett. 101, 131601 (2008), arXiv:0804.1950 arXiv:1903.03145 [hep-ph]. [hep-th]. [525] J.-P. Blaizot and L. Yan, Annals Phys. 412, 167993 [555] E. Witten, Adv. Theor. Math. Phys. 2, 505 (1998), (2020), arXiv:1904.08677 [nucl-th]. arXiv:hep-th/9803131. [526] S. Jaiswal, C. Chattopadhyay, A. Jaiswal, S. Pal, [556] A. Buchel, M. P. Heller, and R. C. Myers, Phys. Rev. and U. Heinz, Phys. Rev. C 100, 034901 (2019), Lett. 114, 251601 (2015), arXiv:1503.07114 [hep-th]. arXiv:1907.07965 [nucl-th]. [557] R. A. Janik, G. Plewa, H. Soltanpanahi, and M. Spalin- [527] J. Brewer, L. Yan, and Y. Yin, (2019), ski, Phys. Rev. D 91, 126013 (2015), arXiv:1503.07149 arXiv:1910.00021 [nucl-th]. [hep-th]. [528] A. Behtash, S. Kamata, M. Martinez, and H. Shi, [558] M. Attems, J. Casalderrey-Solana, D. Mateos, (2019), arXiv:1911.06406 [hep-th]. D. Santos-Oliván, C. F. Sopuerta, M. Triana, and [529] C. Chattopadhyay and U. W. Heinz, Phys. Lett. B 801, M. Zilhão, JHEP 01, 026 (2017), arXiv:1604.06439 [hep- 135158 (2020), arXiv:1911.07765 [nucl-th]. th]. [530] A. Dash and V. Roy, (2020), arXiv:2001.10756 [nucl- [559] M. Attems, J. Casalderrey-Solana, D. Mateos, th]. D. Santos-Oliván, C. F. Sopuerta, M. Triana, and [531] M. Shokri and F. Taghinavaz, (2020), arXiv:2002.04719 M. Zilhão, JHEP 06, 154 (2017), arXiv:1703.09681 [hep- [hep-th]. th]. [532] M. P. Heller, R. Jefferson, M. Spaliński, and [560] M. Attems, Y. Bea, J. Casalderrey-Solana, D. Mateos, V. Svensson, Phys. Rev. Lett. 125, 132301 (2020), M. Triana, and M. Zilhão, Phys. Rev. Lett. 121, 261601 arXiv:2003.07368 [hep-th]. (2018), arXiv:1807.05175 [hep-th]. [533] A. R. Liddle, P. Parsons, and J. D. Barrow, Phys. Rev. [561] B. Craps, E. J. Lindgren, and A. Taliotis, JHEP 12, D 50, 7222 (1994), arXiv:astro-ph/9408015. 116 (2015), arXiv:1511.00859 [hep-th]. [534] W. van der Schee, Phys. Rev. D 87, 061901 (2013), [562] H. Bantilan, P. Figueras, and D. Mateos, Phys. Rev. arXiv:1211.2218 [hep-th]. Lett. 124, 191601 (2020), arXiv:2001.05476 [hep-th]. [535] J. Casalderrey-Solana, M. P. Heller, D. Mateos, and [563] S. S. Gubser, I. R. Klebanov, and A. A. Tseytlin, Nucl. W. van der Schee, Phys. Rev. Lett. 112, 221602 (2014), Phys. B 534, 202 (1998), arXiv:hep-th/9805156. arXiv:1312.2956 [hep-th]. [564] A. Buchel, Nucl. Phys. B 802, 281 (2008), [536] P. M. Chesler, N. Kilbertus, and W. van der Schee, arXiv:0801.4421 [hep-th]. JHEP 11, 135 (2015), arXiv:1507.02548 [hep-th]. [565] A. Buchel, Nucl. Phys. B 803, 166 (2008), 70

