Nonlinear chiral transport from holography

Nonlinear transport phenomena induced by the chiral anomaly are explored within a 4D field theory defined holographically as U(1)V × U(1)A Maxwell-Chern-Simons theory in Schwarzschild-AdS5. First, in presence of external electromagnetic fields, a general form of vector and axial currents is derived. Then, within the gradient expansion up to third order, we analytically compute all (over 50) transport coefficients. A wealth of higher order (nonlinear) transport phenomena induced by chiral anomaly are found beyond the Chiral Magnetic and Chiral Separation Effects. Some of the higher order terms are relaxation time corrections to the lowest order nonlinear effects. The charge diffusion constant and dispersion relation of the Chiral Magnetic Wave are found to receive anomaly-induced non-linear corrections due to e/m background fields. Furthermore, there emerges a new gapless mode, which we refer to as Chiral Hall Density Wave, propagating along the background Poynting vector.


Introduction
Hydrodynamics [1,2] is an effective low energy description of many interacting QFTs near thermal equilibrium. Historically, hydrodynamics has been always associated with a long wavelength limit of the underlying microscopic theory, while over the last decade or so there is an increased number of works addressing "hydronization" relaxing the long wavelength approximation. Rather, hydrodynamics is defined as an effective theory of conserved currents, such as stress tensor and/or charge currents, assuming their algebra is closed on a relevant set of near-equilibrium states.
Dynamics of the theory is governed by conservation equations (continuity equations) of the currents. The simplest example is ∂ t ρ = − ∇ · J, which is a time evolution equation for the charge density ρ sourced by three-current J. However, this equation cannot be solved as an initial value problem without additional input, the current J. In hydrodynamics, J has to be expressed in terms of thermodynamical variables, such as ρ itself, temperature, and possibly external fields if present. This is known as constitutive relation. Traditionally, in the long wavelength limit, constitutive relations are presented as a (truncated) gradient expansion. At any given order, this expansion is fixed by thermodynamic considerations and symmetries, up to a finite number of transport coefficients (TCs). The latter should be either computed from underlying microscopic theory or deduced experimentally. Diffusion constant, DC conductivity or shear viscosity are examples of the lowest order TCs.
It is well known, however, that in relativistic theory truncation of the gradient expansion at any fixed order leads to serious conceptual problems such as violation of causality.
In this paper we continue exploring hydrodynamic regime of relativistic plasma with chiral asymmetries. We closely follow previous works [24,25] focusing on massless fermion plasma with two Maxwell gauge fields, U(1) V × U(1) A . As a result of chiral anomaly, which appears in relativistic QFTs with massless fermions, global U(1) A current coupled to external electromagnetic fields is no longer conserved. The continuity equations turn into where J µ /J µ 5 are vector/axial currents and κ is an anomaly coefficient (κ = eN c /(24π 2 ) for SU(N c ) gauge theory with a massless Dirac fermion in fundamental representation and e is electric charge, which will be set to unit from now on). E and B are vector electromagnetic fields. Non-conservation of the axial current in (1.1) receives extra contribution if external axial electromagnetic fields are turned on. Throughout this work, however, we will not consider external axial fields (they were considered in ref. [24]). Chiral plasma plays a major role in a number of fundamental research areas, historically starting from primordial plasma in the early universe [26][27][28][29][30]. During the last decade, macroscopic effects induced by the chiral anomaly were found to be of relevance in relativistic heavy ion collisions [31][32][33], and have been searched intensively at LHC [34][35][36][37][38]. Finally, (pseudo-)relativistic systems in condensed matter physics, such as Dirac and Weyl semimetals, display anomaly-induced phenomena, which were recently observed experimentally [39][40][41][42][43][44][45] and can be studied via similar theoretical methods [46][47][48][49].
A hydrodynamic description of (chiral) plasma amounts to solving a set of coupled equations. As has been mentioned earlier, the continuity equations (1.1) have to be supplemented by constitutive relations describing plasma medium effects. Generically, these are of the type J = J [ρ, ρ 5 , T, E, B]; where ρ 5 is the axial charge density and T stands for the temperature. 1 In a sense, the constitutive relations (1.2) are "off-shell" relations, because they treat the charge density ρ (ρ 5 ) as independent of J ( J 5 ). Employing (1.1), the currents (1.2) are put into "on-shell". In (1.2), the fields E, B are assumed to be external. However, the charges and currents induce e/m fields of their own. Thus, the external electromagnetic JHEP01(2019)078 fields E, B have to be promoted into dynamical ones, satisfying Maxwell equations (in Gaussian units), 2 ∇ · E = 4πρ tot , where ρ tot and J tot are the total charge density and total current, a sum of external sources (ρ ext , J ext ) and induced part (ρ, J), which is the one that enters the constitutive relations (1.2). The external sources could be absent when a fully isolated system is considered. A typical example would be primordial plasma in the early Universe frequently studied using magneto-hydrodynamics (MHD). MHD, along with many other effective theories of the type, also involves neutral flow dynamics. That is, in addition to the charge current sector discussed above, one has to simultaneously consider energy-momentum conservation. Generically, the two dynamical sectors are coupled. However, in the discussion below, we will consider the probe limit, under which one ignores back-reaction of the charge sector on the energy-momentum conservation. This implies ε + p µρ + µ 5 ρ 5 with ε, p being the fluid's energy density and pressure.
A self-consistent evolution of the system is determined by solving together (1.1), (1.2), (1.3) given some initial conditions. While the equations (1.1), (1.3) are exact, the constitutive relations (1.2) are the ones where various hydrodynamic approximations are applied. A great deal of modelling normally enters (1.2), such as truncated gradient expansion, weak field approximation, etc. As a result of a full simulation, one sometimes finds instabilities leading to exponential growths of some quantities, such as of dynamical magnetic fields. It thus becomes mandatory to check if the original approximations made for the constitutive relations are consistent with the solutions found. If not, the hydrodynamical model has to be revised.
We just outlined a general setup for a hydrodynamical problem, but it is not our goal here to carry it over for any realistic system. Instead, motivated by the discussion above we would like to focus on the nature of the constitutive relations (1.2), which are well known to receive contributions induced by the chiral anomaly. The most familiar example is the chiral magnetic effect (CME) [50,51]: a vector current is generated along an external magnetic field when a chiral imbalance between left-and right-handed fermions is present ( J ∼ ρ 5 B). There is a vast literature on CME, which we cannot review here in full. The chiral magnetic conductivity was computed in perturbative QCD in [52][53][54][55][56][57]. In [58][59][60][61][62][63][64][65][66][67][68][69][70][71][72][73][74] it was evaluated for the strong coupling regime using AdS/CFT correspondence [75][76][77]. CME emerged via arguments based on the second law of thermodynamics, that is positivity of entropy production [78,79], and also within the chiral kinetic theory (CKT) [80][81][82][83][84]. Finally, numerical evidence based on lattice gauge theory for CME can be found in [85][86][87][88][89][90]. We would like to comment by passing that CME is believed to be a strict non-equilibrium 2 In principle, the axial sources (ρ5, J5), through another set of chiral anomaly-modified Maxwell's equations, would also generate classical axial e/m fields. In their turn, the axial e/m fields would enter and modify the constitutive relations (1.2), see e.g. [24].

