PHYSICAL REVIEW D 101, 063530 (2020)

Viable gauge choices in cosmologies with nonlinear structures

Timothy Clifton, Christopher S. Gallagher , Sophia Goldberg, and Karim A. Malik
School of Physics & Astronomy, Queen Mary University of London, London E1 4NS, United Kingdom

(Received 2 January 2020; accepted 26 February 2020; published 25 March 2020)

A variety of gauges are used in cosmological perturbation theory. These are often chosen in order to
attribute physical properties to a particular choice of coordinates, or otherwise to simplify the form of the
resultant equations. Calculations are then performed with the understanding that they could have been done
in any gauge, and that transformations between different gauges can be made at will. We show that this
logic can be extended to the domain of large density contrasts, where different types of perturbative
expansion are required, but that the way in which gauges can be chosen in the presence of such structures is
severely constrained. In particular, most gauges that are commonly considered in the cosmology literature
are found to be unviable in the presence of nonlinear structures. This includes spatially flat gauge,
synchronous gauge, comoving orthogonal gauge, total matter gauge, N-body gauge, and the uniform
density gauge. In contrast, we find that the longitudinal gauge and the Newtonian motion gauge are both
viable choices in both standard cosmological perturbation theory, and in the post-Newtonian perturbative
expansions that are required in order to model non-linear structures.

DOI: 10.1103/PhysRevD.101.063530

I. INTRODUCTION curve of uμ . We have used letters from the Greek alphabet

to denote spacetime indices, lower case Latin letters to
Cosmological observations now span an enormous range
denote space indices, and chosen units such that c ¼ 1.
of scales, from individual galaxies all the way up to the
Treating the geometry and matter content of the Universe in
entire observable Universe. The theory of general relativity
this way are justified by their compatibility with a wide
is widely held to govern the gravitational interaction, and
array of cosmological observables [1–5] (although it is not
therefore the dynamics of matter, over this entire range.
entirely without controversy [6–8]).
However, within this theory the type of perturbative
In order to include small inhomogeneities in the geom-
expansions used to model small fluctuations existing on
etry of space-time one then writes the metric as
very large scales are quite different from those which
should be used on small scales. This paper investigates the
mathematical structure of these different types of expan- gμν ¼ ḡμν þ δgμν ; ð3Þ
sions, and uses the results to identify the choices of gauge
that are viable on both large and small scales in cosmology. where the condition of smallness is enforced by the
Throughout this article we will make the assumption that requirement δgμν ≪ ḡμν . While the coordinates used in
space-time is everywhere close to a single Friedmann- writing the line-element in Eq. (1) are in some sense unique
Lemaître-Robertson-Walker (FLRW) geometry, which can (up to spatial translations and rotations), the same cannot
be written as be said of the coordinates that one uses to express δgμν :
There is a “gauge” dependence, which can be viewed as a
ds2 ¼ ḡμν dxμ dxν ¼ a2 ðτÞð−dτ2 þ dx2 þ dy2 þ dz2 Þ; ð1Þ freedom to make an infinitesimal change in coordinates
xμ → xμ þ ξμ (or, equivalently, as a change in mapping
where we have assumed spatial flatness (small amounts of between the perturbed and unperturbed space-times).
spatial curvature can be included perturbatively). We also Without a single preferred set of coordinates one is forced
take the matter content of the space-time to be well to make a choice, and it is this that leads to the gauge
modeled by a single perfect fluid, such that its stress- problem in cosmology.
energy tensor can be written There are a wide array of gauges routinely used in
cosmology to associate the coordinates with either pre-
T μν ¼ ðρ þ pÞuμ uν þ pgμν ; ð2Þ ferred properties of the geometry or matter content. These
may be, but are not limited to, coordinate-induced folia-
where ρ and p are the energy density and isotropic pressure tions that are spatially flat or orthogonal to the world-lines
measured by an observer whose worldline is an integral of observers, time coordinates that correspond to proper