arXiv:0805.2683 [hep-th]. [592] R. Bellwied, S. Borsanyi, Z. Fodor, J. N. Guenther, [566] A. Buchel, R. C. Myers, and A. Sinha, JHEP 03, 084 J. Noronha-Hostler, P. Parotto, A. Pasztor, C. Ratti, (2009), arXiv:0812.2521 [hep-th]. and J. M. Stafford, Phys. Rev. D101, 034506 (2020), [567] Y. Kats and P. Petrov, JHEP 01, 044 (2009), arXiv:1910.14592 [hep-lat]. arXiv:0712.0743 [hep-th]. [593] A. Bazavov, F. Karsch, S. Mukherjee, and P. Pe- [568] X. O. Camanho, J. D. Edelstein, J. Maldacena, and treczky (USQCD), Eur. Phys. J. A55, 194 (2019), A. Zhiboedov, JHEP 02, 020 (2016), arXiv:1407.5597 arXiv:1904.09951 [hep-lat]. [hep-th]. [594] L.-G. Pang, K. Zhou, N. Su, H. Petersen, H. Stöcker, [569] Grozdanov, Saˇsoand van der Schee, Wilke, Phys. Rev. and X.-N. Wang, Nature Commun. 9, 210 (2018), Lett. 119, 011601 (2017), arXiv:1610.08976 [hep-th]. arXiv:1612.04262 [hep-ph]. [570] Folkestad, Asmund and Grozdanov, Saˇso and Ra- [595] L. R. Weih, M. Hanauske, and L. Rezzolla, Phys. Rev. jagopal, Krishna and van der Schee, Wilke, JHEP 12, Lett. 124, 171103 (2020), arXiv:1912.09340 [gr-qc]. 093 (2019), arXiv:1907.13134 [hep-th]. [596] J.-P. Blaizot, Rept. Prog. Phys. 80, 032301 (2017), [571] J. Cayuso, N. Ortiz, and L. Lehner, Phys. Rev. D 96, arXiv:1607.04448 [hep-ph]. 084043 (2017), arXiv:1706.07421 [gr-qc]. [597] H. Mäntysaari, N. Mueller, F. Salazar, and B. Schenke, [572] J. L. Ripley and F. Pretorius, Class. Quant. Grav. 36, Phys. Rev. Lett. 124, 112301 (2020), arXiv:1912.05586 134001 (2019), arXiv:1903.07543 [gr-qc]. [nucl-th]. [573] J. L. Ripley and F. Pretorius, Phys. Rev. D 101, 044015 [598] H. Mäntysaari and R. Venugopalan, Phys. Lett. B781, (2020), arXiv:1911.11027 [gr-qc]. 664 (2018), arXiv:1712.02508 [nucl-th]. [574] J. L. Ripley and F. Pretorius, Class. Quant. Grav. 37, [599] J. Adolfsson et al. (2020) arXiv:2003.10997 [hep-ph]. 155003 (2020), arXiv:2005.05417 [gr-qc]. [600] J. Churchill, L. Yan, S. Jeon, and C. Gale, in 28th [575] E. Iancu and A. Mukhopadhyay, JHEP 06, 003 (2015), International Conference on Ultrarelativistic Nucleus- arXiv:1410.6448 [hep-th]. Nucleus Collisions (Quark Matter 2019) Wuhan, China, [576] A. Mukhopadhyay, F. Preis, A. Rebhan, and S. A. November 4-9, 2019 (2020) arXiv:2001.11110 [hep-ph]. Stricker, JHEP 05, 141 (2016), arXiv:1512.06445 [hep- [601] L. Oliva, M. Ruggieri, S. Plumari, F. Scardina, G. X. th]. Peng, and V. Greco, Phys. Rev. C96, 014914 (2017), [577] A. Kurkela, A. Mukhopadhyay, F. Preis, A. Reb- arXiv:1703.00116 [nucl-th]. han, and A. Soloviev, JHEP 08, 054 (2018), [602] J. Berges, K. Reygers, N. Tanji, and R. Venugopalan, arXiv:1805.05213 [hep-ph]. Phys. Rev. C95, 054904 (2017), arXiv:1701.05064 [nucl- [578] C. Ecker, A. Mukhopadhyay, F. Preis, A. Rebhan, and