WorksPhysics and Astronomy
Thermalization of non-abelian gauge theories at next-to-leading order
Abstract
We provide the first next-to-leading-order (NLO) weak-coupling description of the thermalization process of far-from-equilibrium systems in non-abelian gauge theory. We study isotropic systems starting from either over- or under-occupied initial conditions and follow their time evolution towards thermal equilibrium by numerically solving the QCD effective kinetic theory at NLO accuracy. We find that the NLO corrections remain well under control for a wide range of couplings and that the overall effect of NLO corrections is to reduce the time needed to reach thermal equilibrium in the systems considered.
Introduction
How do non-abelian gauge fields pushed far from equilibrium approach the thermal state is a central question in several branches of physics. In cosmology, far-from-equilibrium configurations of non-abelian fields may be produced during (p)reheating [1, 2, 3], caused by first order transitions [1, 4, 5], and are a necessary ingredient for baryogenesis [6, 7]. In all of these cases, an understanding of thermalization rates is required for quantitative descriptions of these phenomena [8, 9]. In the early stages of ultra-relativistic heavy-ion collisions a far-from-equilibrium system of gluons and quarks is created. If and how this system reaches local thermal equilibrium plays a crucial part in the phenomenological modeling of the collisions. The recent discussion about the physical origin of collectivity in smaller collision systems created in p-Pb and light-ion collisions [10] further emphasizes the importance of a quantitative understanding of thermalization in far-from equilibrium systems. Furthermore, connections between systems created in atomic physics experiments and gauge field models are being actively studied (see, e.g., [11, 12, 13]).
While first-principles non-perturbative lattice simulation of far-from-equilibrium quantum systems remains elusive, the past years have witnessed progress in methods relying on different approximations — see [14, 15] for recent reviews. On one hand, holographic methods have been successful in the description of Super Yang-Mills theory in the limit of large number of colors and large t’Hooft coupling . These studies have advanced to a mature level, even including sub-leading corrections in the t’Hooft coupling [16, 17]. On the other hand, weak-coupling methods are available for generic theories and have also been widely studied. The first works studying thermalization of pure Yang-Mills theory from simple initial conditions [18] have been extended to Quantum Chromodynamics (QCD) [19, 20, 21, 22] and calculations based on this physical picture have been extended to describe systems of enough complexity to be used in realistic phenomenological modelling of heavy-ion collisions [23, 24] and even in light-ion collisions [25]. This picture has also been applied for extracting estimates of thermalization times during reheating [26, 27, 28]. These studies have, however, been at best limited to leading order (LO) in the coupling constant and it is important to improve the accuracy — and in particular, to test the validity and robustness of the weak-coupling expansion — by finding the first subleading corrections to the weak-coupling results. In this paper we provide the first numerical description of thermalization from simple, isotropic initial conditions at next-to-leading order (NLO).
A direct diagrammatic description of thermalization is prohibitively difficult due to a need to resum diagrams of all loop orders even to obtain a LO result in [29]. At this order, this resummation can be elegantly performed by considering an effective kinetic theory (EKT) that contains all the necessary processes required for a leading-order description of the evolution of the particle distribution functions [30]. In gauge theories, the derivation of the of the collision kernels required for the EKT is further non-perturbative [31]. This arises from the Bose-enhancement of “soft” infrared modes at the plasma screening scale , whose interactions with the typical “hard” particles (with ) are non-perturbative. This, combined with the well-known soft and collinear divergences of the unresummed QCD cross sections, necessitates a resummation that incorporates the physics of in-medium screening [32] and Landau-Pomeranchuk-Migdal (LPM) [33, 34, 35] suppression in the QCD effective kinetic theory [31].
The physical picture of EKT can be extended to next-to-leading-order accuracy. The NLO corrections arise from the interactions among the soft modes. The resulting terms are suppressed only by , in contrast to in vacuum field theory. While various NLO corrections to equilibrium and near-equilibrium quantities have been computed [36, 37, 38, 39, 40, 41], the framework has not until now been pushed to study thermalization of far-from-equilibrium systems.
