EFI-18-18

Effective field theory of magnetohydrodynamics from generalized global symmetries

Paolo Glorioso and Dam Thanh Son
Kadanoff Center for Theoretical Physics, University of Chicago, Chicago, IL 60637, USA

We introduce an effective action for non-dissipative magnetohydrodynamics. A crucial guiding

principle is the generalized global symmetry of electrodynamics, which naturally leads to introducing
a “dual photon” as the degree of freedom responsible for the electromagnetic component of the
fluid. The formalism includes additional degrees of freedom and symmetries which characterize the
hydrodynamic regime. By suitably enhancing one of the symmetries, the theory becomes force-free
electrodynamics. The symmetries furthermore allow to systematize local and non-local conserved
helicities. We also discuss higher-derivative corrections.
arXiv:1811.04879v1 [hep-th] 12 Nov 2018

I. INTRODUCTION dictated by the conservation equation of the dual field-

Magnetohydrodynamics (MHD) [1] combines Navier-
Stokes hydrodynamics and Maxwell electrodynamics into
a theory that can describes the long-time, long-distance
behavior of charged fluids. It is applicable in systems
of sizes ranging from laboratory systems to the whole
Universe. In the standard formulation, the MHD equa-
tions describe the time evolution of the hydrodynamic
variables (density, fluid velocity, temperature) and the
magnetic field.
Recently, MHD has been recast into an alternative
form using a generalized global symmetry [2] associated
with the conservation of the dual of the Maxwell field
strength:1
1 µναβ
∂µ F̃ µν = 0, F̃ µν = ε Fαβ .
2
In this formulation, MHD is a hydrodynamic theory with
a conserve charge, but the conserved current in this case
is a 2-form, rather than a 1-form [3]. Equation (1.1) is
the statement that the magnetic flux, defined as the in-
tegral of the magnetic field over a codimension-2 surface,
is conserved. This approach has the advantage that all
equations of MHD are treated in the same footings, i.e.,
as conservation equations.
In this paper, we provide a formulation of MHD in
terms of the action principle. An action formulation has
an advantage of potentially giving a clear interpretation
of conserved quantities as the consequence of symmetries.
In particular, in our formulation the conservation of the
2-form current is ensured by a an abelian global symme-
try, in which the parameter of a symmetry transforma-
tion is a 1-form, rather than a scalar. This symmetry
turns out to be the main guiding principle to write the
most general equations compatible with the standard as-
sumptions of MHD, i.e., quasi-neutrality of the plasma
and Debye screening of the electric field. This new sym-
metry suggests to include a “dual photon” to describe the
dynamics of the electromagnetic part of the fluid. This
is natural as the dynamics of the electromagnetic field is
strength (1.1). The other degrees of freedom are (the
relativistic) Lagrangian coordinates of the fluid.
Previous action formulations of MHD have been imple-
mented mainly by generalizing the approach of Clebsch
potentials [4–6]. Other descriptions used Lagrangian co-
ordinates [7, 8], although there are no dynamical degrees
of freedom associated to the electromagnetic field. The
advantage of introducing the dual photon is the pres-
ence of an enhanced set of symmetries, which allows one
to systematically find various conserved helicities, pro-
viding a more transparent geometric structure. Another
advantage of our formulation is that it conformes nicely
with the philosophy of effective field theory, in which one
writes down systematically all possible terms of the La-
grangian compatible with the required symmetries, fol-
(1.1) lowing a power counting given by derivative expansion.
The symmetries guarantee that each of these terms is
consistent with the basic physical assumptions of MHD.
It is worth mentioning that, since this approach is based
on a standard action principle, it will not capture dissipa-
tive contributions, such as conductivity and viscosities.2
The paper is organized as follows. We begin Sec. II
by introducing the degrees of freedom and the symme-
tries, and then present the general action of ideal MHD
and show that the stress-energy tensor and the dual field-
strength, as well as as possible additional conserved cur-
rents, acquire the form expected from MHD. In Sec. III
we show how a variational formulation of force-free elec-
trodynamics (FFE) can be obtained by enhancing one
of the symmetries. We also show that, intriguingly, the
equations of FFE can be recovered as a limit of axion elec-
trodynamics. In Sec. IV we describe the Noether currents
associated with the symmetries, and show that particu-
lar subgroups of such symmetries give rise to conserved
helicities. In Sec. V we discuss higher-derivative correc-
tions. We state our conclusions in Sec. VI.

2 In order to capture dissipation, one needs an enhanced varia-

tional principle where degrees of freedom are doubled [9] (see
1 We choose the convention ε0123 = √1 . also Refs. [10] and [11]).
−g
 2

II. ACTION PRINCIPLE In analogy with the effective theory formulation of hy-
drodynamics [9–16], we propose that the degrees of free-
We first describe the building blocks for our action dom responsible for the conservation equations (2.1) are
formulation: the degrees of freedom and the symmetries. the parameters of such transformations, or more pre-
We then use these ingredients to write down the most cisely, undergo shifts under these transformations. To
general action at leading order in derivatives, which will parameterize diffeomorphisms, we introduce the fields
correspond to ideal MHD, and relate it to thermodynam- σ a (x), which we regard as mappings from the physical
ics. spacetime, with coordinates xµ , to an internal spacetime,
with coordinates σ a . To parameterize 1-form U (1) trans-
formations, we introduce the 1-form ϕa (x).3 We shall see
A. Kinematics later that ϕa can be interpreted as a “dual photon.” We
emphasize that σ a , ϕa are not Goldstone modes, as they
In the 2-form formulation of Ref. [2], MHD is stated in are not associated with the spontaneous symmetry break-
terms of the conservation equations ing of a symmetry defined in the microscopic system.4
To ensure that Eqs. (2.1) follow from the variation of
1 ναβ the action with respect to σ a and ϕa , we demand the
∇µ T µν = H Jαβ , ∇µ J µν = 0 , (2.1)
2 action to depend on the background sources gµν , bµν and
the degrees of freedom σ a , ϕa through the following com-
where T µν is the stress-energy tensor and J µν is a 2- binations:
form current which can be identified with the dual field
strength F̃ µν defined in (1.1). In the above we also in- hab = K µa K νb gµν
cluded Bab = K µa K νb bµν + 2K µ[a ∂µ ϕb] (2.7)
Hµνρ = ∂µ bνρ + ∂ν bρµ + ∂ρ bµν , (2.2) = K µa K νb bµν + 2∂[a ϕb]
µ
∂x 5
where bµν is the background source coupled to J µν . The where K µa = ∂σa . Requiring
field strength Hµνρ can be thought of as the dual of an
S = S[hab , Bab ] (2.8)
external current which probes the electromagnetic field,
i.e. ensures that the equations of motion for σ a and ϕa imply
1 precisely Eqs. (2.1). The combinations (2.7) also guar-
σ
Jext = − εσαβγ Hαβγ , (2.3) antee that the action is invariant under diffeomorphisms
6
and 1-form U (1) transformations, given by Eqs. (2.6) to-
so that gether with
1 ναβ σ a (x) → σ a (y(x))
H Jαβ = Jµ,ext F νµ , (2.4) (2.9)
2 ϕa (x) → ϕa (y(x)) + K µa χµ (x) .
i.e., the right-hand side of the first equation in (2.1) is
µ Now that we ensured conservation of stress-energy and
the Lorentz force generated by Jext .
flux, we come to the second fundamental ingredient of
To find an action S whose equations of motion are the
MHD: The explicit expressions of T µν and J µν should
conservation equations (2.1), the first step is to require
depend only on the fluid velocity uµ , the temperature
one can couple the theory to a background metric gµν
T , and the chemical potential for the 1-form charge µhµ .
and 2-form bµν , and that the stress-energy tensor and
Following Ref. [2] we have factorized the 1-form chemical
the 2-form current are obtained by varying the action
potential into a scalar µ and a 1-form hµ such that
with respect to the respective sources, i.e.,
uµ hµ = 0, hµ hµ = 1 . (2.10)
µν 2 δS
T =√
−g δgµν The most general action S of the form (2.8) does not
(2.5) satisfy the above requirement, and the conserved cur-
2 δS
J µν =√ . rents, in general, have a rather generic dependence on
−g δbµν
The conservation of these currents is then a consequence
of the invariance of the action with respect to diffeomor- 3 Note that we chose to take the indices of ϕa to live in the σa
phism and 1-form U (1): spacetime: as we will see later, this choice makes more transpar-
ent the tensor structure of various equations. Defining the 1-form
gµν (x) → ∂µ y α ∂ν y β gαβ (y(x)) ϕµ in physical spacetime leads to an equivalent set of equations
(2.6) of motion after the identification ϕµ = ∂µ σa ϕa .
bµν (x) → ∂µ y α ∂ν y β bαβ (y(x)) − 2∂[µ χν] , 4 See [17] for an alternative perspective of this point.
5 One can arbitrarily choose either xµ or σa to be the dynam-
where y µ (x) and χµ (x) denote coordinate and one-form ical degrees of freedom of the theory responsible for energy-
U (1) transformations, respectively. momentum conservation.