th]. A. Soloviev, JHEP 08, 074 (2018), arXiv:1806.01850 [603] O. Garcia-Montero, (2019), arXiv:1909.12294 [hep-ph]. [hep-th]. [604] K. Boguslavski, A. Kurkela, T. Lappi, and J. Peuron, [579] T. Banks, M. R. Douglas, G. T. Horowitz, and E. J. (2020), arXiv:2005.02418 [hep-ph]. Martinec, (1998), arXiv:hep-th/9808016. [605] S. Mrówczyński, Eur. Phys. J. A 54, 43 (2018), [580] K. Skenderis and B. C. van Rees, Phys. Rev. Lett. 101, arXiv:1706.03127 [nucl-th]. 081601 (2008), arXiv:0805.0150 [hep-th]. [606] M. E. Carrington, A. Czajka, and S. Mrówczyński, [581] K. Skenderis and B. C. van Rees, JHEP 05, 085 (2009), (2020), arXiv:2001.05074 [nucl-th]. arXiv:0812.2909 [hep-th]. [607] J. H. Liu, S. Plumari, S. K. Das, V. Greco, and M. Rug- [582] P. M. Chesler and D. Teaney, (2011), arXiv:1112.6196 gieri, (2019), arXiv:1911.02480 [nucl-th]. [hep-th]. [608] R. Rapp and H. van Hees, in Quark-gluon plasma 4 [583] P. M. Chesler and D. Teaney, (2012), arXiv:1211.0343 (2010) pp. 111–206, arXiv:0903.1096 [hep-ph]. [hep-th]. [609] N. Brambilla, M. A. Escobedo, A. Vairo, and [584] V. Keranen and P. Kleinert, JHEP 04, 119 (2015), P. Vander Griend, Phys. Rev. D100, 054025 (2019), arXiv:1412.2806 [hep-th]. arXiv:1903.08063 [hep-ph]. [585] V. Keranen and P. Kleinert, Phys. Rev. D 94, 026010 [610] Y. Akamatsu, M. Asakawa, S. Kajimoto, and (2016), arXiv:1511.08187 [hep-th]. A. Rothkopf, JHEP 07, 029 (2018), arXiv:1805.00167 [586] J. Louko, D. Marolf, and S. F. Ross, Phys. Rev. D 62, [nucl-th]. 044041 (2000), arXiv:hep-th/0002111. [611] J.-P. Blaizot, J. Liao, and Y. Mehtar-Tani, Pro- [587] M. Headrick, V. E. Hubeny, A. Lawrence, and ceedings, 25th International Conference on Ultra- M. Rangamani, JHEP 12, 162 (2014), arXiv:1408.6300 Relativistic Nucleus-Nucleus Collisions (Quark Matter [hep-th]. 2015): Kobe, Japan, September 27-October 3, 2015, [588] V. Balasubramanian, A. Bernamonti, J. de Boer, Nucl. Phys. A956, 561 (2016), arXiv:1601.00308 [nucl- N. Copland, B. Craps, E. Keski-Vakkuri, B. Muller, th]. A. Schafer, M. Shigemori, and W. Staessens, Phys. Rev. [612] J.-P. Blaizot, E. Iancu, and Y. Mehtar-Tani, Phys. Rev. Lett. 106, 191601 (2011), arXiv:1012.4753 [hep-th]. Lett. 111, 052001 (2013), arXiv:1301.6102 [hep-ph]. [589] V. Balasubramanian, A. Bernamonti, J. de Boer, [613] A. Dumitru, Y. Nara, B. Schenke, and M. Strickland, N. Copland, B. Craps, E. Keski-Vakkuri, B. Muller, Phys. Rev. C78, 024909 (2008), arXiv:0710.1223 [hep- A. Schafer, M. Shigemori, and W. Staessens, Phys. Rev. ph]. D 84, 026010 (2011), arXiv:1103.2683 [hep-th]. [614] M. Asakawa, S. A. Bass, and B. Muller, Saturation [590] H.