In this letter we extend the NLO formulation of EKT to isotropic far-from-equilibrium systems and apply it to numerically describe thermalization of two specific systems initialized with either under- or overoccupied initial conditions studied in LO in [18]. In the idealized limit of weak-coupling, thermalization of under-occupied systems (including those created in heavy-ion collisions) proceeds through the process of bottom-up thermalization [42, 43]. The starting point of bottom-up thermalization is an ensemble of too few particles particles with too high momenta compared to thermal equilibrium with the final temperature . In the bottom-up process, the collisions among these few hard particles lead to soft radiation that forms a soft thermal bath with a temperature . The further interaction between the hard particles and soft thermal bath eventually causes a radiational break-up of the hard particles that heats the soft thermal bath to its final temperature . We will consider how this picture is quantitatively changed when pushing to finite and small values of . We see that the NLO corrections are under quantitative control for , and we observe that the NLO corrections make thermalization faster.
As a second system, we consider an overoccupied, initial state in its self-similar scaling solution, that is, a non-thermal, time-dependent fixed point that is rapidly reached from any overoccupied initial condition – see [43, 44, 45, 46, 47, 48]. We find that in this case too NLO corrections bring about a faster thermalization and that, while a bit larger than for the underoccupied scenario, they remain under control over a wide range of couplings.
Setup
Leading Order Kinetic Theory
In the weak coupling limit , the evolution of modes with perturbative occupancies and whose momenta are larger than the screening scale can be described to leading order in by an effective kinetic equation for the color averaged gauge boson distribution function [31]
The elastic scattering and collinear splitting parts of the collision operator — whose precise forms are given in App. § 2.12 — depend respectively on effective matrix elements and splitting rates which have been discussed in detail in refs. [18, 31, 48, 49]. The elastic collision term includes LO screening effects by consistently regulating the Coulombic divergence in and channels at the scale . The splitting kernel includes the effects of LPM suppression [33, 34, 35, 50, 51, 52] which regulate collinear divergences. These effects depend on and an effective temperature
which are self-consistently calculated during the simulation. The effective theory contains no free parameters besides the coupling constant . Our numerical implementation is the discrete- method of [48].
Next-to-Leading Order Kinetic Theory
NLO corrections to this kinetic picture have been derived in [53] for a dilute set of high-energy “jet” partons interacting with a thermal medium and in [40] at first order in the departure from equilibrium, suited for the determination of transport coefficients. These corrections arise from the self-interactions of soft gluons with appearing in the internal lines in the diagrammatic computation of the collision kernels. At this order, these soft gluons can be treated as classical fields, retaining only the -enhanced part of their equilibrium distribution, and their contributions can be treated within the Hard Thermal Loop (HTL) effective theory [32]. Furthermore, they can be treated analytically without recurring to brute-force HTL computations, owing to the light-cone techniques introduced in [37, 53, 54] (see [55] for a more pedagogical exposition).
These calculations can be extended also to some far-from-equilibrium systems. As is known (see e.g. [18, 40, 56, 57, 58]), for , the collinear splittings are very effective and rapidly build up a soft thermal tail. That is, they ensure that . This, in turn, implies that, in cases with isotropic initial conditions, the collision operator can naturally accommodate the NLO corrections derived in [40, 53]. The NLO corrections are suppressed — with respect to the LO terms in Eq. (1) — by a factor of . This arises from the product of the naive suppression factor for loops with the occupation number at the scale , that is .3 Isotropy further ensures that the terms which have not been determined in the “almost NLO” determination of [40] do not contribute here, guaranteeing that what we are presenting is the full set of NLO modifications.
These contributions, which we discuss in more detail in App. § 2.1.§ 5.2, consist of new scattering processes and modifications to the LO ones, as shown in [40, 53]. The rate of soft scattering is modified. This modification, and an correction to the in-medium dispersion, also provide an shift in the rate. This splitting rate must furthermore be corrected wherever one participant becomes soft or when the opening angle becomes less collinear.
A rather general property of kinetic theory resummations is that it is possible to construct collision operators that are equivalent up to a given order but differ by subleading corrections. This was exploited in [18, 48] to construct a LO implementation that is numerically well-behaved, thanks to a partial resummation of higher-order effects: a subtraction will thus be needed to ensure that no double-counting takes place.