JHEP01(2019)078
phenomenon. In other words, different arguments indicate that CME must vanish in equilibrium [32,46,91,92]. 3 Another important transport phenomenon induced by the chiral anomaly is the chiral separation effect (CSE) [93,94]: left and right charges get separated along applied external magnetic field ( J 5 ∼ B). Combined, CME and CSE lead to a new gapless excitation called chiral magnetic wave (CMW) [95]. This is a propagating wave along the magnetic field. While signature of CME/CSE has not yet been confirmed in heavy ion collision experiments [34][35][36][37], a large negative longitudinal magneto-resistance observed in Dirac/Weyl semimetals can be attributed to CME [43][44][45].
Just like in refs. [24,25], our playground will be a holographic model, namely U(1) V × U(1) A Maxwell-Chern-Simons theory in Schwarzschild-AdS 5 [59,64] to be introduced in detail in section 3. For some sort of universality, we hope to learn from this model about both generic structures of the currents and relative strengths of various effects.
Recently, transport phenomena nonlinear in external fields were realised [102] to be of critical importance in having self-consistent evolution of chiral plasma. Combined with the causality arguments mentioned earlier, the conclusion is that the constitutive relations (1.2) should contain some "nonlinear" transport coefficients so to guarantee their applicability in a broader regime. Particularly, traditional MHD is strongly affected by anomalous transports [97][98][99][100][101], which necessitates a development of a fully self-consistent chiral MHD. This triggered strong interest in nonlinear chiral transport phenomena within CKT [103][104][105][106], to which we will compare some of our findings below. Previous works on the subject of nonlinear anomalous transports include [107] based on the entropy current approach and [108] based on the fluid-gravity correspondence.
The main objective of the series of publications [24,25,96] and the present work is to explore the constitutive relations (1.2) under various approximations, primarily zooming on transport phenomena induced by the chiral anomaly. In the present publication, the following new directions are explored. First, we derive general expressions for the vector and axial currents, see (2.11), (2.12), which do not involve any approximations. This clarifies the concept of "non-renormalisation" of CME/CSE [24,25,109,110] when electromagnetic fields can be both strong and inhomogeneous in spacetime. Second,within the holographic model, we complete the calculation of all second order nonlinear transport coefficients and compare with those obtained in CKT [104]. Finally, and this is the main novel part in this publication, all third order transport coefficients are computed analytically, including relaxation time corrections to some second order transport terms (see section 4). This paves a way for the gradient resummation project released in [96]: some of the third order transport coefficients become all order frequency/momentum-dependent functions.
In the next section, we will review our results including connections to the previous works [24,25] and the forthcoming publication [96]. The remaining sections present details of the calculations.