-T. Ding, F. Karsch, and S. Mukherjee, in Quark- the color glass condensate and the glasma: What have Gluon Plasma 5 , edited by X.-N. Wang (2016) pp. 1–65. we learned from RHIC? Proceedings, Workshop, Upton, [591] F. G. Gardim, G. Giacalone, M. Luzum, and Brookhaven, USA, May 10-12, 2010, Nucl. Phys. A854, J.-Y. Ollitrault, (2019), 10.1038/s41567-020-0846-4, 76 (2011), arXiv:1008.3496 [nucl-th]. arXiv:1908.09728 [nucl-th]. [615] M. E. Carrington, S. Mrówczyński, and B. Schenke, 71

Phys. Rev. C95, 024906 (2017), arXiv:1607.02359 [hep- [643] M. Rangamani and T. Takayanagi, Holographic ph]. Entanglement Entropy, Vol. 931 (Springer, 2017) [616] A. Ipp, D. I. Müller, and D. Schuh, (2020), arXiv:1609.01287 [hep-th]. arXiv:2001.10001 [hep-ph]. [644] D. Gross and F. Wilczek, Phys. Rev. D 8, 3633 (1973). [617] O. Garcia-Montero, N. Löher, A. Mazeliauskas, [645] H. D. Politzer, Phys. Rev. Lett. 30, 1346 (1973). J. Berges, and K. Reygers, (2019), arXiv:1909.12246 [646] L. Kofman, Lect. Notes Phys. 738, 55 (2008). [hep-ph]. [647] I. Bloch, J. Dalibard, and W. Zwerger, Reviews of Mod- [618] A. Dumitru, K. Dusling, F. Gelis, J. Jalilian-Marian, ern Physics 80, 885–964 (2008). T. Lappi, and R. Venugopalan, Phys. Lett. B697, 21 [648] E. Haller, R. Hart, M. J. Mark, J. G. Danzl, L. Re- (2011), arXiv:1009.5295 [hep-ph]. ichsöllner,M. Gustavsson, M. Dalmonte, G. Pupillo, [619] T. Altinoluk and N. Armesto, (2020), arXiv:2004.08185 and H.-C. Nägerl, Nature 466, 597 (2010). [hep-ph]. [649] M. Gring, M. Kuhnert, T. Langen, T. Kitagawa, [620] E. Shuryak and I. Zahed, Phys. Rev. C 88, 044915 B. Rauer, M. Schreitl, I. E. Mazets, D. Adu Smith, (2013), arXiv:1301.4470 [hep-ph]. E. Demler, and J. Schmiedmayer, Science 337, 1318 [621] A. Bzdak and K. Dusling, Phys. Rev. C 94, 044918 (2012). (2016), arXiv:1607.03219 [hep-ph]. [650] C.-L. Hung, V. Gurarie, and C. Chin, Science 341, 1213 [622] K. Dusling and R. Venugopalan, Phys. Rev. D 87, (2013). 094034 (2013), arXiv:1302.7018 [hep-ph]. [651] T. Langen, S. Erne, R. Geiger, B. Rauer, T. Schweigler, [623] M. Martinez, M. D. Sievert, and D. E. Wertepny, JHEP M. Kuhnert, W. Rohringer, I. E. Mazets, T. Gasenzer, 02, 024 (2019), arXiv:1808.04896 [hep-ph]. and J. Schmiedmayer, Science 348, 207 (2015). [624] A. Dumitru, V. Skokov, and T. Stebel, Phys. Rev. D [652] N. Navon, A. L. Gaunt, R. P. Smith, and Z. Hadzibabic, 101, 054004 (2020), arXiv:2001.04516 [hep-ph]. Science 347, 167 (2015). [625] J. L. Albacete and C. Marquet, Prog. Part. Nucl. Phys. [653] N. Navon, A. L. Gaunt, R. P. Smith, and Z. Hadzibabic, 