We exploit this same property at NLO: as we shall show in detail in App A2, we construct two separate collision operators, both including all effects but differing at higher orders. We call these two schemes scheme 1 and scheme 2. The difference in the results obtained from these two, as well as their spread from the LO results, can be taken as an estimate of the uncertainty, in particular when extrapolating towards regions where the expansion parameters are no longer small. One such region is thus , while another arises in the region where . As is known (see the detailed discussion in [49]), the formation time for a collinear splitting process grows with , making the splitting process sensitive not just to the frequent soft scatterings exchanging , but also to the rarer higher-momentum exchanges. For our form of the LO and NLO rate, which only includes scatterings, becomes inaccurate. As we elaborate in App A2, our first implementation, scheme 1, treats these processes with no partial resummation of higher order effects and the collision kernel is more prone to extrapolate to (unphysical) negative values than our second, non-strict implementation, scheme 2.
initial conditions
For the underoccupied initial condition we will use a gaussian form centered around a characteristic momentum scale , as in [18]. In order to mimic the situation in the last stage of bottom-up thermalization (and for numerical stability), we embed this distribution of hard particles in a soft thermal bath that carries 10% of the total energy density
where and are and . is the equilibrium Bose–Einstein distribution.
In the overoccupied case we let the system evolve from the scaling solution [48]
where and . For this initial condition one has and a direct energy cascade from the IR to the UV takes place. We choose and an initial time such that .
RESULTS
The thermalization processes of systems initialised with Eqs. (3) and (4) are displayed in Figure 1 for and for the underoccupied case (left panel) and for the overoccupied case (right panel). Both are evolved with the scheme 2 prescription.

Figure 1. Time evolution from the initial conditions (3)–(4) in solid blue lines to the final equilibrium state in solid black. The dotted and dashed lines show intermediate steps upon solving the NLO kinetic theory (scheme 2). The values of the couplings are and respectively.
The NLO evolutions of these systems exhibit the same qualitative features as their LO counterparts. In the case of underoccupied initial conditions, the NLO evolution shows the characteristic features of bottom-up thermalization: one can see the hard particles lose energy through the radiational cascade heating the soft thermal bath. Eventually the system thermalizes as the hard particles are quenched in the thermal bath [43]. In the case of the overoccupied initial conditions, the direct energy cascade to the UV seen at LO is also seen at NLO. The departure from the scaling solution takes place once , corresponding to .
In order to determine thermalization times of these systems, we characterise them in terms of effective temperatures
which all coincide with in equilibrium but differ for non-equilibrium systems. We then define a (kinetic) thermalization time by demanding that the different effective temperatures are sufficiently close to each other. Specifically, we define the (kinetic) thermalization time using the condition [20]
where we use “+” and “-” for under- and overoccupied systems, respectively. For the underoccupied (overoccupied) system in Figure 1, this condition is fulfilled for (), denoted by the green dashed line. At this point most of the energy is in the thermal bath, rather than in the initial UV (IR) structure.
We have determined this thermalization time for different values of the coupling constant and, in the underoccupied case, a variety of initial momenta , using both the LO as well as the two NLO schemes; the under- and overoccupied-case results are documented in Tab. I and II and displayed in Figure 2. Our main findings are that

Figure 2. Equilibration times as a function of the coupling and, in the underoccupied case, the initial UV scale. The shaded band between the two NLO schemes can be taken as a first indication of the theory uncertainty. The coupling for which in thermal equilibrium is indicated by a vertical gray line.
the qualitative effect of the NLO corrections is to reduce the time required for thermalization
and that NLO corrections are well under control for a wide range of coupling constants.
In the regime of small values of — corresponding to in equilibrium, so that the scale separations assumed in the derivation of the kinetic theory are fulfilled — the NLO corrections constitute merely a 5% and 20% reduction of the thermalization time in the under- and overoccupied cases. It is reassuring to observe that, in both scenarios, results from the two NLO schemes are close to each other compared to the overall size of the NLO correction. In the -limit, the difference between the two NLO schemes vanishes faster than their difference to LO. This demonstrates that the observed differences from the LO are true NLO corrections and are not contaminated by the scheme differences that affect the result beyond the NLO accuracy.
Extrapolating to higher values of , we see that in the underoccupied case the difference between the two NLO schemes becomes comparable to the size of the NLO correction itself. This indicates quantitative sensitivity to corrections beyond NLO. However, taking the difference of the two schemes as an estimate of the uncertainty, we observe that, strikingly, the corrections remain below 10%-level even for these large value of the coupling. In the overoccupied case the correction reaches 40%-level, with only a moderate spread between the two schemes.