2470-0010=2020=101(6)=063530(16) 063530-1 © 2020 American Physical Society

CLIFTON, GALLAGHER, GOLDBERG, and MALIK PHYS. REV. D 101, 063530 (2020)

time of a class of observers, or choices that reduce the field δgij ¼ a2 ð−2ψδij þ 2E;ij þ 2Fði;jÞ þ hij Þ; ð6Þ
equations to some desirable form. Choosing such a set of
coordinates reduces the number of degrees of freedom that where Si and Fi are divergenceless vector field compo-
need to be solved for in a given physical problem, and nents, and where hij is divergenceless and tracefree. In
removes the possibility of spurious gauge artefacts being
terms of these new quantities we can expand the 4-velocity
introduced into the solutions. It also often allows the
of the fluid as
equations that describe that problem to be written in a
simplified way. It is therefore highly desirable to under-

1 1
stand the gauges that are possible, in any given situation. uμ ¼ 1 − ϕ þ v2 ; vi ; ð7Þ
a 2
We will investigate the viable gauge choices in cosmol-
ogy for the following two weak-field expansions:
where vi is the 3-velocity of the fluid. The Latin labels here
(i) Cosmological perturbation theory,
correspond to spatial indices, and the factors of two and
(ii) Post-Newtonian theory,
a ¼ aðτÞ are introduced for convenience only. Both cos-
which will both be discussed in detail in the next section.
mological perturbation theory and post-Newtonian theory
The former of these expansions is valid in a wide array of
are examples of weak field expansions, as they both typically
scenarios, as long as the density contrast and peculiar
have gravitational potentials of magnitude ≲10−4 . We will
velocities of matter fields remain small, while the second is
consider these two expansions in what follows.
valid for arbitrarily large density contrasts, but only on
If we are attempting to model a particular physical
small spatial scales. It is therefore the latter that should
situation with a weak-field expansion, then we need to
be used to describe the gravitational fields of the highly
identify which type of expansion(s) are applicable in that
nonlinear structures that exist at late times on scales
situation. Crucial factors in making such an assessment are
≲100 h−1 Mpc. We find that most of the gauges that are the magnitude of the velocity and magnitude of the density
commonly used in cosmology are not compatible with contrast of matter in the system. In systems with small
post-Newtonian theory, and therefore should not be used density contrasts we typically find v ∼ 10−4 (such that
when modeling nonlinear structures in the late universe. ϕ ∼ v), whereas in systems with highly nonlinear matter we
Exceptional cases are the longitudinal gauge and the typically have v ∼ 10−2 (such that ϕ ∼ v2 ). The former of
Newtonian motion gauge, which are both valid in the these relationships follows directly from being in a sit-
presence of nonlinear structures. uation in which a linearized version of the field equations
We begin in Sec. II with a general discussion of weak are applicable, such that all deviations from the homo-
field expansions in cosmology, and their application in geneous and isotropic FLRW background are “small”
the form of cosmological perturbation theory and post- (including the density contrast). On the other hand, the
Newtonian gravity (suitably adapted for cosmology). In latter can be readily identified from the virial theorem,
Sec. III we then discuss gauge transformations in these two which is derived in the Newtonian limit of Einstein’s
types of theory, before progressing to a discussion of the equations, and which is thought to be a good approxima-
commonly used gauges in cosmology in Sec. IV. Section V tions in most nonlinear situations where ϕ and v are
then contains a detailed analysis of the Newtonian motion both small.
gauge, where we find that this idea can be implemented in When considering the applicability of different
post-Newtonian gravity, but not in cosmological perturba- approaches to studying weak-field gravity, an important
tion theory. We then conclude in Sec. VI. factor is the spatial extent of the domain under consid-
eration. On spatial scales that are in some sense “small,”
II. WEAK FIELD EXPANSIONS and on which we have slowly moving matter, one can take a
IN COSMOLOGY limit of a post-Minkowski expansion in order to find that
the solutions to the field equations (which in general have
Weak field expansions are applicable in cosmology null characteristic curves) can be well approximated by the
when the geometry of space-time is, in some sense, close solutions to Poisson equations (which have support on
to a known exact solution (usually FLRW). If this is the spacelike hypersurfaces). This is what happens in post-
case, then we can write the metric of the space-time as that Newtonian theory, and it formally changes the structure of
of the background universe plus a small perturbation, as in the perturbative expansion by producing a hierarchy of
Eq. (3). If required, we can then irreducibly decompose the PDEs in spatial variables only. In general, no such
perturbations δgμν using Helmholtz theorem, so that they simplification can be made on horizon-sized scales, and
can be written as so the system of equations we are required to solve remains
a set of PDEs in both space and time. In this case post-
δg00 ¼ −2a2 ϕ ð4Þ Newtonian theory can no longer be applied.
Figure 1 shows the domains of applicability of both
δg0i ¼ a2 ðB;i − Si Þ ð5Þ cosmological perturbation theory and post-Newtonian