Generalities
This subsection briefly summarises the series of works [24,25,96,111] including the present one, so to help the reader to navigate between various studies and results. We write down the most comprehensive constitutive relation and indicate specific approximations applied in each individual work.
Following [24,25], the charge densities and external fields are split into constant backgrounds and space-time dependent fluctuations whereρ,ρ 5 , E and B are the backgrounds, while δρ, δρ 5 , δ E and δ B stand for the fluctuations. Here is a formal expansion parameter to be used below. Furthermore, being unable to perform calculations for arbitrary background fields for most of the time, we introduce an expansion in the field strengths where α is a corresponding expansion parameter. Below we will introduce yet another expansion parameter λ, which will correspond to the hydrodynamical gradient expansion. For the purpose of gradient counting, e/m fields will be frequently considered as O(λ 1 ). The constitutive relations (1.2) can be formally Taylor expanded in all its arguments. This includes the gradient (λ), , and α expansions. Parametrically, a generic term entering (1.2) looks likeρ which is multiplied by a transport coefficient. 4 k, k 5 , n E , n B , m t , m x , l, l 5 , l E , l B are integers. The most general constitutive relations correspond to a sum of all possible terms like (2.3). 5 Obviously, we do not intend to consider all possible terms in (2.3). Instead, most of the results obtained in the present and early works [24,25] can be combined in a compact constitutive relation (focusing on the vector current J ), (2.4) 4 In fact, each term in (2.3) corresponds to a large number of terms obtained by different actions of the derivatives and index contractions. 5 The asymptotic nature of the gradient expansion and problems related to resummation of the series have been a hot topic over the last few years, see recent works [13,14,112]. In our approach, however, we never attempt to actually sum the series and thus these discussions are of no relevance to our formalism.

JHEP01(2019)078
The coefficients γ i are most general O(3) scalars which could be constructed from three vectors ∇, E, and B. That is, γ i are scalar functions of E 2 and B 2 , and pseudo-scalar functions of E · B. Furthermore, γ i are scalar functionals of derivative operators ∂ t , ∇ 2 , E · ∇, and pseudo-scalar functionals of B · ∇, . Some of the terms in (2.4) are well recognisable, such as diffusion (γ 1 ), electrical conductivity (γ 3 ), or CME (γ 4 ). Some other terms might be less familiar and we will discuss them below in detail.
As explained in the Introduction, the purpose of [24,25,96] and the present work is to systematically explore (2.3) under different approximations. We first briefly summarise them using both the notations of (2.3) and (2.4), and then deepen our presentation of the current study. The remaining γ i have not been probed in the study. γ i (∂ t , ∇) correspond to the gradient resummation. Thanks to the linearisation, the constitutive relations could be conveniently expressed in Fourier space. Then, the functionals of the derivatives are turned into functions of frequency and space momenta, (∂ t , ∇) → (−iω, i q). We refer to γ i (−iω, q 2 ) as transport coefficients functions (TCFs) [6]. TCFs contain information about infinitely many derivatives and associated transport coefficients. In practice, TCFs are not computed as a series resummation of order-by-order hydrodynamic expansion, and are in fact exact to all orders. TCFs go beyond the hydrodynamic low frequency/momentum limit and contain collective effects of non-hydrodynamic modes. Fourier transformed back into real space, TCFs become memory functions. Diffusion and shear viscosity memory functions were previously computed in [7,111].
• Ref. [25], study 2. Dependence of longitudinal electric conductivity on arbitrary strong constant magnetic field. A time-varying electric field is assumed to be weak.
• In the present work, we relax some of the approximations made in [24,25] and derive constitutive relations for the currents, up to third order in the gradient expansion.
In the next subsection, we summarise the main results of the present work, leaving all the technical details in the main text and appendix.