At leading order, the underoccupied thermalization time is parametrically (up to logarithms) of order [43], related to the democratic splitting time of the particles at the scale in a thermal bath with temperature . At NLO, corrections are expected to arise at the relative order . We find that LO thermalization time given in Eq. (5) is well described for by a fit3
For small and the NLO correction in both schemes is approximately given by
and similarly for the overoccupied case
Conclusions
The poor convergence of the perturbative series for several different quantities has limited its usefulness in many phenomenological applications. The soft corrections studied here are responsible for this poor convergence for many observables such as transport coefficients [40, 41] or momentum broadening coefficients [36, 37]. For these quantities NLO corrections completely overtake the LO results for . On the contrary, in the present case of isotropic thermalisation, these soft corrections seem to be well under control; the corrections are at most of order 40% for the overoccupied case at . These findings are ostensibly in sharp contrast.
However, it is important to note that [40] found NLO corrections to transport coefficients to be numerically dominated by the NLO contribution to the isotropization rate governed by the transverse momentum broadening coefficient (which obtains a large positive NLO correction [37]). The key difference with respect to the present case is that, in an isotropic setting, the dependence on is significantly reduced. Instead of explicitly entering the calculation as an anisotropisation rate, only appears in our case as the source of splittings; it does make their rate larger, but its numerical effect is moderated by the fact that, parametrically, the LPM-suppressed splitting rate is , whereas isotropisation is . Furthermore, the other NLO corrections to splitting arising from a soft participant, a wider-angle emission or a rarer larger-momentum radiation-inducing scattering tend to decrease the rate, partially cancelling the -driven increase. This partial cancellation was already seen in the thermal photon production rate — another isotropic observable — which also shows moderate NLO corrections [39]. This is suggestive of a pattern which we think deserves further investigations. We note that some of these issues may be ameliorated in thermal equilibrium by non-perturbative determination of the soft contributions developed in [59, 60, 61, 62]. However, it is currently not known how these methods could be extended to far-from-equilibrium systems.
Lastly, we point out that, when trying to apply our methods to anisotropic systems, such as one undergoing Bjorken (1D) expansion, we would necessarily need to include the isotropizing effect of transverse momentum broadening, further compounded by the emergence of plasma instabilities [43, 63, 64, 65, 66, 67]. However, in the final stages of the bottom-up thermalization of heavy-ion collisions, the hard particles interact mainly with the isotropic soft thermal bath. This suggests that the methods developed here may be extended to improve the phenomenological description of the bottom-up hydrodynamization in heavy-ion collision.
| 20 | 1 | 503.4 | 465.2 | 473.2 | 35 | 0.1 | 623.4 | 614.5 | 615.7 |
| 40 | 1 | 818.7 | 784.1 | 791.8 | 35 | 0.5 | 707.5 | 683.3 | 687.6 |
| 60 | 1 | 1060.0 | 1039.1 | 1044.4 | 35 | 1 | 749.3 | 712.5 | 720.7 |
| 80 | 1 | 1263.9 | 1261.5 | 1263.2 | 35 | 5 | 859.4 | 764.5 | 803.4 |
| 100 | 1 | 1443.4 | 1462.2 | 1459.8 | 35 | 10 | 910.5 | 774.3 | 849.5 |
| 20 | 5 | 588.4 | 489.5 | 528.8 | 50 | 0.1 | 798.9 | 791.7 | 793.3 |
| 40 | 5 | 934.6 | 845.5 | 882.4 | 50 | 0.5 | 897.3 | 878.6 | 882.0 |
| 60 | 5 | 1193.8 | 1142.4 | 1163.5 | 50 | 1 | 945.5 | 916.9 | 923.6 |
| 80 | 5 | 1409.5 | 1410.4 | 1408.9 | 50 | 5 | 1070.9 | 998.6 | 1028.7 |
| 100 | 5 | 1599.2 | 1661.6 | 1630.4 | 50 | 10 | 1129.1 | 1027.6 | 1086.4 |
Table 1. Table of thermalization times of underoccupied initial conditions with different and values of the coupling .