063530-2
VIABLE GAUGE CHOICES IN COSMOLOGIES WITH … PHYS. REV. D 101, 063530 (2020)

cosmological perturbation theory and post-Newtonian

theory separately, in order to identify suitable gauge
choices in each case.

A. Cosmological perturbation theory

Cosmological perturbation theory treats all perturbative
objects on an equal footing, and expands all relevant
equations order-by-order in every variable. This necessarily
results in taking both the peculiar velocities and density
contrasts to be “small.” It is a strict application of the
more general concept of perturbation theory to the case
of almost-FLRW cosmologies. As such, every quantity is
assigned an order-of-smallness using a parameter ϵ. This
FIG. 1. Domains of applicability of cosmological perturbation includes both perturbations to the geometry, as well as
theory (left of the blue line), and post-Newtonian theory (under perturbations to the background stress-energy tensor, such
the red line). Both formalisms are valid on small scales, when that in a perfect fluid-filled universe we have
velocities (and hence density contrasts) are small. The character-
istic spatial scale of a system is denoted L, and the Hubble scale is ϕ ∼ ψ ∼ v ∼ δ ∼ Si ∼ B;i ∼ Fi;j ∼ E;ij ∼ hij ∼ ϵ ≪ 1; ð8Þ
denoted LH .
where v is the peculiar velocity of the matter fields (as a
fraction of the speed of light), δ is the density contrast, and
theory. Cosmological perturbation theory can (in principle)
all geometric variables are defined in Eqs. (4)–(6). The
by applied to any spatial scale, as long as the magnitude of
reader will note that appropriate derivatives have been
all fluctuations from the FLRW background remain small.
added to B, E, and Fi , in order to make them dimension-
In the early universe this is widely thought to correspond to
less. The comparison between spatial tensor, vector, and
all spatial scales (unless primordial black holes are present),
scalar modes in this expression should be taken to mean
but in the late universe it means that scales on which the
that each component of the vectors and tensors are of the
density contrast have become nonlinear should not be
same order of magnitude as each of the scalars.
expected to be well modeled using such an approach. On
An important feature of cosmological perturbation
the other hand, for post-Newtonian theory to be applicable
theory is that, once the smallness of the quantities in
the range of spatial scales is restricted to those that are
Eq. (8) have been identified, all field equations and
“small” compared to the cosmological horizon (this state-
equations of motion can first be expanded to linear order,
ment will be made more precise later on). However, when
then quadratic order, and subsequently to all higher-orders
modeling systems on such scales using post-Newtonian
in these variables. This gives, for the background part of the
theory, the density contrast is only restricted by the
field equations, the usual Friedmann equations
condition that it does not cause the velocity of matter to
violate v ≪ 1. This is valid all the way down to compact 8πG 2 Λ 2
astrophysical objects, such as neutron stars and black holes. H2 ¼ ρ̄a þ a ð9Þ
3 3
In fact, it is of course already standard practice to apply
linear cosmological perturbation theory to fluctuations on and
large scales (or on any scale at early times), and to use the
equations of Newtonian gravity on the small scales where 4πG Λ
H_ ¼ − ðρ̄ þ 3p̄Þa2 þ a2 ; ð10Þ
structures become nonlinear at late times. These are the 3 3
leading-order parts of cosmological perturbation theory and
post-Newtonian theory, respectively. What is not currently and energy conservation equation
done consistently in much of the literature is to use the
next-to-leading order parts of each of these formalisms to ρ̄_ þ 3Hðρ̄ þ p̄Þ ¼ 0; ð11Þ
simultaneously calculate relativistic gravitational effects
from structures in each of their respective domains of where Λ is the cosmological constant. For the scalar part of
applicability. Instead, cosmological perturbation theory the field equations, to linear-order in ϵ, we find the
is extrapolated down into the nonlinear regime [9], or constraint equations
post-Newtonian gravity is extended up to horizon-sized
scales [10]. A more comprehensive approach, in which ∇2 ψ − 3Hðψ_ þ HϕÞ þ H∇2 σ ¼ 4πGa2 δρ ð12Þ
both expansions are deployed together, was only recently
introduced in Refs. [11–14]. In this paper we will consider ψ_ þ Hϕ ¼ −4πGa2 ðρ̄ þ p̄Þðv þ BÞ; ð13Þ