76, 1 (2014), arXiv:1401.4866 [hep-ph]. Nature 539, 72 (2016). [626] C. Andres, N. Armesto, H. Niemi, R. Paatelainen, [654] M. F. Parsons, A. Mazurenko, C. S. Chiu, G. Ji, and C. A. Salgado, Phys. Lett. B 803, 135318 (2020), D. Greif, and M. Greiner, Science 353, 1253 (2016). arXiv:1902.03231 [hep-ph]. [655] T. Schweigler, V. Kasper, S. Erne, I. Mazets, B. Rauer, [627] A. Huang, Y. Jiang, S. Shi, J. Liao, and P. Zhuang, F. Cataldini, T. Langen, T. Gasenzer, J. Berges, and Phys. Lett. B 777, 177 (2018), arXiv:1703.08856 [hep- J. Schmiedmayer, Nature 545, 323 (2017). ph]. [656] H. Bernien, S. Schwartz, A. Keesling, H. Levine, A. Om- [628] F. Becattini and M. A. Lisa, Ann. Rev. Nucl. Part. Sci. ran, H. Pichler, S. Choi, A. S. Zibrov, M. Endres, 70, 395 (2020), arXiv:2003.03640 [nucl-ex]. M. Greiner, et al., Nature 551, 579 (2017). [629] R. K. Ellis et al., (2019), arXiv:1910.11775 [hep-ex]. [657] S. Eckel, A. Kumar, T. Jacobson, I. B. Spielman, and [630] G. Roland (sPHENIX), PoS HardProbes2018, 013 G. K. Campbell, Phys. Rev. X 8, 021021 (2018). (2019). [658] J. Hu, L. Feng, Z. Zhang, and C. Chin, Nature Physics [631] C. Gogolin and J. Eisert, Reports on Progress in Physics 15, 785 (2019). 79, 056001 (2016). [659] L. Feng, J. Hu, L. W. Clark, and C. Chin, Science 363, [632] L. D’Alessio, Y. Kafri, A. Polkovnikov, and M. Rigol, 521 (2019). Advances in Physics 65, 239–362 (2016). [660] P. A. Murthy, N. Defenu, L. Bayha, M. Holten, P. M. [633] F. Borgonovi, F. Izrailev, L. Santos, and V. Zelevinsky, Preiss, T. Enss, and S. Jochim, Science 365, 268 (2019). Physics Reports 626, 1–58 (2016). [661] A. Keesling, A. Omran, H. Levine, H. Bernien, H. Pich- [634] K. Dusling, T. Epelbaum, F. Gelis, and R. Venu- ler, S. Choi, R. Samajdar, S. Schwartz, P. Silvi, gopalan, Phys. Rev. D 86, 085040 (2012), S. Sachdev, P. Zoller, M. Endres, M. Greiner, V. Vuletić, arXiv:1206.3336 [hep-ph]. and M. D. Lukin, Nature 568, 207 (2019). [635] D. I. Podolsky, G. N. Felder, L. Kofman, and M. Peloso, [662] M. Prüfer,T. V. Zache, P. Kunkel, S. Lannig, A. Bon- Phys. Rev. D 73, 023501 (2006), arXiv:hep-ph/0507096. nin, H. Strobel, J. Berges, and M. K. Oberthaler, [636] M. Moeckel and S. Kehrein, Physical Review Letters arXiv:1909.05120 (2019). 100 (2008), 10.1103/physrevlett.100.175702. [663] T. V. Zache, T. Schweigler, S. Erne, J. Schmied- [637] T. Langen, T. Gasenzer, and J. Schmiedmayer, Journal mayer, and J. Berges, Phys. Rev. X 10, 011020 (2020), of Statistical Mechanics: Theory and Experiment 2016, arXiv:1909.12815 [cond-mat.quant-gas]. 