Main results
This subsection is further split into three parts. The first one contains the general form of the currents. The second part focuses on the second order nonlinear transport and comparison with similar results obtained in the CKT. The third order terms constituting the bulk of our new results are presented towards the end of this subsection.

General form of the currents
A formal expression for the constitutive relations for the vector and axial currents is derived in section 4, having the following form where ρ, ρ 5 are generic vector/axial charge densities and µ, µ 5 are corresponding chemical potentials. The external e/m fields E, B are generic as well, that is none of the approximations introduced in [24,25] is assumed. The lowest order TCs -charge diffusion constant D 0 , DC electrical conductivity σ 0 e and DC CME/CSE conductivity σ 0 χ are [24,111] where from here on we set πT = 1 for convenience. Proper powers of πT for dimensionfull quantities could be easily recovered given their physical dimensions. The σ 0 χ -terms in J and J 5 are standard CME and CSE, respectively, in agreement with "non-renormalisability" of CME [95,109,110]. The new element here, which we find important to emphasise, is that (2.11), (2.12) are exact, that is they are derived relaxing all the approximations undertaken in [24,25,109,110]. Nonlinearity of CME/CSE in external fields E and B is completely absorbed into the chemical potentials µ, µ 5 .
The corrections δ J and δ J 5 are formally defined in (A.7), which consist of higher derivative terms starting from the second order only. These terms are built from powers and derivatives of E, B, µ and µ 5 . δ J and δ J 5 are not known analytically, but could be worked out perturbatively. δ J and δ J 5 introduce new effects, particularly additional contributions to the currents along the direction of B proportional to derivatives of the chemical potentials. These effects introduce very important modifications to the original CME/CSE. As will be clear later, external e/m fields make corrections to D 0 and σ 0 e , and even generalise them into tensor-type TCs. While in principle an axial analogue of σ 0 e (i.e., a term proportional to E-term) in J 5 is also possible, it does not appear in our calculations due to the probe limit [113].
We have mentioned earlier a discussion about vanishing equilibrium CME, which might appear in tension with (2.11). In principle, since U(1) A is not a symmetry, axial gauge potential A µ itself could be regarded as another external field. Our calculations, however, are performed assuming vanishing A µ . Had we introduced a non-vanishing constant background for the time component, A t = 0, CME conductivity σ 0 χ would be shifted 14)

JHEP01(2019)078
due to a Chern-Simons contribution [24,64]. In order to have CME vanish, it suffices to impose A t =μ 5 [46,64]. In [46] it was indeed argued that the equality A t =μ 5 must be satisfied in equilibrium. While we do not have much to add to this discussion, we notice that a constant A t does not lead to any new effect beyond the shift (2.14) in CME. This is because the bulk dynamics underlying our model is expressed entirely in terms of the vector and axial field strengths.