| 0.01 | 0.03 | 0.06 | 0.1 | 0.3 | 0.6 | 1 | 3 | 6 | 10 | |
| 40.9 | 46.2 | 50.4 | 54.0 | 63.2 | 70.4 | 76.2 | 89.5 | 97.1 | 101.0 | |
| 40.5 | 45.4 | 49.1 | 52.0 | 58.8 | 62.9 | 65.4 | 67.1 | 64.1 | 59.3 | |
| 40.5 | 45.4 | 49.1 | 52.1 | 59.1 | 63.6 | 66.6 | 71.1 | 71.7 | 70.7 |
Table 2. Table of thermalization times of overoccupied initial conditions with different values of the coupling .
Acknowledgments
J.G. acknowledges support by a PULSAR grant from the Région Pays de la Loire. S.I. and Y.F. were supported in part by the National Natural Science Foundation of China under Grant Nos. 11935007, 11221504, 11890714 and 11861131009. We are grateful to Peter Arnold for useful conversations.
Definitions and implementations of the kinetic theory
Leading order kinetic theory
The precise form of the LO collision operator reads4
The elastic kernel given in Eq. (9) depends on the effective in-medium matrix element . As the vacuum elastic scattering has a (and ) infrared divergence, with momentum transfer , it makes the soft small angle scattering contribution to the scattering kernel diverge. This divergence is, however, regulated by the physics of in-medium screening. A prescription that is accurate to leading order was given in [48] by the replacement
where at LO is fixed to , so as to reproduce the LO longitudinal momentum diffusion coefficient [53, 55].
The effective medium-induced collinear splitting/merging matrix element is given by [31, 68]
with the momentum fraction and where resums an arbitrary number of soft elastic scatterings with the medium. It depends on two dimensionless variables
where is the LO mass for gluons with . Parametrically is the ratio squared of the formation time of the splitting process and of the elastic scattering rate . is the solution to this differential equation [31, 53, 68]
is the Fourier transform of the soft scattering rate,
In an isotropic medium it reads
By rescaling and , the coefficient of the second line fo Eq. (14) becomes proportional to . The method presented in [69] is then used for the numerical solution.
Next-to-leading order kinetic theory
Let us start by discussing the corrections to Eq. (12). As shown in [53], its form remains valid at NLO, but the LPM resummation in Eq. (14) must include two corrections. The dispersion relation gets shifted to and the soft scattering kernel is modified in . For an isotropic state with a soft thermal tail, the equilibrium results for [38] and [37, 39] can be used with the replacement , . The former reads
In our first implementation, i.e. scheme 1, we treat and as perturbations to their LO counterparts. Hence is perturbed as , and the latter is computed exactly as in App. E of [53].5 The resulting can become appreciable when extrapolated to large values of and . As per its definition, large values of correspond to formation times larger than the mean free time for soft scatterings, so that rarer, harder scatterings, which are not included in the form Eq. (16) of the scattering kernel, would have a chance to occur. As shown in [49], for scatterings with would need to be included, which is far from trivial in an off-equilibrium setting. At LO one can however expect, as in equilibrium, that the approximation introduced by extrapolating Eq. (16) to amounts to an overestimate of at the 10–20% level. That happens because large values of privilege the small- form of , which at leading order is approximated by , with a coefficient that varies in equilibrium by 25% between and .
At NLO this translates for large into a strong sensitivity on , which is the Fourier transform of the subleading, , form of the collision kernel for . Its negative coefficient, for large enough and , makes negative. We thus propose a second implementation, scheme 2, so that the difference between the two can be taken as a proxy for the reliability of these extrapolations. In this second implementation, we do not treat and as perturbations. We rather solve
where we have defined the mass self-consistently as
i.e. the positive solution to , so that, by resumming some higher-order terms, it stays positive at large . In a similar spirit, we have implemented the collision kernel as
so that is not treated as a perturbation in this scheme. Hence, the difference between the two schemes, in particular at small to moderate values of and large values of , is a measure of the uncertainty caused by the lack of harder scatterings in the implementation of LPM resummation.
The remaining genuine NLO corrections are
wider-angle “semi-collinear” processes,
contributions to longitudinal momentum diffusion arising from soft legs in processes and from soft loops in processes.