σ a and ϕa . We now show that the requirement can be
satisfied by imposing additional symmetries, which, in a
certain sense, will truly define the geometric nature of our
MHD theory. In order to introduce these symmetries, we
need to develop intuition on the the physical meaning of
the degrees of freedom σ a and ϕa and their relationship
with the hydrodynamic variables uµ , T , and µhµ .
The fields σ a are the relativistic Lagrangian fluid coor-
dinates. Heuristically, a fixed value of the spatial coordi-
nates σ i identifies a given fluid element, whose trajectory
in spacetime is described by the function xµ (σ 0 , σ i ) as σ 0
varies, and σ 0 represents the internal “clock” of the fluid
element. It is then natural to define the velocity of the
fluid as the normalized tangent vector to such trajectory:
1 ∂xµ p that the action should be invariant under (2.16). In other
uµ = , b = −h00 . (2.11)
b ∂σ 0
Consider now a system in a homogeneous configura-
tion at temperature T0 . Placing the system in a slowly-
varying curved background induces a redshift in √ the lo-
cal temperature, so that the latter is T (x) = T0 / −g00 .
This invites to take the following relation
√ between the
local temperature and σ a : T (x) = T0 / −h00 . Further-
more, a global rescaling of time x0 → γx0 induces a
rescaling of the asymptotic temperature T0 → T0 /γ. We
can use this to fix T0 = 1 and write
1
T (x) = . (2.12)
b
Finally, ϕa can be thought of as a “1-form phase”
associated to each fluid element, which we want to re-
late to the 1-form chemical potential. The correspond-
ing charges are the magnetic flux across 2-dimensional
surfaces at constant time. In the σ a -coordinate system,
where σ 0 singles out the time direction, the charges are
√
Z
Qi = d2 x −gJ 0i , (2.13)
Σ(i)
where Σ(i) is the plane at constant values of σ 0 and σ i ,
for i = 1, 2, 3. The thermodynamic partition function is
then
1
Z = Tr(e− T (H−µi Q ) ) ,
i
(2.14)
where H is the vacuum Hamiltonian of the system, and
µi is the chemical potential conjugated to Qi . Compar-
ing Eqs. (2.13), (2.14), and (2.5), we are then led to
identify B0i with the 1-form chemical potential µi in the
σ a -coordinates. Pulling back to xµ coordinates, this be-
comes
1 ∂σ i
µhµ = B0i , (2.15)
b ∂xµ
where the prefactor of 1b comes from that B0i behaves like
a scalar density in the time direction. We will develop
more details on this object in the next section.
3
We now come to the symmetries. As discussed above,
the spatial coordinates σ i can be thought of as giving a
system of coordinates embedded in the fluid. However,
there is a redundancy in this identification. Given a fluid
configuration, one can arbitrarily change the system of
coordinates used to parametrize fluid elements without
changing physical quantities, i.e. temperature, velocity
and one-form U (1) chemical potential. This is in contrast
to, e.g., solids, where there is a single-out coordinate sys-
tems tied to the lattice. The redundancy characteristic
of fluids is captured by the spatial diffeomorphisms
σ i → f i (σ j ) , (2.16)
where f i is an arbitrary function of σ j . We then require
words, one has the freedom of relabeling fluid elements
at a constant time slice σ 0 = const. Analogously, the
choice of initial value of the internal time σ 0 should be
arbitrary for each fluid element. This leads to requiring
invariance under time shifts:
σ 0 → σ 0 + f (σ i ) . (2.17)
where f is an arbitrary function of σ i . Lastly, one should
also have the freedom of arbitrary choice of the one-form
ϕa at the initial time. This leads to demanding invariance
with respect to time-independent shifts of the one-form
phase
ϕa → ϕa + λa (σ i ) , (2.18)
where ψa is a time-independent one-form with arbitrary
dependence on σ i .
Finally, given that the action depends on ϕa through
the second line of Eq. (2.7), one has the additional sym-
metry
ϕa → ϕa + ∂a α(σ) , (2.19)
where α is an arbitrary scalar function of σ a . This sym-
metry is time-dependent, contrary to the previous ones.
We shall see later that it will lead to a generalization of
the constraint on the magnetic field ∂i B i = 0.
Note that T, uµ , µ and hα , as defined in Eqs. (2.11),
(2.12), and (2.15) are invariant under (2.16)–(2.18). In
fact, as we will show, the stress-energy tensor and the
2-form current coming from an action invariant under
(2.16)–(2.18), depend only on T , uµ , and µα . This will
be a crucial property to verify the consistency of our for-
mulation.
To make the symmetries (2.16)–(2.18) manifest, it is
convenient to parametrize the fields (2.7) in terms of new
variables b, vi , aij , mi , and mij
hab dσ a dσ b = − b2 (dσ 0 − vi dσ i )2 + aij dσ i dσ j
Bab dσ a ∧ dσ b =2bmi (dσ 0 − vj dσ j ) ∧ dσ i (2.20)
i j
+ mij dσ ∧ dσ .
 4

The objects in (2.20) transform covariantly under (2.16): If we identify

b → b, (2.21) s = ∂T F , (2.31)
j j
mi → ∂i f mj , vi → ∂i f vj (2.22)
where s is the entropy density, we find the thermody-
aij → ∂i f k ∂j f l akl , mij → ∂i f k ∂j f l mkl (2.23) namic relation
while under time shifts (2.18), only vi transforms non-
trivially p + ε = sT + µρ , (2.32)

vi → vi − ∂i f, (2.24) which was first obtained in [2].