Second order results: comparison with the CKT
At second order, the results read where the TCs take the following values The TCs in (2.15), (2.16) could be related to Taylor expansion of some γ i 's in (2.4), schematically indicated as follows To the best of our knowledge, in a holographic model, σ 0 aχH is computed here for the first time. The rest of the TCs in (2.15), (2.16) have been previously computed in [24,25]. While in the constitutive relations these terms were already introduced in [24,25], they did not contribute to dynamics (continuity equations) due to the static/homogeneous field approximations assumed in these earlier publications. For instance, to O( 0 ) considered in [24], the currents do not contain any gradient terms, resulting in dynamical instability (linear growth of ρ 5 with time). Inclusion of the external e/m field fluctuations (i.e., δ E and δ B) and associated gradient terms in (2.15), (2.16) regularizes the instability. Our results (2.15), (2.16) reveal new dynamical effects and thus are novel in a sense of a much broader range of applicability. Physical interpretation of the terms in (2.15), (2.16) and their dissipative properties are discussed below.
Our second order results could be compared with similar results obtained in CKT [103,104]. To this goal, we first put the currents on-shell eliminating ∂ t ρ and ∂ t ρ 5 JHEP01(2019)078 using the continuity equations (1.1). Second, we replace the densities by the corresponding chemical potentials. In the holographic model, the chemical potentials µ, µ 5 are computed analytically in the hydrodynamic limit. At second order in the gradient expansion, Eventually, the on-shell currents are Now lets discuss the physics of each term in (2.23), (2.24), primarily focusing on J on−shell . The first term in (2.23) is CME. The next one introduces relaxation into CME induced by time variation of the magnetic field, with τ χ being a relaxation time originally computed in [24]. τ χ was recently re-examined numerically in [114] within a quite similar holographic model but beyond probe limit. The third and fourth terms are just the classic Ohm's and diffusion currents accompanied by another relaxation effect associated with time varying electric field. The corresponding relaxation time τ e was originally computed in [115]. Note that in (2.15), (2.16) there are two additional relaxation time terms. The first one with τ D enters the diffusion current [111]. Finally, τχ is yet another relaxation time associated with generalised CME. Note the difference between τ χ and τχ: while the former is a TC responding to time varying external magnetic field, the latter is related to relaxation of the axial charge density. In (2.23), (2.24) both terms appear as O(∂ 3 ). σ 0 χ and τχ are the first two coefficients in the gradient expansion of a resummed TCF σχ [96]. Instead of a full resummation, which is a complicated numerical problem, one could use the relaxation time τχ in order to build a causal model for σχ in a spirit of MIS. In this sense, the relaxation times, such as τ χ and τχ, are of special importance.
The E × B-term in (2.23), (2.24) looks very similar to the usual Hall effect, which is, however, absent in our holographic model because of the probe limit. The term that we do find is induced by the chiral anomaly. To distinguish it from the normal Hall effect, it is referred to as chiral Hall effect [116] with σ 0 χH being its TC. Notice that σ 0 χH ∝ κ 2 . Contrary to purely anomaly-induced effects, which are normally odd in κ, the terms even in κ appear as anomaly-induced corrections to normal transports [107]. The D 0 H -term generates current perpendicular to both the magnetic field and gradients of chemical potentials. In [104] this effect was called Hall diffusion. This term can be JHEP01(2019)078 regarded as an example (we will expand on this below) in which the diffusion constant is turned into a non-trivial diffusion tensor depending on the magnetic field.
The σ 0 aχH -term in (2.23) induces flow perpendicular to both the electric field and gradient of the axial chemical potential. It was referred to as anomalous chiral Hall effect in [104]. 7 Finally, the last term in (2.23) corresponds to another anomaly-induced correction to a normal current. Normal transport due to rotor of magnetic field was first analysed in [111]. At second order in the gradient expansion under discussion now, the "normal" transport coefficient was found to be identically zero. Thus, the entire effect at this order arises from the anomaly alone.
Dissipative nature of each term entering (2.23), (2.24) is of interest. While we do not intend to dwell in this question here, we note in passing that the TCs σ 0 χ , τ e , σ 0 χH , D 0 H , σ 0 m , and τ D are all time reversal T -even and thus non-dissipative. The remaining terms in (2.23), (2.24) are dissipative.
Starting from CSE, the various terms in J on−shell 5 could be simply understood as axial analogues of those in J on−shell .
For the sake of a more detailed comparison of our results with parallel ones in CKT, we quote here the expression for the vector current as appears in [104] (2.26) Here τ is a parameter of dimension of time introduced in relaxation time approximation (RTA) of CKT. Confronting with (2.23) we notice absence of ∇µ 5 term in J on−shell . Similarly, there are no terms proportional to ∇µ, E, ∂ t E in J on−shell

5
. All these terms are expected to arise beyond the probe limit. On the other hand, the magnetic conductivity term ∇ × B is missing in (2.25). All the remaining terms appear in perfect agreement, at least as far as general structures are concerned.
Because in principle the two models describe two different regimes (strong vs weak coupling), the transport coefficients are not expected to agree. It is nevertheless instructive to pursue such a comparison. For this goal, we need to fix the parameter τ of the CKT. Obviously, there is no unique way to fix τ . We chose to set CME as a benchmark. That is, we equate the CME conductivities and the associated relaxation times in two models. This results in κ = 1 24π 2 , τ = 3 log 2.