063530-3
CLIFTON, GALLAGHER, GOLDBERG, and MALIK PHYS. REV. D 101, 063530 (2020)

where v and σ ¼ E_ − B are the scalar parts of the velocity To solve the linear Newton-Poisson equation, and then
and shear, respectively. We have also used dots to denote solve the nonlinear Eulerian equations of motion in the
differentiation with respect to τ, and written the Hubble rate resultant gravitational field. In this situation it is clear that
_
as H ¼ a=a. The perturbed conservation equations, which the Newtonian limit of general relativity cannot be readily
give the evolution equations for δ and v, can be written as recovered from cosmological perturbation theory, and that
it is not a limiting case of that approach. Let us now
_
δρ_ þ 3Hðδρ þ δpÞ ¼ ðρ̄ þ p̄Þ½3ψ_ − ∇2 ðv þ EÞ ð14Þ consider how the Newtonian limit can be realized in
cosmologies with nonlinear structures, as part of a con-
∂ τ ½ðρ̄ þ p̄Þðv þ BÞ þ δp ¼ −ðρ̄ þ p̄Þ½ϕ þ 4Hðv þ BÞ sistent weak-field expansion.
ð15Þ
where p̄ and δp are the background and perturbation to the B. Post-Newtonian theory
isotropic pressure. We note the evolution equation for σ can In contrast to cosmological perturbation theory, post-
be obtained from the linearized Einstein equations, and can Newtonian theory requires that perturbations to the various
be written as σ_ þ 2Hσ − ϕ þ ψ ¼ 0. geometric and matter variables appear at different orders in
The linear cosmological perturbation equations (12)–(15) the expansion of the field equations and equations of
have a number of well-known properties, which greatly aid motion. It also requires that time-derivatives and space
one in finding and understanding their solutions. First, it can derivatives have different orders-of-magnitude associated
be seen that one does not need to know anything about the with them. These departures from the usual approach used
divergenceless vector or tensor degrees of freedom in the in standard cosmological perturbation theory increases its
metric, in order to write down a consistent set of equations complexity, but has the very considerable advantage that it
that can be used to solve for the scalar parts of the results in a theory that is valid in the presence of extremely
gravitational field. This is a result of the decomposition large density contrasts (which are not formally part of this
theorem, which holds for all linear equations in cosmological particular weak-field expansion at all). In this section we
perturbation theory (but does not hold at any higher order). will outline how post-Newtonian theory can be used in
A second property is that any derivatives acting on any cosmology.
quantities do not change its order of magnitude in the An essential property of the post-Newtonian expansion
expansion, and neither does multiplication or division by the is that it is a slow-motion, as well as weak-field, expansion.
background quantities H (the Hubble rate) or ρ̄ (the back- The condition of being slow-motion can be understood by
ground energy density). This property is very important for identifying the length scales involved in the type of
the theory, as together with the fact that all of the quantities physical systems we wish to model [15]. Let us start by
in Eq. (8) are dimensionless, it means that there is no limit to identifying a characteristic timescale for such a system, tc .
the spatial scales to which the theory is applied, unless that This could correspond to the orbital period for two bodies
limit also happens to imply that one of the quantities in in close proximity, or to the timescale required for a large
Eq. (8) is no longer small. body (such as a cluster) to assemble itself. The gravitational
What we cannot do with cosmological perturbation field is known to propagate at the speed of light, c, which
theory, however, is expect it to provide accurate solutions means that we can associate a characteristic length scale
when one of the quantities in Eq. (8) is no longer ≪1. This with the variations of our system:
is readily apparent from our attempts to model nonlinear
structures in the late universe. While we can use cosmo- λc ¼ cτc : ð16Þ
logical perturbation theory to extrapolate results into the
mildly nonlinear regime, it becomes highly problematic to Due to the high propagational speed of light, the value of λc
try and use it find results when the density contrast becomes is typically very large compared to the spatial scale of the
highly nonlinear (which is exactly the reason cosmologists system itself, which we denote as rc . Such a system will
use Newtonian N-body simulations). This problem arises typically contain matter that has 3-velocities of magnitude
because of the structure of the differential equations that vc ¼ rc =τc , so it can immediately be seen that the slow-
result from applying cosmological perturbation theory, motion condition vc ≪ c is equivalent to the condition that
which at each order of the expansion produce a linear rc ≪ λc . That is, systems that are considered slow motion
equation (or set of equations) in the new variables at that should exist on scales that are much less than the character-
order, and which means that even the mild (quadratic) istic length scale of the gravitational fields that are
nonlinearity that exists in the Newtonian equations of associated with them.
motion requires an infinite number of orders in perturbation In cosmology we are interested in structures that grow
theory in order to approach the true value (if the theory is over timescales that are comparable to the age of the
convergent at all). Universe (or less), so we have τc ∼ H−1 and therefore λc ∼
In the highly nonlinear regime it is much easier to follow cH−1 [11]. The characteristic length scale is therefore that
the approach prescribed in Newtonian N-body simulations: of the observable universe, and the slow motion condition