and only mij transforms nontrivially under (2.19):
mij → mij + 2∂[i λj] . (2.25) C. Connection to the traditional formulation of
MHD
From the discussion around (2.15) we also see that mi
is the one-form chemical potential in σ a -coordinates. In In the usual formulation of relativistic ideal MHD [18],
terms of mi , eq. (2.15) reads the stress tensor is the sum of fluid and Maxwell stress
1 ∂σ i
q tensors:
hµ = m i , µ = mi mj aij . (2.26)
µ ∂xµ µν µν
T µν = Tfluid + Tmaxwell , (2.33)
Note that the second equation in (2.10) is automatically
guaranteed, indeed where
µν
1 ∂xµ ∂σ i µ Tfluid = p0 (T )g µν + (ε0 (T ) + p0 (T ))uµ uν
u µ hµ = µ mi = δ0i mi = 0 . (2.27)
b ∂σ 0 ∂xµ b µν 1 (2.34)
Tmaxwell = F µα F να − F 2 g µν ,
4
B. Ideal MHD and where one imposes the vanishing of the electric field
in the fluid rest frame F µν uν = 0, which allows to rewrite
In this Section we shall study the most general ac-
tion at leading derivative order which is compatible with µν 1
Tmaxwell = B 2 uµ uν + B 2 g µν − B µ B ν , (2.35)
the symmetry requirements discussed above. The leading 2
derivative order contains no derivatives at all, meaning
that we construct the action based on the fields in the where B µ = 21 εµνρσ uν Fρσ is the magnetic field. In
µν
decomposition (2.20). The most general action at this or- this case Tfluid is completely disentangled from the elec-
der turns out to depend only on b and mi . Using (2.12) tromagnetic part, as the thermodynamic parameters
and (2.26), it is given by p0 (T ), ε0 (T ) are only functions of the temperature. Ad-
ditionally,
√
Z
S = d4 x −gF (T, µ) , (2.28)
F̃ µν = 2u[µ B ν] , (2.36)
where F is an arbitrary function of T and µ. Varying and the complete set of equations of motion is given by
(2.28) with respect to the background sources as in (2.5)
we obtain stress-energy tensor and two-form current:6 ∇µ T µν = 0, ∇µ F̃ µν = 0 . (2.37)
2 δS
T µν = √ = pg µν + (ε + p)uµ uν − µρhµ hν We then see that the total stress tensor (2.33) and the
−g δgµν
dual field strength (2.36) coincide with T µν and J µν given
2 δS in (2.29),(2.30), respectively, upon identifying
J µν = √ = 2ρu[µ hν] ,
−g δbµν
(2.29) 1
p = p0 + B 2 . (2.38)
2
where p and ε are pressure and energy density, and ρ
denotes a scalar associated to the 2-form density. They The traditional formulation of MHD can indeed be seen
are related to F through as a particular limit of (2.29), where the dynamics of the
electromagnetic field is dominated by the Maxwell La-
p = F, ρ = ∂µ F, ε = T ∂T F + µ∂µ F − F . (2.30)
grangian. For sufficiently large electromagnetic fields one
expects thermal and quantum fluctuations to become im-
portant, thus generating a nontrivial dependence on B 2
6 See Appendix A for the variations of various fields with respect in the pressure, which is captured by the general expres-
to background sources. sions in (2.29).

D. MHD with number conservation
In many situations one may need to include an addi-
tional conserved charge in MHD. The conservation equa-
tion for the corresponding vector current is then
∇µ J µ = 0 . (2.39)
We emphasize that J µ is not the current associated to
the electromagnetic U (1). Depending on the context,
J µ could be the baryon number current, or simply the
particle number current, e.g. when dealing with a non-
relativistic system, where the particle number is con-
served. Such J µ can also be the axial current, which
is relevant when the system is coupled to chiral matter.7
We couple the theory to a background gauge field Cµ so
that J µ is obtained by the variation
1 δS
Jµ = √ . (2.40)
−g δCµ
Note that in the presence of J µ , the first eq. in (2.1) is
modified to
1 ναβ
∇µ T µν = H Jαβ + Dνα Jα , (2.41)
2
where Dµν = ∂µ Cν − ∂ν Cµ , and the conservation of the
2-form is unaffected.
The conservation equation (2.39) is associated to in-
variance of S under the background gauge transforma-
tion
Cµ → Cµ + ∂µ λ , (2.42)
where λ(x) is a function. In parallel with sec. II A, we
then introduce a scalar field ϕ(x) which parameterizes
such transformations. To ensure that eq. (2.39) follows
from the varying the action with respect to ϕ, we re-
quire that the action depends on Cµ and ϕ through the
combination
Ba = Kaµ (Cµ + ∂µ ϕ) . (2.43)
This combination ensures that the action will be invari-
ant under the symmetries discussed before, as well as
(2.42), provided we simultaneously transform
ϕ→ϕ−λ . (2.44)
Finally, we demand the action to be invariant under the
transformation
ϕ → ϕ + χ(σ i ) (2.45)
(keeping Cµ fixed), where χ(σ i ) is an arbitrary function
of the spatial fluid coordinates. This corresponds to re-
quiring the freedom of relabeling the U (1) phase of each
7 See Ref. [19] for a discussion on chiral MHD and the dramatic
consequences due to quantum anomalies.
5
fluid element at a given time slice. The same symmetry
was required in the formulation of charged fluids in [13],
where the transformation (2.45) was dubbed “chemical
shift”.
It will be convenient to decompose Ba as
Ba dσ a = bµN (dσ 0 − vi dσ i ) + bi dσ i , (2.46)
where we shall interpret µN as the chemical potential
associated to the U (1) charge. This can be motivated
by a similar argument as that around eq. (2.15). The
variables µN and bi transform covariantly under (2.16)-
(2.19), while under (2.45),
µN → µN , bi → bi + ∂i χ . (2.47)
At ideal level, the most general action is then
Z
√
S = d4 x −g F (T, µ, µN ) , (2.48)
where F is an arbitrary function. Varying S with respect
to Cµ gives the current
J µ = nuµ , (2.49)
where n = ∂µN F is the number density. The other con-
stitutive relations (2.29) are unmodified, as well as (2.30),
except that now
ε = T ∂T F + µ∂µ F + µN ∂µN F − F , (2.50)
leading to the thermodynamic relation
p + ε = sT + µρ + µN n . (2.51)
III. FORCE-FREE ELECTRODYNAMICS
An important limit of MHD if force-free electrodynam-
ics. In this limit the contribution of electromagnetic fields
to the dynamics is much more relevant than the contribu-
tion of matter, the latter essentially decouples from the
equations but the nonlinear structure of MHD remains.
This is a very useful approximation in astrophysics to
study e.g. the solar corona or other types of astrophysical
plasma, such as black hole and neutron star atmospheres
[20, 21].
In FFE, the stress-energy tensor is approximated with
that coming from Maxwell Lagrangian − 41 F µν Fµν , which
implies that
F µν Jµ = 0 , (3.1)
µ
where J is the electromagnetic current produced by the
plasma. Using Maxwell’s equations, we can then write
FFE in the following form
∇µ F̃ µν = 0, ∇µ F µν Fνρ = 0 , (3.2)
where we are left with are left with a non-linear system
which determines the evolution of Fµν .
Below we shall see that an action for FFE can be
obtained by enhancing the symmetry requirements dis-
cussed in the previous Section. We will also show that,
intriguingly, eqs. (3.2) can be obtained from a limit of
axion electrodynamics.
 6