We implement the two together, following [40]. This amounts to the addition of this extra splitting rate
where
In a nutshell, this implementation subtracts the single-scattering term of Eq. (12) — the second term in is precisely in Eq. (15) — and replaces it with a form that keeps track not only of the medium-induced changes in the transverse momentum of the particles undergoing splitting, but also of the changes in the small light-cone component of the momentum, i.e. for . Indeed, as shown in [39, 40, 53], for larger emission angles these changes are no longer negligible with respect to those in transverse momentum, and give rise to the form shown here. The soft gluon carries and is no longer kinematically constrained to mediating space-like only interactions with the medium.
Finally, as anticipated in the main text, we need to avoid double countings. The collision kernel in Eq. (9) integrates over values of , , that can be of order , with as well. In this region the formulation in Eq. (9) is no longer accurate. These slices of phase space can be shown to be an contribution [40, 53], though obtained with an improper treatment for these soft modes. Thus, this contribution needs to be subtracted, as it is properly included in the NLO contribution to longitudinal momentum diffusion, incorporated in Eq. (21). This subtraction is analogous to that discussed in App. B.3 of [40]. Here we perform it by shifting the value of to . We recall that the LO value of is fixed by imposing that the expansion of Eq. (9) with the replacement Eq. (11) for and much smaller than and matches the LO Hard Loop evaluation of that limit, which is proportional to the LO longitudinal momentum diffusion coefficient [53]. To get we must now also expand for , , generating a term of relative order . We then impose that cancels this term, yielding
where the l.h.s. is what we impose, i.e. the Hard Loop form, with some UV cutoff , corresponding to the LO longitudinal momentum diffusion term, while the r.h.s. contains the terms arising from the explicit expansion of Eq. (9). Keeping only the first, leading term we recover . We solve Eq. (23) self-consistently, finding in terms of the Lambert function as
References
- [1]J. H. Traschen and R. H. Brandenberger, Phys. Rev. D 42, 2491 (1990).
- [2]L. Kofman, A. D. Linde, and A. A. Starobinsky, Phys. Rev. Lett. 73, 3195 (1994), arXiv:hep-th/9405187.DOI
- [3]L. Kofman, L. Kofman, et al., Notes Phys. 738, 55 (2008).
- [4]A. D. Linde, Rept. Prog. Phys. 42, 389 (1979).
- [5]A. Mazumdar and G. White, Rept. Prog. Phys. 82, 076901 (2019), arXiv:1811.01948 [hep-ph].DOI
- [6]A. D. Sakharov, Pisma Zh. Eksp. Teor. Fiz. 5, 32 (1967).
- [7]P. Di Bari, (2021), arXiv:2107.13750 [hep-ph].
- [8]D. G. Figueroa, A. Florio, F. Torrenti, and W. Valkenburg, JCAP 04, 035 (2021), arXiv:2006.15122 [astro-ph.CO].DOI
- [9]D. G. Figueroa, A. Florio, F. Torrenti, and W. Valkenburg, (2021), arXiv:2102.01031 [astro-ph.CO].DOI
- [10]Z. Citron et al., CERN Yellow Rep. Monogr. 7, 1159 (2019), arXiv:1812.06772 [hep-ph].arxiv.org/abs/1812.06772
- [11]D. Banerjee, M. Bögli, M. Dalmonte, E. Rico, P. Stebler, U. J. Wiese, and P. Zoller, Phys. Rev. Lett. 110, 125303 (2013), arXiv:1211.2242 [cond-mat.quant-gas].DOI
- [12]J. Berges, K. Boguslavski, S. Schlichting, and R. Venugopalan, Phys. Rev. Lett. 114, 061601 (2015), arXiv:1408.1670 [hep-ph].DOI
- [13]D. Paulsen et al., PRX Quantum 2, 030334 (2021), arXiv:2008.09252 [quant-ph].