063530-4
VIABLE GAUGE CHOICES IN COSMOLOGIES WITH … PHYS. REV. D 101, 063530 (2020)

restricts us to considering systems that have a spatial extent where we have used ∇ to denote a spatial derivative, and
that is much smaller than that scale. More precisely, we are where similar results should hold for derivatives of v and p.
limited by If we again choose units such that rc ∼ 1, then we find that
time derivatives of matter variables add an extra order-of-
rc vc smallness in η.
∼ ∼ 10−2 ; ð17Þ
λc c Given that we expect gravitational perturbations on small
scales to inherit the time dependence of the stress-energy
where we have extracted a typical velocity from the virial tensor components that source them, this rule should extend
relation ϕ ∼ v2 =c2 , and used the empirical observation to metric perturbations as well. This leads to the general
that we have at most ϕ ∼ 10−4 for all systems of interest. rule
For a Universe with λc ∼ cH−1 ∼ 104 Mpc this gives us
rc ∼ 100 Mpc. This shows that if we wish to apply a slow- ∂=∂τ
∼ η; ð23Þ
motion condition, in the context of a weak-field expansion, ∂=∂x
then we should restrict the domain of applicability of such
an approach to systems that have a spatial scale that are i.e., that every time derivative adds an extra order-of-
∼100 Mpc (or less). smallness, when acting on either matter fields or gravita-
Let us now consider what the slow-motion condition tional fields. This rule means that the field equations that
implies for the form of the field equations, and their would normally correspond to null wave equations can
solutions. From Eq. (2) we can immediately identify instead be written at leading-order as Poisson equations:
T 00 ≃ ρc2 , T 0i ≃ ρvi c, and T ij ≃ ρvi vj þ pδij . These
1 1
immediately imply □δgμν ∝ T μν − gμν T ⇒ ∇2 δgμν ∝ T μν − gμν T;
2 2
T 0i vc T ij v2c
∼ and ∼ ; ð18Þ where □ ¼ ḡμν ∂ μ ∂ ν and ∇2 ¼ ḡij ∂ i ∂ j and T ¼ T μ μ .
T 00 c T 00 c2 The support for the integral that gives the function
which from the field equations implies δgμν ðt; xÞ in Eq. (24) should really be taken to be on the
past light cone L of the point P at position x. This shows
δg0i vc v2c the causal nature of Einstein’s theory, and the fact that
∼ and δg00 ∼ δgij ∼ : ð19Þ gravitational interactions propagate at the speed of light.
δg00 c c2
However, such an approach would be problematic to apply
in cosmology, as the integral for the gravitational fields at
If we now recall that ϕ ∼ v2 =c2 , and choose units such that
each point in space would have its own distinct domain
rc ∼ 1, then we see that we can write
(i.e., its own past lightcone). A fortunate consequence of
the slow-motion expansion is that on scales r ≲ rc we can
ϕ ∼ ψ ∼ Fi;j ∼ E;ij ∼ hij ∼ ρ ∼ η2 ð20Þ
approximate the past light cone of a point as being given
by a spacelike surface S of constant τ [15], as shown in
and
Fig. 2. This is because the time taken for a null signal to go
from one side of such a domain to the other is negligible
Si ∼ B;i ∼ η3 and p ∼ η4 ; ð21Þ
compared to τc , and means that we can find solutions
for δgμν ðt; xÞ at some time τ by simply integrating over
where we have introduced η ¼ vc =c ≪ 1 as the order of
a suitable region of a hypersurface of constant τ. The
smallness in this expansion. The reader will note that
integrals for the gravitational field value at neighboring
different geometric and matter perturbations appear at
points in space then have their support on overlapping
different orders in the expansion, and that the density
domains, and the whole process of finding solutions is
contrast does not have to be small at all.
considerably simplified.
From purely kinematic considerations we can immedi-
The lowest order fields, using the rules outlined above to
ately identify that the slow-motion criterion also has
consequences for the order-of-smallness of quantities that Oðη2 Þ, then give us the following constraint and evolution
equations [16]:
contain derivatives. If the constituent parts of a system are
moving slowly, then this immediately implies that the time- 2 8πG 2 Λ 2
variation of state variables such as energy density and H2 þ ∇2 ψ ¼ ρa þ a þ Oðη4 Þ ð24Þ
3 3 3
pressure will also be only slowly varying. This can be
quantified in terms of η as follows: and
ρ_ r 1 4πG Λ
∼ c ∼ η;
∇ρ τc
ð22Þ H_ − ∇2 ϕ ¼ − ðρ þ 3p̄Þa2 þ a2 þ Oðη4 Þ; ð25Þ
3 3 3