A. FFE as symmetry-enhanced MHD Now, using the identity

Vµ εαβγδ Fαβ Fγδ = −4εαβγδ Fµα Vβ Fγδ , (3.10)
We now require, instead of (2.17), that the action be
invariant under the larger group of transformations one finds

σ 0 → f (σ 0 , σ i ) , (3.3) Fρν J ν = κεµναβ ∂ν θFρµ Fαβ

1 (3.11)
where f is an arbitrary function. Clearly, this transfor- = − κ∂ρ θεµναβ Fµν Fαβ = 0 .
4
mation implies that σ 0 decouples from the action. Under
(3.3), the temperature (2.12) transforms as We thus recovered precisely the equations of force-free
electrodynamics! The current in Eq. (3.9) may appear

∂f
−1 to possess a more specific form than that in FFE, but it
T →T , (3.4) turns out that this not the case. Indeed, from Eq. (3.11)
∂σ 0
we know that Fµν , being antisymmetric, has rank 2, and
and thus, the most general action at leading derivative thus one can write
order is Fµν = V[µ Wν] , J µ Vµ = J µ Wµ = 0 , (3.12)
√
Z
S = d4 x −g F (µ) . (3.5) where Vµ , Wµ are two vector fields. Hence, there exist a
vector field Sµ such that
The stress-energy tensor and two-form current obtained J µ = εµαβγ Vα Wβ Sγ = εµαβγ Fαβ Sγ , (3.13)
from this action are the same as those in Eqs. (2.29) and
(2.30), except that now ∂T F = 0. The interesting obser- and, due to the Bianchi identity,
vation is that such system can be shown to be equivalent
to a generalization FFE. ∇µ J µ = εµαβγ Fαβ ∇µ Sγ = 0 , (3.14)
Identifying J µν = 21 εµναβ Fαη , the Maxwell energy- which in turn implies ∇[µ Sν] = 0. We then must have
momentum tensor can be written as Sµ = ∂µ α, for some function α. This then shows that
1 Eq. (3.9) describes the most general current in FFE, and
T µν = − J 2 g µν + J µα J να , (3.6) we conclude that the equations from the axion electro-
4
dynamics action (3.7) are completely equivalent to FFE.
which has the form of (2.29) upon using (2.30), with The action (3.7) should be considered as a formal
F (µ) = 12 µ2 . The action (3.5) can then be viewed as rewriting of FFE, and θ as an unphysical Legendre multi-
the generalization of FFE to non-linear electrodynamics. plier. One may ask, however, if there is a regime in which
We also note that the structure of the above constitutive axion electrodynamics, with an axion field of interest for
relations has similarities with the zero-temperature limit particle physics, reduces to FFE. We now show that the
of [2]. neglection of the kinetic term and the mass term in the
Lagrangian (3.7) can be justified only at very large values
of the magnetic field. To see that, we add these terms
B. FFE as a limit of axion electrodynamics into the Lagrangian

In this section we describe a completely different action fa2

− [(∂µ θ)2 + m2 θ2 ] (3.15)
principle for FFE, which can be thought of as a limit of 2
axion electrodynamics. This action is where fa and ma are the decay constant and mass of the
axion, respectively, and requires that the effect of these
1 √
Z
d4 x −g F µν Fµν + κθεµναβ Fµν Fαβ , new terms on the equation of motion is negligible. Con-


S=−
4 sider a mostly magnetic solution where E ≪ B. From
(3.7) an equation of motion
where κ is a constant, and we regard the axion θ to be
dynamical. Note that both the kinetic term and the po- ∂i E i = κB i ∂i θ (3.16)
tential term for θ are absent, so θ is now a Lagrange
multiplier. it follows that θ ∼ E/(κB). The new terms in the action
Varying Aµ and θ, one finds the equations of motion can then be neglected when
θ
εµναβ Fµν Fαβ = 0 fa2 , fa2 m2a θ ≪ κEB (3.17)
(3.8) L2
∇µ F µν = J ν ,
where L is the length scale over which the fields vary.
µ
where the current J is defined by This condition can be rewritten as
f a ma f a
J µ ≡ κεµναβ ∂ν θFαβ . (3.9) B≫ , (3.18)
κ κL
 7

For QCD axions fa maa ∼ Λ2QCD and κ ∼ α, and one under some of the symmetries (2.16)-(2.19). The issue of
needs magnetic fields stronger than 1020 G. We leave finding conserved helicities and understanding if they can
open the possibility that FFE can be realized in some be written locally in Eulerian variables is a topic of inter-
regime of the axion electrodynamics relevant for cer- est in MHD [4, 5, 25]; helicities are useful to characterize
tain time-reversal breaking topological insulators, where solutions and have wide applications to numerical simu-
θ plays the role of a dynamical magnetization [22–24]. lations. The discussion of this subsection aims at giving
a systematic derivation of local and non-local helicities
using the relation between locality of a given conserved
IV. NOETHER CHARGES quantity and invariance of the latter with respect to the
symmetries (2.16)–(2.18) and (2.45). Below we will con-
Equations (2.16)–(2.18) and (2.45) constitute infinitely sider specific subgroups of our symmetries and find that
many distinct symmetries, which will give infinitely many they recover a generalization of known conserved helici-
conserved currents. These can be conveniently written in ties, and we will see how these helicities become local in
the σ a -coordinates: specific limits.
√ Let us start by considering fluid relabelings of the form
Jfa = asf δ0a (4.1)
1
√ η i = √ ǫijk ∂j zk , zk = zk (σ i ) , (4.7)
 
ρ as
Jηai = a svi + nbi + mj mij η i δ0a (4.2)
µ √
√ where we included a so that η i is a spatial vector (rather
aρ i a
Jλai = m λi δ0 (4.3) than a vector density).8 The independence of η i on σ 0
µ
√ is guaranteed as far as the conservation of (4.1) is im-
Jχa = anχδ0a , (4.4) posed. Note that, up to the prefactor of √1as which can
where f, λi and χ are the functions introduced in (2.17), be fixed to 1 using (2.16), η i in (4.7) is divergenceless,
(2.18) and (2.45), respectively, and η i is an infinitesimal i.e. it is the infinitesimal generator of volume-preserving
transformation defined through f i = σ i + η i , where f i diffeomorphisms, or special diffeomorphisms.9 The asso-
was introduced in (2.16). Conservation of the currents is ciated charge is obtained by integrating over a σ 0 = c
then ∂a J a = 0. The conservation of the first two currents slice, where c is a constant,
is entropy and momentum conservation. Conservation of  
n ρ l
Z
the third current is a generalization of the “frozen-in” Hsdiff = d3 σ vi + bi + m mil ǫijk ∂j zk . (4.8)
condition for the magnetic field. The last one corresponds s sµ
to number conservation. The symmetry (2.19) gives the Under (2.21)-(2.25), the above quantity transforms as
constraint
√
Z
Hsdiff → d3 σ (vi − ∂i f

aρ i
∂i m =0, (4.5)
µ 
n ρ l
+ (bi + ∂i χ) + m (mil + 2∂[i λl] ) ǫijk ∂j zk .
which reduces to the constraint on the magnetic field s sµ
∂i B i = 0. One can easily verify that the conservation (4.9)
of (4.1)-(4.4) together with (4.5) are equivalent to the
equations of motion (2.1),(2.39) and (2.41), so (4.1)-(4.4) Invariance then imposes the conditions:
constitute a complete basis of conserved quantities. n 
ρ [l i]jk