- [14]J. Berges, M. P. Heller, A. Mazeliauskas, and R. Venugopalan, (2020), arXiv:2005.12299 [hep-th].DOI
- [15]S. Schlichting and D. Teaney, Ann. Rev. Nucl. Part. Sci. 69, 447 (2019), arXiv:1908.02113 [nucl-th].DOI
- [16]S. Grozdanov and W. van der Schee, Phys. Rev. Lett. 119, 011601 (2017), arXiv:1610.08976 [hep-th].DOI
- [17]A. Folkestad, S. Grozdanov, K. Rajagopal, and W. van der Schee, JHEP 12, 093 (2019), arXiv:1907.13134 [hep-th].DOI
- [18]A. Kurkela and E. Lu, Phys. Rev. Lett. 113, 182301 (2014), arXiv:1405.6318 [hep-ph].DOI
- [19]A. Kurkela and A. Mazeliauskas, Phys. Rev. Lett. 122, 142301 (2019), arXiv:1811.03040 [hep-ph].DOI
- [20]A. Kurkela and A. Mazeliauskas, Phys. Rev. D 99, 054018 (2019), arXiv:1811.03068 [hep-ph].DOI
- [21]X. Du and S. Schlichting, (2020), arXiv:2012.09079 [hep-ph].DOI
- [22]X. Du and S. Schlichting, (2020), arXiv:2012.09068 [hep-ph].DOI
- [23]A. Kurkela, A. Mazeliauskas, J.-F. Paquet, S. Schlichting, and D. Teaney, Phys. Rev. Lett. 122, 122302 (2019), arXiv:1805.01604 [hep-ph].DOI
- [24]A. Kurkela, A. Mazeliauskas, J.-F. Paquet, S. Schlichting, and D. Teaney, Phys. Rev. C 99, 034910 (2019), arXiv:1805.00961 [hep-ph].DOI
- [25]A. Kurkela, A. Mazeliauskas, and R. Törnkvist, (2021), arXiv:2104.08179 [hep-ph].DOI
- [26]S. Davidson and S. Sarkar, JHEP 11, 012 (2000), arXiv:hep-ph/0009078 [hep-ph].DOI
- [27]K. Harigaya and K. Mukaida, JHEP 05, 006 (2014), arXiv:1312.8097 [hep-ph].
- [28]K. Mukaida and M. Yamada, JCAP 1602, 003 (2016), arXiv:1506.07661 [hep-ph].DOI
- [29]S. Jeon, Phys. Rev. D 52, 3591 (1995), arXiv:hep-ph/9409250.DOI
- [30]S. Jeon and L. G. Yaffe, Phys. Rev. D 53, 5799 (1996), arXiv:hep-ph/9512263.DOI
- [31]P. B. Arnold, G. D. Moore, and L. G. Yaffe, JHEP 01, 030 (2003), arXiv:hep-ph/0209353 [hep-ph].DOI
- [32]E. Braaten and R. D. Pisarski, Nucl. Phys. B337, 569 (1990).
- [33]L. D. Landau and I. Pomeranchuk, Dokl. Akad. Nauk Ser. Fiz. 92, 735 (1953).
- [34]L. D. Landau and I. Pomeranchuk, Dokl. Akad. Nauk Ser. Fiz. 92, 535 (1953).
- [35]A. B. Migdal, Dokl. Akad. Nauk Ser. Fiz. 105, 77 (1955).
- [36]S. Caron-Huot and G. D. Moore, Phys.Rev.Lett. 100, 052301 (2008), arXiv:0708.4232 [hep-ph].
- [37]S. Caron-Huot, Phys.Rev. D79, 065039 (2009), arXiv:0811.1603 [hep-ph].
- [38]S. Caron-Huot, Phys.Rev. D79, 125002 (2009), arXiv:0808.0155 [hep-th].
- [39]J. Ghiglieri, J. Hong, A. Kurkela, E. Lu, G. D. Moore, and D. Teaney, JHEP 1305, 010 (2013), arXiv:1302.5970 [hep-ph].DOI
- [40]J. Ghiglieri, G. D. Moore, and D. Teaney, JHEP 03, 179 (2018), arXiv:1802.09535 [hep-ph].DOI
- [41]J. Ghiglieri, G. D. Moore, and D. Teaney, Phys. Rev. Lett. 121, 052302 (2018), arXiv:1805.02663 [hep-ph].DOI
- [42]R. Baier, A. H. Mueller, D. Schiff, and D. T. Son, Phys. Lett. B502, 51 (2001), arXiv:hep-ph/0009237 [hep-ph].DOI
- [43]A. Kurkela and G. D. Moore, JHEP 12, 044 (2011), arXiv:1107.5050 [hep-ph].DOI
- [44]J. Berges, S. Schlichting, and D. Sexty, Phys. Lett. B681, 362 (2009), arXiv:0811.4293 [hep-ph].DOI
- [45]J. Berges, S. Schlichting, and D. Sexty, Phys. Rev. D86, 074006 (2012), arXiv:1203.4646 [hep-ph].DOI
- [46]S. Schlichting, Phys. Rev. D86, 065008 (2012), arXiv:1207.1450 [hep-ph].DOI
- [47]A. Kurkela and G. D. Moore, Phys. Rev. D86, 056008 (2012), arXiv:1207.1663 [hep-ph].DOI
- [48]M. C. Abraao York, A. Kurkela, E. Lu, and G. D. Moore, Phys. Rev. D 89, 074036 (2014), arXiv:1401.3751 [hep-ph].