063530-5
CLIFTON, GALLAGHER, GOLDBERG, and MALIK PHYS. REV. D 101, 063530 (2020)

should expect strong cosmological backreaction, and a

violation of our initial ansatz of a perturbed FLRW space-
time, but we will not consider this further here.
It is noteworthy that the Friedmann equations and the
Newton-Poisson equations occur at the same order of
magnitude in this expansion. This shows the well known
fact that post-Newtonian expansions are not (strictly speak-
ing) a direct application of perturbation theory, a fact that
is already obvious from the leading-order conservation
equations:

ρ_ þ 3Hρ þ ∂ i ðρvi Þ ¼ 0; ð28Þ

and
FIG. 2. The past lightcone L of a point P following a worldline
W. The support for the metric perturbations at P can be ρv_j þ ρvi ∂ i vj þ ρHvj ¼ −ρ∂ j ϕ − ∂ j p; ð29Þ
approximated as being located on the spacelike hypersurface
S, as long as rc ≪ λc . which are the standard equations of Newtonian gravity on
an expanding background. These equations are clearly
nonlinear, and therefore cannot be considered as being
where H ∼ τ−1 _ −2 2
c ∼ η and H ∼ τc ∼ η (in units such that the result of an application of perturbation theory, in the
rc ∼ 1). These equations are a combination of the Hubble strict sense outlined in the previous section. Nevertheless,
equations and the Newton-Poisson equations for ϕ and ψ, they are well-defined, and the post-Newtonian expansion
which both occur at the same order in this expansion. itself constitutes a well-defined expansion of the field and
Within a region of space S, of scale r ≲ rc , they can be conservation equations, which has been extensively applied
transformed to the usual Newtonian equations through a in other areas of gravitational physics.
suitable choice of coordinates. It is also known that many All equations in this section, as well as higher-order
such regions can be patched together to form a cosmology equations, can be obtained by direct coordinate trans-
described by a line-element that is close to a single global formation from their form in the post-Minkowski approach
FLRW solution [17], as in Eq. (1). [17]. Their existence shows the direct correspondence
If we integrate Eqs. (24)–(25) over S, and divide by the (through an isomorphism) of the expansion about a
spatial volume of that region, we recover the standard Friedmann space that we have just outlined, and the
Friedmann equations (9)–(10), as well as the Newton- extremely well studied expansions that are usually per-
Poisson equations in an expanding background, formed around Minkowski. They can be used to further
justify the order of magnitude we have associated with the
∇2 ϕ ¼ ∇2 ψ ¼ 4πGδρa2 ; ð26Þ various quantities we have required, as well as under-
standing some of the features that have become apparent.
as long as we choose the boundary condition First, the applicability to scales r ≪ H−1 can be seen to
Z correspond directly to the requirement that v ≪ c. Second,
∇ϕ · dS ¼ 4πGhδρia2 ; ð27Þ the mixing of Friedmann and Newton-Poisson equations
∂S can be shown to be a result of the leading-order part of the