ijk
The currents (4.1)-(4.4) are all non-local when written ∂i ǫ ∂j zk = 0, ∂i m ǫ ∂j zk = 0 .
in terms of the Eulerian variables T, µ, µN , uµ and hµ , in s sµ
the sense that (4.1)-(4.4) have a dependence on the initial (4.10)
values of such variables. For example, (4.2) depends on When these conditions are met, Hsdiff can be written as
the degree of freedom ϕ through bi = ∂i ϕ + vi ∂0 ϕ, where a local quantity. To show this, it is convenient to choose
we used (2.43) and (2.46), and we switched off the gauge a gauge for the symmetry (2.17) in which σ 0 = x0 on the
field Cµ = 0 for simplicity. Using (2.43) and (2.46), one slice σ 0 = c, upon which ∂I σ 0 , ∂i x0 = 0, and
∂ϕ
has ∂σ 0 = µN /T , which allows to express ϕ in terms of ǫIJK
Eulerian variables, ǫijk = ∂I σ i ∂J σ j ∂K σ k , (4.11)
J
σ0
µN (ζ, σ i )
Z
ϕ(σ 0 , σ i ) = dζ , (4.6)
−∞ T (ζ, σ i ) 8 We define ǫ123 = 1.
9 More explicitly,
where we took ϕ = 0 in the far past. We then see that √ one can fix the fluid relabeling symmetry (2.16)
by requiring as = 1 on a given constant-σ0 slice. Thanks to
(4.2) depends non-locally on the usual hydrodynamic (4.1), this condition is then preserved along all the flow, so that
variables. As we will see, this non-locality is related to one recovers the well-known special diffeomorphism invariance of
the fact that the currents (4.1)-(4.4) are not invariant ideal fluids.
 8

where I, J, K denote spatial indices in the xµ spacetime, Let us assume again that the plasma is isentropic
∂σi n/s =const. The above charge is invariant under all the
and where J is the determinant of ∂x J , and moreover
symmetries (2.16)-(2.19). In particular, under (2.17) one
1 finds, using the first eq. in (2.24),
∂α σ l vl = uα + ∂α σ 0 , (4.12)
b √
aρ i
Z
Hmshift → Hmshift − d3 σ ∂i f m η0 , (4.17)
and, for simplicity, we set all the background sources to µ
zero, i.e. gµν = ηµν , bµν = 0 and Cµ = 0. Pulling-back
(4.8) to xµ -coordinates then gives where the integrand can be easily shown to be a total
derivative, thanks to the conservation of (4.5), leading to
the vanishing of the integral. The same holds for trans-
 
h + ρµ µuL BL
Z
(isen) 3 IJK
Hsdiff = d x ǫ ∂J uI − BI z K , formation (2.45) acting on bi . Since η0 is an arbitrary
s sρ(u0 )2
(4.13) function satisfying the second eq. in (4.15), the conser-
where h = T s + µN n is the enthalpy density, and vation of Hmshift is equivalent to the conservation of
Bµ = ∂µ σ i µρ mi is the magnetic field. One physically Z
ρ  n 
relevant case in which condition (4.10) is satisfied is that dℓ mi vi + bi , (4.18)
M µ s
of isentropic fluids, for which n/s =const. [26], with no
electromagnetic contribution ml mil = 0. We leave it as where M is a line in the σ i -space generated by the vector
an open question whether there are other physically in- mi , and dℓ is the associated line element. In other words,
teresting situations where (4.10) is satisfied. for each magnetic flux tube we have an associated con-
Let us now set again ml mil = 0. As pointed out above, served quantity. Pulling back to xµ spacetime, eq. (4.16)
if the fluid is not isentropic the conserved helicity (4.8) becomes
is not a local expression of Eulerian variables. To under- Z
stand this better let us take again ml mil = 0 for sim- Hmshift = d3 x hbI uI η0 , (4.19)
plicity, without assuming isentropicity. The pull-back to
physical spacetime of Hsdiff is
where h is the enthalpy density defined before, bI = J 0I
Z 
h
 is the magnetic field in the “lab frame”, and where we
3 IJK
Hsdiff = d x ǫ ∂J uI zK set again to zero background sources, gµν = ηµν , Cµ = 0,
s
(4.14) bµν = 0. Eq. (4.19) is precisely the standard cross-helicity
3 n
Z
+ d x ∂I ϕǫ IJK
∂J zK . for an isentropic fluid. Relaxing isentropicity, eq. (4.16)
s cannot in general be written as a local quantity, and gives
the cross-helicity for general MHD. A discussion on non-
Clearly, if n/s is not constant the second term will in gen- local cross-helicity for traditional MHD can be found in
eral contribute, making Hsdiff non-local when expressed [25].
in terms of Eulerian variables, as discussed around (4.6). Lastly, we consider the generalized FFE action (3.5).
See Ref. [25] for an alternative discussion of this non- There is an accidental infinitesimal symmetry in this
local invariant in the case of vanishing electromagnetic case, given by the following time-dependent fluid rela-
fields ml mil = 0. A derivation of Kelvin’s theorem for belings η i :
fluids from symmetry was discussed in Ref. [12]. The
conservation of (4.8) is the extension of Kelvin’s circula- ǫijk
tion theorem to relativistic magnetohydrodynamics, and ηi = √ mj ∂k g, ∂0 g = 0, mi ∂i g = 0 . (4.20)
aµρ
includes the non-barotropic case.
Now let us consider the following transformations The associated conserved charge is
√ √
aρ
Z Z
1 aρ i
η i (σ j ) = η0 (σ j ) √ m, mi ∂i η0 = 0 . (4.15) Hffe = d3 σJη0i = d3 σ mij mi η j
as µ µ
1
Z Z
These fluid relabelings generate a shift along the mag- = d σmij ǫ ∂k g = − d3 σǫijk ∂k mij g ,
3 ijk
2
netic field, where the amount of shift is the same along (4.21)
the same magnetic field line. We can thus view them as
“magnetic shifts.” Time-independence of η i is guaran- where we used the identity
teed after accounting for the conservation
√
of (4.1) and
(4.3), which imply that √1as and µaρ mi are both time- ǫijk Mil Vj Wk = ǫijk W[i Vl] Mjk , (4.22)
independent. The conserved charge associated to this where Mjk is an antisymmetric tensor. Upon using the
symmetry is equations of motion, this quantity is time-independent:
√
n  aρ i
Z  Z
3
Hmshift ≡ d σ vi + bi m η0 . (4.16) ∂0 Hffe = d3 σ∂0 mij ǫijk ∂k g = 0 , (4.23)
s µ
 9
√  √
where we used that ∂0 aρ j
µ m mij = aρ j
µ m ∂0 mij = 0, has a non-trivial transformation under (2.17). In σ a -
coordinates:
and mi ∂i g = 0, together with the antisymmetry of ǫijk .
To get more physical intuition on this quantity, let us
Z Z
specify to Maxwell force-free electrodynamics. Pulling- δ d σa2 ǫ ∂0 mij vk = − d4 σa2 εijk ∂0 mij ∂k f
4 ijk

back to physical space-time and again fixing σ 0 = x0 on Z

= d4 σ (∂0 a2 )εijk mij ∂k f − ∂0 (a2 εijk mij ∂k f ) .


the hypersurface of integration, we find
Z (5.4)
Hffe = − d3 xgǫIJK ∂I JJK
In order for the above to vanish we then require a2 to
(4.24)
be a constant, independent of T . It is interesting that
Z Z
3 I 3
= −2 d xg∂I E = −2 d xgρq , the action formulation constraints a2 to be a constant.
A similar type of restriction was observed before in [14].
where ρq is the electric charge of the system. In the Comparing with the table below, we see that all the cou-
above steps, we used that J µν = 12 ǫµνρσ Fρσ and Gauss’ plings in (5.4), being linear in the electromagnetic field
law ∂i E i = ρq . In other words, Hffe is the electric charge break charge conjugation, while parity and time-reversal
integrated along magnetic flux tubes. are preserved. The coefficients a1 and a3 survive in the
static limit, and are thus of thermodynamic nature.