JHEP01(2019)078
Then, the transport coefficients in (2.23) are compared to those in (2.25) While some of the coefficients came out to be of the same order, the electrical conductivity σ CKT e in CKT is strongly suppressed (by order 10 −2 ) compared to the holographic model. On the other hand, the anomaly-induced coefficients σ 0 χH and D 0 H are highly suppressed (by order 10 −3 ) in holography. 8 A complimentary way of looking at (2.11), (2.12), (2.15), (2.16) is by separately collecting terms proportional to ∇ρ and ∇ρ 5 . All these terms constitute a diffusive current, which to the lowest order in the gradients is where Much like in MHD, the diffusion constants are turned into tensors, which in fact depend non-linearly on the external e/m fields E and B. Furthermore, when higher order gradients are resummed, these diffusion tensors become momenta dependent tensor functions [96].

The third order results and collective excitations
Third order corrections in δ J and δ J 5 contain a few dozens of new terms with corresponding TCs, all of which are computed analytically (see section 4 for more details). Since the complete set of the results for δ J [3] and δ J Denoted as δ J [3] l and δ J [3] l 5 , the linear in ρ, ρ 5 terms at third order are 8 In the presented comparison, we have expressed the constitutive relations in terms of the chemical potentials, so to have them in the same form as in [104]. We could have done oppositely and that is to compare the results expressed in terms of the charge densities. To this goal we could use the relation between the charge densities and the chemical potentials of [104], which remarkably are quite similar to (2.21), (2.22). While on both sides of the comparison the TCs get modified, we have checked that the ratios (2.28) remain intact.
The underlined terms τ 13 , τ 14 include anomaly-induced B 2 -, E 2 -corrections to the charge diffusion constant D 0 : We note that both corrections are negative, see (2.34). E 2 -correction is new whereas B 2correction was first calculated in [25]. Obviously, there will be higher powers in E 2 , B 2 corrections to D 0 . In the forthcoming publication [117], we will compute the charge diffusion constant, as a function of constant e/m fields relaxing the weak field approximation. The transport coefficients τ 6 , τ 7 , τ 9 are due to spacetime inhomogeneity of ρ, ρ 5 . τ 6 , τ 9 correspond to second order expansion of the generalised CME/CSE conductivity function σχ to be computed in the forthcoming paper [96].
The terms τ 11 , τ 12 , τ 12 represent mixing effect between magnetic field and spatial gradients of ρ, ρ 5 . They were first considered in [25]. The TCs τ 15−18 , τ 20 have similar structure as the Hall diffusion and Hall effect, but the former are induced by time-varying densities and electromagnetic fields. Thus, the τ 15−18 , τ 20 -terms are relaxation times corrections to the Hall diffusion and Hall effect.
The third order gradient corrections (2.31), (2.32) contribute to various collective excitations of the holographic chiral medium, particularly they modify the dispersion relation JHEP01(2019)078 of CMW [95]. For constant electromagnetic fields, When E = 0, the dispersion relation (2.35) reduces to the one obtained in [25]. The first term (∼ q · B) in (2.35) represents nonlinear corrections to the speed of CMW, which are negative making the wave to propagate slower. The second term (∼ q · E) in (2.35) corresponds to a wave mode propagating along the electric field. It is called density wave [33] or chiral electric wave [116]. Since the chiral electric separation effect vanishes in the probe limit, here this effect is mimicked by the second term in (2.35) which is induced by the chiral anomaly as a nonlinear correction. Its presence is conditional to E not being orthogonal to B. We find the third term (∼ q · S) of special interest because it corresponds to a new phenomenon. It corresponds to a wave propagating along the direction of the energy flux S = E × B, which can be referred to as chiral Hall density wave (CHDW). The remaining terms in (2.35) are decay rates of various wave modes. The rest of this paper is structured as follows. Section 3 is about the holographic model. Section 4 is devoted to the main part of our study. Section 5 contains some closing remarks. Appendix A collects more technical details.

Holographic setup: U(1) V × U(1) A
The holographic model is Maxwell-Chern-Simons theory in the Schwarzschild-AdS 5 . The bulk action is 2) and the counter-term action S c.t. is The gauge Chern-Simons terms (∼ κ) in the bulk action mimic the chiral anomaly of the boundary field theory. Note M N P QR is the Levi-Civita symbol with the convention rtxyz = +1, while the Levi-Civita tensor is M N P QR / √ −g. The counter-term action (3.3) is specified based on minimal subtraction, which excludes finite contribution to the boundary currents from the counter-term.