- [49]P. B. Arnold and C. Dogan, Phys. Rev. D 78, 065008 (2008), arXiv:0804.3359 [hep-ph].DOI
- [50]A. B. Migdal, Phys. Rev. 103, 1811 (1956).
- [51]R. Baier, Y. L. Dokshitzer, A. H. Mueller, S. Peigne, and D. Schiff, Nucl. Phys. B483, 291 (1997), arXiv:hep-ph/9607355 [hep-ph].
- [52]B. Zakharov, JETP Lett. 63, 952 (1996), arXiv:hep-ph/9607440 [hep-ph].
- [53]J. Ghiglieri, G. D. Moore, and D. Teaney, JHEP 03, 095 (2016), arXiv:1509.07773 [hep-ph].DOI
- [54]P. Aurenche, F. Gelis, and H. Zaraket, JHEP 05, 043 (2002), arXiv:hep-ph/0204146 [hep-ph].
- [55]J. Ghiglieri and D. Teaney, Int. J. Mod. Phys. E24, 1530013 (2015), arXiv:1502.03730 [hep-ph].DOI
- [56]J. Hong and D. Teaney, Phys.Rev. C82, 044908 (2010), arXiv:1003.0699 [nucl-th].DOI
- [57]M. C. Abraao York, A. Kurkela, E. Lu, and G. D. Moore, Phys. Rev. D 89, 074036 (2014), arXiv:1401.3751 [hep-ph].DOI
- [58]J.-P. Blaizot, J. Liao, and Y. Mehtar-Tani, Nucl. Phys. A 961, 37 (2017), arXiv:1609.02580 [hep-ph].DOI
- [59]M. Panero, K. Rummukainen, and A. Schäfer, Phys. Rev. Lett. 112, 162001 (2014), arXiv:1307.5850 [hep-ph].DOI
- [60]G. D. Moore and N. Schlusser, Phys. Rev. D 101, 014505 (2020), [Erratum: Phys.Rev.D 101, 059903 (2020)],
- [61]G. D. Moore and N. Schlusser, Phys. Rev. D 102, 094512 (2020), arXiv:2009.06614 [hep-lat].DOI
- [62]G. D. Moore, S. Schlichting, N. Schlusser, and I. Soudi, (2021), arXiv:2105.01679 [hep-ph].DOI
- [63]S. Mrowczynski, Arles Multipart.Dyn.1988:0499, Phys. Lett. B214, 587 (1988), [Erratum: Phys. Lett.B56,273(2007)].
- [64]S. Mrowczynski and M. H. Thoma, Phys. Rev. D62, 036011 (2000), arXiv:hep-ph/0001164 [hep-ph].DOI
- [65]A. Kurkela and G. D. Moore, JHEP 11, 120 (2011), arXiv:1108.4684 [hep-ph].DOI
- [66]S. Hauksson, S. Jeon, and C. Gale, Phys. Rev. C 103, 064904 (2021), arXiv:2012.03640 [hep-ph].
- [67]S. Hauksson, S. Jeon, and C. Gale, (2021), arXiv:2109.04575 [hep-ph].DOI
- [68]P. B. Arnold, G. D. Moore, and L. G. Yaffe, JHEP 0206, 030 (2002), arXiv:hep-ph/0204343 [hep-ph].DOI
- [69]J. Ghiglieri and G. D. Moore, JHEP 12, 029 (2014), arXiv:1410.4203 [hep-ph].DOI