T µ vi aij mi mij
V. ACTION AT FIRST ORDER
T + + − + − +
P + + − + + −
In this section we discuss first-derivative contributions
to the MHD action. This entails writing the list of cou- C + + + + − −
plings which contain one derivative and are invariant un-
Using the variations given in appendix A, we find the
der the symmetries discussed in sec. II A. To make the
following first order corrections to the constitutive rela-
symmetries (2.16)-(2.19) more manifest, it will be con-
tions:
venient to use derivatives performed with respect to σ a
instead of xµ . To this aim, we introduce a “covariant” µν
T(1) =δεuµ uν + 2u(µ q ν)
derivative in the σ a -spacetime. Given a scalar ϕ(σ a ), its µν
(5.5)
ordinary σ i -derivative is not invariant under (2.17): J(1) =2u[µ κν] + j µν ,

∂i ϕ → ∂i ϕ + ∂i f ∂0 ϕ . (5.1) where q µ , κµ , j µν are transverse to uµ , and10

This has the opposite transformation compared to vi in δε = − 16 (T ∂T a3 − a3 )εαβγδ uα Hβγδ

(2.24), and thus we are lead to introducing the covariant + µ(T 2 ∂T a1 + 2T ∂T a3 − 2a3 )hα Ωα
derivative q = − (T ∂T a1 − a1 + 2∂T a3 − 2 aT3 )sµ1 + T a1 B µ
µ
 
di ϕ ≡ ∂i ϕ + vi ∂0 ϕ . (5.2) − 2 a1 + (1 − sTρµ )(a 2 + a3
T ) v1µ
1 (5.6)
One can verify that di ϕ transforms as a vector under − (a2 T + a3 )εαβγδ hα Hβγδ hµ
6
(2.16), and does not transform under any of the other
+ 2(a2 T + a3 )B α hα hµ
symmetries. The time derivative ∂0 ϕ does not transform
under any symmetry. κµ =a1 T Ωµ
We can now write down the general MHD action at j µν =2(a2 + ∂T a3 )εµναβ uα ∂β T
first derivative order, which is given by:
+ 2(T a2 + a3 )εµναβ uα ∂uβ ,
√
Z
S(1) = d4 x −g a1 εijk mi dj vk with
(5.3)
+ a2 T εijk ∂0 mij vk + a3 εijk di mjk ,

v1µ = µεµαβδ uα hβ (∂δ T + T ∂uδ )
Ωµ = εµαβγ uα ∂β uγ , B µ = εµαβγ uα ∂β (µhγ ) (5.7)
√
where εijk = ǫijk / a. For simplicity, we considered sµ1 = µεµαβγ uα hβ ∂γ T .
MHD without additional conserved currents, and we in-
cluded only terms that are linear in the electromagnetic
field. We also restricted to terms that satisfy parity
and time-reversal invariance. See Appendix B for MHD 10 Given that we consider only linear order in the magnetic field,
with one additional conserved current. Here, a1 and a3 we refrain from further decomposing those tensors into transverse
are arbitrary functions of T . The term multiplying a2 tensors to both uµ and hµ .
 10

The tensor structures in (5.7) are all independent after while the other operators defined in (5.12) are indepen-
imposing the ideal equations of motion. We rewrite these dent from the first order operators used in (5.10). The
constitutive relations in the hydrodynamic frame used in comparison then gives
[2, 27]:
ρµT  ρ sT a3

µν ˜ µν + δτ hµ hν + 2ℓ(µ hν) + tµν r̃⊥ = −2 a 1 − 2 (a
µ(ε+p) 2 + )
T(1) = δf ∆ ε + p ε+p T
(5.15)
µν
(5.8) η̃k = 0, η̃⊥ = 0 .
J(1) = 2m[µ hν] + sµν .
In other words, except r̃⊥ , the first order terms of [27] are
To go from frame (5.5) to frame (5.8) we redefine
not reproduced by the action (5.3). The fact that not all
uµ → uµ + δuµ , hµ + δhµ non-dissipative transport is captured by the action was
(5.9) already observed e.g. in 2+1 parity-breaking systems
T → T + δT, µ → µ + δµ ,
[15].11 Finally we note that [27] did not consider the
where δuµ , δhµ , δT, δµ are first order expressions, with most general first order constitutive relations, which is
uµ δuµ = hµ δhµ = 0. For a suitable choice of these ex- why in (5.10) we have additional new terms.
pressions, we find
   
δf = 16 ∂p T 2
∂ a3 αβγδ
ε u H − T ∂p a1 VI. CONCLUSIONS
∂ε ρ T T α βγδ ∂ρ ε µ

In this paper we constructed a variational principle
 
∂p
+ ∂ε T µ∂T (a1 + 2 T ) hα Ωα
a3
ρ for MHD following the approach of effective field theory.
   This theory extends previous variational formulations of
1 2 ∂µ a3 αβγδ
δτ = − 3 ρT ∂ε ρ ∂T T ε uα Hβγδ + T µ (µ hydrodynamics. The main new ingredients here are the
      implementation of the 1-form U (1) symmetry and the in-
+ρ ∂ρ ∂µ a1 2 ∂µ
∂T (a1 + 2 T ) hα Ωα
a3 troduction of the dual photon. We studied in detail the
µ + ρT ∂ε
ε ρ action for ideal MHD, and how, by enhancing the time
µ
ℓ = ˜ µ Ων
a 1 T µ∆ shift symmetry (2.17) to arbitrary time reparametriza-
ν
µν tions (3.3) one obtains a variational principle for FFE.
t =0
  We also observed that FFE can be recovered from a par-
T2
mµ = − ε+p 2 µs ∂T aT3 − ρ∂T aT1 sµ1 ticular limit of axion electrodynamics. It would be in-
  teresting to further investigate further this point, which
ρ
+ 2 ε+p a1 − 2 µ(ε+p)sT
(a2 + aT3 ) v1µ + T a1 ∆µν B ν might potentially lead to new connections between low-
density plasmas in astrophysics and topological insula-
sµν = 2(a2 + ∂T a3 )εµναβ uα hβ hγ ∂γ T tors in condensed matter.
2µ(T a2 + a3 )εµναβ uα hβ hγ ∂uγ . We then showed that certain subgroups of the sym-
(5.10) metries imposed on the action lead to various conserved
quantities, which are a generalization of local and non-
In [27], the charge-odd part of first order constitutive local helicities discussed in the literature. Curiously, we
relations is have not yet been able to recover the conserved magnetic
µν
helicity in ideal MHD. In the traditional formulation such
ℓµ = −η̃k Σ̃µ , tµν = −η̃⊥ σσ⊥ , mα = −r̃⊥ Ỹ α , quantity is the integral of an expression depending on the
(5.11) gauge potential. The latter is non-locally related to the
where dual photon used here, so it could perhaps be that the
magnetic helicity in this formulation cannot be written
Ỹ µ = εµναβ uν hα T ∂ Tµ + uγ hσ Hγβσ − hγ ∇γ (µhβ )