JHEP01(2019)078
In the ingoing Eddington-Finkelstein coordinate, the Schwarzschild-AdS 5 is where f (r) = 1 − 1/r 4 . Thus, the Hawking temperature (identified as temperature of the boundary theory) is normalised to πT = 1. On the hypersurface Σ of constant r, the induced metric γ µν is It is convenient to split the bulk equations into dynamical and constraint components, dynamical equations : constraint equations : The boundary currents are defined as which, in terms of the bulk fields, are where n M is the outpointing unit normal vector with respect to the slice Σ, and ∇ is compatible with the induced metric γ µν . The radial gauge V r = A r = 0 will be assumed throughout this work. As a result, in order to determine the boundary currents (3.11) it is sufficient to solve dynamical equations (3.6) only, leaving the constraints aside. Indeed, the constraint equations (3.7) give rise to continuity equations (1.1) In this way, the currents' constitutive relations to be derived below are off-shell. Practically, it is more instructive to relate the currents (3.11) to the coefficients of near boundary asymptotic expansion of the bulk gauge fields. Near r = ∞, (3.14) A possible constant term for A µ in (3.13) has been set to zero, in accordance with the fact that no axial external fields is assumed to be present in the current study. V µ is the gauge potential of external electromagnetic fields E and B, Dynamical equations (3.6) are sufficient to derive (3.13), (3.14), where the near-boundary data V (2) µ and A (2) µ have to be determined by completely solving (3.6) from the horizon to the boundary. The currents (3.11) become As the remainder of this section, we outline the strategy for deriving the constitutive relations for J µ and J µ 5 . To this end, we turn on finite vector/axial charge densities for the dual field theory, which are also exposed to external electromagnetic fields. Holographically, the charge densities and external fields are encoded in asymptotic behaviors of the bulk gauge fields. In the bulk, we will solve the dynamical equations (3.6) assuming the charge densities and external fields as given, but without specifying them explicitly.
Following [111] we start with the most general static and homogeneous profiles for the bulk gauge fields satisfying the dynamical equations (3.6), where V µ , ρ, ρ 5 are all constants for the moment. Regularity at r = 1 has been used to fix one integration constant for each V i and A i . As explained below (3.14), the constant term in A µ is set to zero. Through (3.16), the boundary currents are Hence, ρ and ρ 5 are identified as the vector/axial charge densities. Next, following the idea of fluid/gravity correspondence [118], we promote V µ , ρ, ρ 5 into arbitrary functions of the boundary coordinates As a result, (3.17) ceases to solve the dynamical equations (3.6). To have them satisfied, suitable corrections in V µ and A µ have to be introduced: where V µ , A µ will be determined by solving (3.6). Appropriate boundary conditions are classified into three types. First, V µ and A µ are regular over the domain r ∈ [1, ∞). Second, at the conformal boundary r = ∞, we require

JHEP01(2019)078
which amounts to fixing external gauge potentials to be V µ and zero (for the axial fields). Additional integration constants will be fixed by the Landau frame convention for the currents, The Landau frame convention corresponds to a residual gauge fixing for the bulk fields. The vector/axial chemical potentials are defined as (3.23) Generically, µ, µ 5 are nonlinear functionals of densities and external fields.
In terms of V µ and A µ , the dynamical equations (3.6) are (3.27) In the next section we will present solutions to (3.24)-(3.27) under approximation discussed in the Introduction.

Nonlinear chiral transport
In this section, we initially explore generic structure of the vector and axial currents (1.2) as emerges within the holographic model of section 3. No assumptions will be made regarding the charge densities ρ, ρ 5 and external fields E, B. While we are not able to solve the dynamical equations (3.24)-(3.27) analytically, we can advance by rewriting them in integral JHEP01(2019)078 forms and extract near-boundary asymptotic expansion for the corrections V µ and A µ . The procedure is rather tedious. Hence all the details are moved to appendix A. Via (3.16), the near-boundary asymptotic behaviors (A.1)-(A.4) yield the results (2.11), (2.12) with δ J and δ J 5 formally given by (A.7). As is clear from (A.5), (A.6), δ J and δ J 5 are composed of higher derivative terms involving E, B and ρ, ρ 5 . Now we continue with the gradient expansion of δ J and δ J 5 . Within the hydrodynamic limit, the dynamical equations (3.24)-(3.27) are solved perturbatively. Let us introduce a formal expansion parameter λ by ∂ µ → λ∂ µ , which counts order of the gradient expansion. Then, V µ and A µ could be expanded in powers of λ, We remind the reader that for this study, E and B are considered to be of O(λ 1 ). At each order in λ, V [n] µ and A [n] µ obey a system of ODEs, which could be analytically solved via direct integration over r. We list the results for V With the second order corrections V [2] µ and A [2] µ (A.13)-(A. 16), we obtain the third order results δ J [3] and δ J [3] 5 . However, nonlinearity makes such calculations rather involved and the number of various terms is very large. For the sake of presentation, we have split the third order corrections into terms that are linear in either ρ or ρ 5 , and the rest.
The linear in the charge densities parts of δ J [3] and δ J [3] 5 , denoted as δ J [3] l and δ J The explicit expressions for a and b are
The τ 24 -term stands for second order expansion of the CME conductivity σ χ of [24] and was first computed there. τ 25 is the relaxation term for the second order Hall diffusion current (see the D 0 H -term in (2.23), (2.24)). The τ 26 , τ 27 , τ 29 -terms rely on the time inhomogeneity of charge densities or magnetic field and could be thought of as extension of the Hall diffusion current. The TC τ 31 is related to the e/m energy flux and also generalises the Hall diffusion current. τ 32 is composed of spatial gradient of charge densities and corresponds to nonlinear charge diffusion process. The τ 36 , τ 39 -terms are e/m field corrections to the nonlinear charge diffusions. The last TC τ 40 is a nonlinear in E, B correction to the charge diffusions. The terms τ 33 , τ 34 are nonlinear in densities corrections to τ 21 , τ 22 .