064009 (2016). [664] E. A. Martinez, C. A. Muschik, P. Schindler, D. Nigg, [638] T. Mori, T. N. Ikeda, E. Kaminishi, and M. Ueda, A. Erhard, M. Heyl, P. Hauke, M. Dalmonte, T. Monz, Journal of Physics B: Atomic, Molecular and Optical P. Zoller, and et al., Nature 534, 516–519 (2016). Physics 51, 112001 (2018). [665] N. Klco, E. F. Dumitrescu, A. J. McCaskey, T. D. [639] D. Smith, M. Gring, T. Langen, M. Kuhnert, B. Rauer, Morris, R. C. Pooser, M. Sanz, E. Solano, P. Lougov- R. Geiger, T. Kitagawa, I. Mazets, E. Demler, and ski, and M. J. Savage, Physical Review A 98 (2018), J. Schmiedmayer, New J. Phys. 15, 075011 (2013), 10.1103/physreva.98.032331. arXiv:1212.4645 [cond-mat.quant-gas]. [666] N. Mueller, A. Tarasov, and R. Venugopalan, (2019), [640] J. Eisert, M. Cramer, and M. Plenio, Rev. Mod. Phys. arXiv:1908.07051 [hep-th]. 82, 277 (2010), arXiv:0808.3773 [quant-ph]. [667] N. Mueller, A. Tarasov, and R. Venugopalan, in 28th [641] P. Calabrese and J. Cardy, J. Phys. A 42, 504005 (2009), International Conference on Ultrarelativistic Nucleus- arXiv:0905.4013 [cond-mat.stat-mech]. Nucleus Collisions (Quark Matter 2019) Wuhan, China, [642] H. Casini and M. Huerta, J. Phys. A 42, 504007 (2009), November 4-9, 2019 (2020) arXiv:2001.11145 [hep-th]. arXiv:0905.2562 [hep-th]. [668] C. Kokail, C. Maier, R. van Bijnen, T. Brydges, M. K. 72

Joshi, P. Jurcevic, C. A. Muschik, P. Silvi, R. Blatt, [699] M. Rigol, V. Dunjko, and M. Olshanii, Nature 452, C. F. Roos, and et al., Nature 569, 355–360 (2019). 854–858 (2008), arXiv:0708.1324 [cond-mat.stat-mech]. [669] A. Mil, T. V. Zache, A. Hegde, A. Xia, R. P. Bhatt, [700] P. Calabrese and J. L. Cardy, J. Stat. Mech. 0504, M. K. Oberthaler, P. Hauke, J. Berges, and F. Jen- P04010 (2005), arXiv:cond-mat/0503393. drzejewski, Science 367, 1128–1130 (2020). [701] J. Abajo-Arrastia, J. Aparicio, and E. Lopez, JHEP [670] C. Chin, R. Grimm, P. Julienne, and E. Tiesinga, Re- 11, 149 (2010), arXiv:1006.4090 [hep-th]. views of Modern Physics 82, 1225–1286 (2010). [702] H. Liu and S. J. Suh, Phys. Rev. D 89, 066012 (2014), [671] C. Eigen, J. A. P. Glidden, R. Lopes, E. A. Cornell, arXiv:1311.1200 [hep-th]. R. P. Smith, and Z. Hadzibabic, Nature 563, 221–224 [703] J. S. Cotler, M. P. Hertzberg, M. Mezei, and M. T. (2018). Mueller, JHEP 11, 166 (2016), arXiv:1609.00872 [hep- [672] K. Aamodt et al. (ALICE), Phys. Rev. Lett. 105, th]. 252302 (2010), arXiv:1011.3914 [nucl-ex]. [704] A. M. Kaufman, M. E. Tai, A. Lukin, M. Rispoli, [673] G. Aad et al. (ATLAS), Phys. Rev. C 86, 014907 (2012), R. Schittko, P. M. Preiss, and M. Greiner, Science 353, arXiv:1203.3087 [hep-ex]. 