locally even using the Lagrangian variables.
µν (µ
σ̃⊥ = σ⊥γ εν)αβγ uα hβ Lastly, we discussed first order corrections to the ac-
˜α tion. These do not capture the full non-dissipative first
Σ̃µ = εµνρσ uν hρ ∆ β
σ h σαβ , order MHD. It would be interesting to understand if part
(5.12) of the missing terms can be described using a WZW-like
and where construction, similar to that of [14]. It is also desirable
µν
σ⊥ ˜ µα ∆
= (∆ ˜ µν ∆
˜ νβ − 1 ∆ ˜ αβ )σαβ to extend this variational principle to include dissipa-
2
tive terms as well as anomalies, which should be possible
σ µν = (∆µα ∆νβ − 12 ∆µν ∆αβ )∇(α uβ) (5.13)
∆˜ µν = ∆µν − hµ hν .
11
Note the (on-shell) relation In [27] it is noted that r̃⊥ is related to the Hall conductivity.
The fact that our action recovers r̃⊥ is then consistent with that
ρµT µ [15] recovers the Hall conductivity (at least for restricted values
v1µ = − Ỹ , (5.14) of the latter).
ε+p
 11

using the formalism introduced in [9]. As one further from which we obtain
application, we hope that this approach can be used to
address stability of objects where the magnetic field plays β
T ∂0 mij = λα γ
i λj (u Hγαβ + 2∇[α (µhβ] ) − 2µh[α ∂uβ] )
a significant role, in a similar spirit as what was done e.g. β
in [28, 29]. d[i vj] = T λα
i λj ∂[α uβ]
β
d[i mj] = λα
i λj ∂[α (µhβ] )
Acknowledgements
d[i mjk] = λµ[i λνj λρk] ( 13 Hµνρ + 2µhµ ∇ν uρ )
T ∂0 bi = λα γ
i (∂α µN + µN ∂uα − u Dαγ )
We are grateful to Hong Liu, Maxim Lyutikov and
β
Ellen Zweibel for insightful conversations. P. G. was d[i bj] = λα
i λj (Dαβ + 2µN ∂[α uβ] )
supported by a Leo Kadanoff Fellowship. The work of (A4)
D.T.S. is supported, in part, by US Department Of En-
ergy grant No. DE-FG02-13ER41958 and a Simons In-
vestigator Grant from the Simons Foundation. While
this work was being completed, we learned of [30], which 3. Ideal equations of motion
has some overlap with our manuscript.
In this section we give the explicit form of the ideal
equations of motion, which are necessary to find the on-
shell independent tensor structures used in the main text.
Appendix A: Useful formulae The ideal equations of motion are obtained by plugging
(2.29) and (2.49) in
1. Variations with respect to background sources
1
∇µ T µν = Hµαβ J αβ +Dνα Jα , ∇µ J µν = 0, ∇µ J µ = 0 .
2
Introduce (A5)
In the scalar sector we have five equations. Three of them
∂xµ ∂xµ are
λµi = i
+ vi 0 = di xµ . (A1)
∂σ ∂σ
uµ ∇ν T µν = −∂ε − (ε + p)θ + µρhµ hν ∇µ uν = 0
Variations with respect to the background are given by µhµ ∇ν J νµ = µ∂ρ + µρθ − ρµhµ hν ∇µ uν = 0 (A6)
µN ∇µ J µ = µN ∂n + nµN θ = 0 .
T µ ν ν) 1
δg T = u u , δg vi = T u(µ λi , δg mi = uµ uν mi
2 2 Summing them, we find
µ µ ν ν µ µN µ ν
δg µ = (u u − h h ), δg µN = u u
2 2
(ν −∂ε + µN ∂n + µ∂ρ + (µN n + µρ − ε − p)θ = −T ∇µ (suµ ),
δg aij = −λi(µ λjν) , δg mij = −2m[i λj] uµ) (A7)
ν) 1 where we used
δg bi = µN u(µ λi , δg εijk = − ∆µν εijk
2
[µ ν] ν]
δb mij = λi λj , δb mi = u[µ λi ε + p = sT + µρ + µN n, dε = T ds + µdρ + µN dn. (A8)
1
δC µN = uµ , δC bi = λµi , The other two scalar equations are
2
(A2)
where δg stands for the variation with respect to gµν , and hµ ∇ν T µν =(ε + p)hµ ∂uµ + ε+p−µρ−µN n µ
h ∂µ T
T
similarly for δb and δC .
− µ∇ν (ρhν ) = nh u Dνα − T ∂ν µTN
ν α

uµ ∇ν J νµ =∇µ (ρhµ ) − ρhµ ∂uµ = 0,

(A9)
2. Pull-backs
where the second equation is a constraint. Rearranging
these two equations gives
From (2.11) and (A1), and using ∂i ∂0 xµ = ∂0 ∂i xµ and
∂i ∂j xµ = ∂j ∂i xµ , we find
∇µ (ρhµ ) = − Tρ hµ ∂µ T
 
∂µ T
T ∂0 vi = λµi (∂µ T + T ∂uµ) hµ ∂uµ + n
hν uα Dνα − T ∂ν µTN .


T = ε+p−µρ
(A3)
T ∂0 λµi = λνi (∇ν uµ + uµ ∂uν ) , (A10)
 12

For the vector sector we have Appendix B: First order MHD with number
conservation

The first order action (5.3) admits additional terms

˜ µα ∇ν J να =ρ∆
∆ ˜ νµ (∂hν − hα ∇α uν ) = 0 when including a conserved number current. The most
 general first order action which is linear in the electro-
∆ ˜ ν ε + p (T ∂uν + ∂ν T ) + nT ∂ν µN
˜ µα ∇ν T να =∆ magnetic fields is
µ
T T
√
 Z
µ
+ ρT ∂ν − µρhα ∇α hν S(1) = d4 x −g a1 εijk mi dj vk
T
(B1)
1 + a2 T εijk ∂0 mij vk + a3 εijk di mjk
= ρuα hβ Hµαβ + nDµν uν + nhµ hα uν Dαν ,
2 + a4 εijk mi dj bk + a5 T εijk ∂0 mij bk ,


(A11)
where a1 , a3 , a4 are generic functions of T, µN . As dis-
cussed around (5.4), invariance under time shifts (2.17)
forces a2 to be constant, and in a similar way one sees
where ∆ ˜ µν = ∆µν − hµ hν . Using the tensor structures that (2.45) imposes constancy of a5 . Assuming that the
(5.7) and (B6), the second equation above can be written number density n does not change sign under neither
as of time-reversal, parity and charge-conjugation, the be-
havior under these transformations of µN and bi is the
same as that of T and vi , respectively, which implies that
the above action preserves parity and time-reversal, and
sv1µ + nv2µ breaks charge-conjugation. The coefficients a1 , a3 , a4 are
ρµ ˜ µ ναβγ of thermodynamic nature as they survive in the static
= ∆ ε uα ( 21 uρ Hρβγ + 2∂β (µhγ ) − 2µhβ ∂uγ ) . limit. To the constitutive relations (5.6) we have addi-
2 ν
(A12) tional contributions:

µN µ
δε = µN µ∂µN (a1 + a3 )hµ Ωµ − ∂µ a3 εαβγδ uα Hβγδ + (T ∂T a4 + µN ∂µN a4 )hµ N µ
12 N 2
q µ = −(T ∂µN a1 − 2∂µN a3 + µN ∂µN a4 )sµ2 − µN T ∂T aT4 sµ1 + a4 µN B µ − µN ( aT4 − ρµ
2s
a5 )v1µ
µN n µ µN
− (a4 + 2(1 − ρµ )a5 + (a2 T + a3 ) 2n α µ
ρµ )v2 + 2µN a5 B hα h − 6 a5 εαβγδ hα Hβγδ hµ (B2)
1
κµ = a4 N µ
2
j µν = 2a5 εµναβ uα (∂β µN + µN ∂uβ − Dβδ uδ ) + 2∂µN a3 εµναβ uα ∂β µN

and the additional first order constitutive relation for the and where
number current:
N µ = εµαβγ uα (Dβγ + 2µN ∂β uγ )
µ µ µ
J(1) = δnu + j , (B3) v2µ = µεµαβδ uα hβ (∂δ µN + µN ∂uδ − Dδσ uσ ) (B6)
sµ2 = µε µαβγ
uα hβ ∂γ µN
where j µ is transverse to uµ , and

δn = µ( 12 T ∂µN a1 + ∂µN a3 )hα Ωα

1 (B4)
− ∂µN a3 εαβγδ uα Hβγδ
12

a4 µ
j µ = −(∂T a4 − T )s1 − ∂µN a3 sµ2
2n
2s
+ a3 B µ − ( aT3 − ρµ a4 )v1µ + a4 v2µ (B5)
ρµ
a4
+ 2a4 B α hα hµ − εαβγδ hα Hβγδ hµ ,
6

[1] See, e.g., L. D. Landau and E. M. Lifshitz, Electro-
dynamics of Continous Media, 2nd ed., Butterworth-
Heinenmann, 1984.
[2] S. Grozdanov, D. M. Hofman and N. Iqbal, “Generalized
global symmetries and dissipative magnetohydrodynam-
ics,” arXiv:1610.07392 [hep-th].
[3] D. Gaiotto, A. Kapustin, N. Seiberg and B. Willett,
JHEP 1502, 172 (2015) doi:10.1007/JHEP02(2015)172
[arXiv:1412.5148 [hep-th]].
[4] J. D. Bekenstein and A. Oron, Phys. Rev. E
62, 5594 (2000) doi:10.1103/PhysRevE.62.5594 [astro-
ph/0002045].
[5] G. M. Webb and S. C. Anco, J. Phys. A 50,
no. 25, 255501 (2017) doi:10.1088/1751-8121/aa7181
[arXiv:1701.00521 [physics.plasm-ph]].
[6] K. Kumar, arXiv:1711.01142 [physics.plasm-ph].
[7] P.J. Morrison, M. Lingam and R. Acevedo, Phys. Plas-
mas 21, 082102 (2014).
[8] Y. Kawazura, Z. Yoshida, and Y. Fukumoto, J. Phys. A
47 (2014).
[9] M. Crossley, P. Glorioso, H. Liu and Y. Wang,
JHEP 1602, 124 (2016) doi:10.1007/JHEP02(2016)124
[arXiv:1504.07611 [hep-th]].
[10] F. M. Haehl, R. Loganayagam and M. Rangamani,
JHEP 1604, 039 (2016) doi:10.1007/JHEP04(2016)039
[arXiv:1511.07809 [hep-th]].
[11] K. Jensen, N. Pinzani-Fokeeva and A. Yarom,
arXiv:1701.07436 [hep-th].
[12] S. Dubovsky, T. Gregoire, A. Nicolis and R. Rat-
tazzi, JHEP 0603, 025 (2006) doi:10.1088/1126-
6708/2006/03/025 [hep-th/0512260].
[13] S. Dubovsky, L. Hui, A. Nicolis and D. T. Son, Phys. Rev.
D 85, 085029 (2012) doi:10.1103/PhysRevD.85.085029
[arXiv:1107.0731 [hep-th]].
[14] M. Geracie and D. T. Son, JHEP 1411, 004 (2014)
doi:10.1007/JHEP11(2014)004 [arXiv:1402.1146 [hep-
th]].
[15] F. M. Haehl and M. Rangamani, JHEP 1310, 074
(2013) doi:10.1007/JHEP10(2013)074 [arXiv:1305.6968
[hep-th]].
13
[16] H. Liu and P. Glorioso, PoS TASI 2017, 008 (2018)
doi:10.22323/1.305.0008 [arXiv:1805.09331 [hep-th]].
[17] J. Armas and A. Jain, arXiv:1808.01939 [hep-th].
[18] A. Lichnerowicz, “Relativistic Hydrodynamics and Mag-
netohydrodynamics,” W. A. Benjamin, Inc., New York
(1967).
[19] Y. Akamatsu and N. Yamamoto, Phys. Rev. Lett.
111, 052002 (2013) doi:10.1103/PhysRevLett.111.052002
[arXiv:1302.2125 [nucl-th]].
[20] G. D. Fleishman and I. N. Toptygin, “Cosmic Electrody-
namics,” Springer, New York (2013).
[21] S. E. Gralla and T. Jacobson, “Spacetime approach
to force-free magnetospheres,” Mon. Not. Roy. Astron.
Soc. 445, no. 3, 2500 (2014) doi:10.1093/mnras/stu1690
[arXiv:1401.6159 [astro-ph.HE]].
[22] R. Li, J. Wang, X. Qi and S. C. Zhang, Nature Phys.
6, 284 (2010) doi:10.1038/nphys1534 [arXiv:0908.1537
[cond-mat.other]].
[23] H. Ooguri and M. Oshikawa, Phys. Rev. Lett.
108, 161803 (2012) doi:10.1103/PhysRevLett.108.161803
[arXiv:1112.1414 [cond-mat.mes-hall]].
[24] J. Wang, B. Lian and S. C. Zhang, Phys. Rev.
B 93, 045115 (2016) doi:10.1103/PhysRevB.93.045115
[arXiv:1512.00534 [cond-mat.mes-hall]].
[25] G. Webb, “Magnetohydrodynamics and Fluid Dynam-
ics: Action Principles and Conservation Laws,” Springer
International Publishing (2018).
[26] E. Gourgoulhon, EAS Publ. Ser. 21, 43 (2006)
doi:10.1051/eas:2006106 [gr-qc/0603009].
[27] J. Hernandez and P. Kovtun, JHEP 1705, 001 (2017)
doi:10.1007/JHEP05(2017)001 [arXiv:1703.08757 [hep-
th]].
[28] J. L. Friedman, Commun. Math. Phys., 62 247-278
(1978).
[29] G. M. Vasil, D. Lecoanet, B. P. Brown, T. S. Wood
and E. G. Zweibel, Astrophys. J. 773, 169 (2013)
doi:10.1088/0004-637X/773/2/169 [arXiv:1303.0005
[astro-ph.SR]].
[30] J. Armas and A. Jain, “One-form superfluids & magne-
tohydrodynamics,” to appear.