JHEP01(2019)078
τ 35 is the third order extension of the anomalous chiral Hall effect, i.e., σ 0 aχH -term in (2.23), (2.24). In [96], we will perform a systematic resummation for certain transports and the term τ 35 will be generalised into a TCF. Similarly, τ 38 can be simply taken as the magnetic analogue of τ 35 and will be extended to a TCF in [96].
Finally, let us mention the dissipative nature for each term in the third order results (4.6), (4.7). Via the criterion of T -symmetry, the TCs τ 24 , τ 30−34 , τ 37−40 are T -even and are thus non-dissipative. The remaining terms are all dissipative.

Conclusion
In this work, we have continued exploration of nonlinear chiral anomaly-induced transport phenomena based on a holographic model with two U(1) fields interacting via a Chern-Simons term. For a finite temperature system, we constructed off-shell constitutive relations for the vector/axial currents. A detailed report on our new results could be found in the Summary section. Here they are in brief: • We demonstrated that both CME and CSE get corrected by higher derivative terms, see (2.11), (2.12). In the hydrodynamic limit, we analytically calculated those gradient corrections up to third order. Comparison with the CKT was presented. New third order results, particularly (2.31), (2.32), extend those that were initially considered in [24,25] and reveal novel effects associated with time-dependence or inhomogeneities of the charge densities and external fields.
• Among new results worth highlighting, in weak field approximation the charge diffusion constant D 0 was found to receive negative anomaly-induced E 2 -and B 2corrections (2.34). It is very interesting to explore the dependence of D 0 on the e/m fields beyond the weak field approximation, that is non-perturbatively. Of particular interest would be a strong field limit. We are pursuing this line of study in the forthcoming paper [117] (see also [95] for similar study but in a different holographic model 9 ).
• Another result we found to be of interest is that the chiral medium is shown to support three types of collective modes: CMW (propagating along B), CEW propagating along E, and a new one, chiral Hall density wave, propagating orthogonal to the other two, that is, along the energy flux E × B.
The follow up paper [96] focuses on another set of approximations. Instead of considering a fixed order gradient expansion adopted here, we compute some TCFs in nonlinear chiral transport phenomena. More specifically, the external electromagnetic fields are assumed to be constant and weak, while the charge densities are split into constant backgrounds and small inhomogeneous fluctuations. The setup is similar to that of [24], but in [96] as opposed to [24], gradient resummation is performed for terms that are linear both in the charge density fluctuations and external fields.

JHEP01(2019)078
We have found a wealth of non-linear phenomena all induced entirely by the chiral anomaly. An important next step in deriving a full chiral MHD would be to abandon the probe limit adopted in this paper and include the dynamics of a neutral flow as well. This will bring into the picture additional effects such as thermoelectric conductivities, normal Hall current, the chiral vortical effect [119,120], and some nonlinear effects discussed in [103]. We plan to address these in the future.

A Supplement for section 4
In this appendix, we collect all calculational details omitted in section 4. Regarding the general structure of the constitutive relations of the vector/axial currents, we present the integral versions of the bulk dynamical equations and explore near boundary asymptotics. We further derive the gradient expansion, and compute analytically all TCs, up to third order.