794–800 (2016), arXiv:1603.04409 [quant-ph]. [674] S. Chatrchyan et al. (CMS), Phys. Rev. C 84, 024906 [705] V. Alba and P. Calabrese, Proceedings of the Na- (2011), arXiv:1102.1957 [nucl-ex]. tional Academy of Sciences 114, 7947–7951 (2017), [675] D. Son, Phys. Rev. D 78, 046003 (2008), arXiv:1608.00614 [cond-mat.str-el]. arXiv:0804.3972 [hep-th]. [706] J. Berges, S. Floerchinger, and R. Venugopalan, Phys. [676] K. Balasubramanian and J. McGreevy, Phys. Rev. Lett. Lett. B 778, 442 (2018), arXiv:1707.05338 [hep-ph]. 101, 061601 (2008), arXiv:0804.4053 [hep-th]. [707] J. Berges, S. Floerchinger, and R. Venugopalan, JHEP [677] A. Adams, L. D. Carr, T. Schäfer, P. Steinberg, 04, 145 (2018), arXiv:1712.09362 [hep-th]. and J. E. Thomas, New J. Phys. 14, 115009 (2012), [708] F. Becattini, Z. Phys. C 69, 485 (1996). arXiv:1205.5180 [hep-th]. [709] A. Andronic, F. Beutler, P. Braun-Munzinger, [678] G. Dvali and C. Gomez, Fortsch. Phys. 61, 742 (2013), K. Redlich, and J. Stachel, Phys. Lett. B 675, 312 arXiv:1112.3359 [hep-th]. (2009), arXiv:0804.4132 [hep-ph]. [679] L. N. Lipatov, Nucl. Phys. B365, 614 (1991). [710] Z. Tu, D. E. Kharzeev, and T. Ullrich, Phys. Rev. Lett. [680] D. Amati, M. Ciafaloni, and G. Veneziano, Phys. Lett. 124, 062001 (2020), arXiv:1904.11974 [hep-ph]. B 197, 81 (1987). [711] C. Ecker, D. Grumiller, P. Stanzer, S. A. Stricker, [681] G. Dvali, C. Gomez, R. S. Isermann, D. Lüst, and W. van der Schee, JHEP 11, 054 (2016), and S. Stieberger, Nucl. Phys. B893, 187 (2015), arXiv:1609.03676 [hep-th]. arXiv:1409.7405 [hep-th]. [712] R. Orus, Annals Phys. 349, 117 (2014), arXiv:1306.2164 [682] H. Kawai, D. C. Lewellen, and S. H. H. Tye, Nucl. [cond-mat.str-el]. Phys. B269, 1 (1986). [713] M. B. Hastings, Physical Review B 73 (2006), [683] A. Addazi, M. Bianchi, and G. Veneziano, JHEP 02, 10.1103/physrevb.73.085115, arXiv:cond-mat/0508554 111 (2017), arXiv:1611.03643 [hep-th]. [cond-mat.str-el]. [684] Z. Bern, J. J. Carrasco, M. Chiodaroli, H. Johansson, [714] G. Vidal, Phys. Rev. Lett. 101, 110501 (2008), and R. Roiban, (2019), arXiv:1909.01358 [hep-th]. arXiv:quant-ph/0610099. [685] A. Sabio Vera, E. Serna Campillo, and M. A. Vazquez- [715] P. Corboz, Phys. Rev. B 94, 035133 (2016), Mozo, JHEP 03, 005 (2012), arXiv:1112.4494 [hep-th]. arXiv:1605.03006 [cond-mat.str-el]. [686] Z. Liu, JHEP 02, 112 (2019), arXiv:1811.11710 [hep-th]. [716] P. Corboz, Phys. Rev. B 93, 045116 (2016). [687] R. Monteiro, D. O’Connell, and C. D. White, JHEP [717] L. Vanderstraeten, J. Haegeman, P. Corboz, and 12, 056 (2014), arXiv:1410.0239 [hep-th]. F. Verstraete, Phys. Rev. B 94, 155123 (2016). [688] W. D. Goldberger and A. K. Ridgway, Phys. Rev. D95, [718] P. Corboz, P. Czarnik, G. Kapteijns, and L. Taglia- 125010 (2017), arXiv:1611.03493 [hep-th]. cozzo, Phys. Rev. X 8, 031031 (2018). [689] G. Dvali and R. Venugopalan, (2021), arXiv:2106.11989 [719] M. Rader and A. M. Läuchli, Phys. Rev. X 8, 031030 [hep-th]. (2018). [690] A. Di Piazza, C. Muller, K. Z. Hatsagortsyan, and [720] M. C. Bañulsand K. Cichy, Rept. Prog. Phys. 83, C. H. Keitel, Rev. Mod. Phys. 84, 1177 (2012), 024401 (2020), arXiv:1910.00257 [hep-lat]. arXiv:1111.3886 [hep-ph]. [721] T. Pichler, M. Dalmonte, E. Rico, P. Zoller, and [691] J. Berges and J. Cox, Phys. Lett. B 517, 369 (2001), S. Montangero, Phys. Rev. X 6, 011023 (2016), arXiv:hep-ph/0006160. arXiv:1505.04440 [cond-mat.quant-gas]. [692] S. Juchem, W. Cassing, and C. Greiner, Nucl. Phys. A [722] B. Buyens, J. Haegeman, H. Verschelde, F. Verstraete, 743, 92 (2004), arXiv:nucl-th/0401046. and K. Van Acoleyen, Phys. Rev. X 6, 041040 (2016), [693] A. Arrizabalaga, J. Smit, and A. Tranberg, Phys. Rev. arXiv:1509.00246 [hep-lat]. D 72, 025014 (2005), arXiv:hep-ph/0503287. [723] B. Buyens, J. Haegeman, F. Hebenstreit, F. Verstraete, [694] J. Berges, S. Borsanyi, and J. Serreau, Nucl. Phys. B and K. Van Acoleyen, Phys. Rev. D 96, 114501 (2017), 660, 51 (2003), arXiv:hep-ph/0212404. arXiv:1612.00739 [hep-lat]. [695] L. Shen, J. Berges, J. Pawlowski, and A. Rothkopf, [724] M. C. Bañuls, M. P. Heller, K. Jansen, J. Knaute, (2020), arXiv:2003.03270 [hep-ph]. and V. Svensson, Phys. Rev. Res. 2, 033301 (2020), [696] G. Aarts, G. F. Bonini, and C. Wetterich, Phys. Rev. arXiv:1912.08836 [hep-th]. D 63, 025012 (2000), arXiv:hep-ph/0007357. [725] S. Ryu and T. Takayanagi, Phys. Rev. Lett. 96, 181602 [697] J. M. Deutsch, Phys. Rev. A 43, 2046 (1991). (2006), arXiv:hep-th/0603001. [698] M. Srednicki, Physical Review E 50, 888–901 (1994), [726] V. E. Hubeny, M. Rangamani, and T. Takayanagi, arXiv:cond-mat/9403051 [cond-mat]. JHEP 07, 062 (2007), arXiv:0705.0016 [hep-th]. 73

[727] A. Lewkowycz and J. Maldacena, JHEP 08, 090 (2013), [730] A. Almheiri, T. Hartman, J. Maldacena, E. Shaghou- arXiv:1304.4926 [hep-th]. lian, and A. Tajdini, JHEP 05, 013 (2020), [728] X. Dong, A. Lewkowycz, and M. Rangamani, JHEP arXiv:1911.12333 [hep-th]. 11, 028 (2016), arXiv:1607.07506 [hep-th]. [731] G. Baym et al., Bear Mountain workshop report (1975). [729] G. Penington, S. H. Shenker, D. Stanford, and Z. Yang, [732] D. Spitz, J. Berges, M. K. Oberthaler, and A. Wien- (2019), arXiv:1911.11977 [hep-th]. hard, (2020), arXiv:2001.02616 [cond-mat.quant-gas].