cosmological expansion arising from the motion of par-
where we have written ρ ¼ ρ̄ þ δρ, and where hδρi is the ticles under the influence of Newtonian gravitational fields
volume averaged value of δρ in the region S. It is important in the perturbed Minkowski approach. We refer the reader
to note that there is no assumption made about the relative to Refs. [16,17] for more details of these observations.
sizes of ρ̄ and δρ here; the post-Newtonian expansion is
specifically constructed to allow for large density contrasts
III. GAUGE TRANSFORMATIONS
to be consistently modeled, and this means that δρ=ρ̄ is
allowed to be much larger than one without signaling any A gauge transformation is a transformation (or set of
breakdown in the weak-field expansion. transformations) that preserve the structure of a theory, and
The left-hand side of Eq. (27) can be set to zero if one that can be used to remove (or fix) redundant degrees of
chooses S to have periodic boundary conditions, which freedom. In general relativity the term gauge is sometimes
also sets the right-hand side to zero (as the average of this used to describe the invariance of the theory under general
spatial domain would automatically be equal to the global coordinate transformations. When applied to a weak-field
average of the cosmology). In general, it seems conceivable expansion of the field equations, however, it refers to the set
that Eq. (27) may not be satisfied. If this is so, then one of transformations that leave the theory that results from

063530-6
VIABLE GAUGE CHOICES IN COSMOLOGIES WITH … PHYS. REV. D 101, 063530 (2020)

that expansion unchanged. In such cases the transforma- ξμ ∼ ϵ: ð36Þ

tions must be small (in the sense of the expansion), and
must provide a map from the full set of perturbed quantities If ξμ were larger than this, then the metric (after the
into itself (see e.g., [18]). gauge transformation), could no longer be written as
A gauge transformation can be said to be either active perturbed FLRW.
or passive. We will use the former of these approaches, Using this information, and the decomposition given
which changes the point in space-time that a given set of in Eqs. (4)–(6), we then find the standard set of gauge
coordinate values identifies. The action of such a trans- transformations:
formation on a tensor field T can be written in the form

T → T̃ ¼ eLξ T; ð30Þ ϕ → ϕ þ Hξ0 þ ξ_0

B → B þ ζ_ − ξ0
where Lξ is the Lie derivative with respect to the gauge
generator ξ, and where a tilde denotes the field T after the Si → Si − ζ_i
transformation. The exponential map is used here to ensure ψ → ψ − Hξ0
that the group structure of the diffeomorphisms associated
with the transformations is preserved. E→Eþζ
Treating the coordinates on the manifold as a set of four Fi → Fi þ ζ i
scalar fields, the gauge transformation in Eq. (30) can be
seen to be equivalent to hij → hij ; ð37Þ

α∂
xμ ðpÞ → xμ ðqÞ ¼ eξ α jp xμ ðpÞ; ð31Þ where we have decomposed the spatial part of the gauge
generator, such that ξi ¼ ζ ;i þ ζ i , where ζi is divergence-
where xμ ¼ xμ ðpÞ on the right-hand side is evaluated at less. It can be seen that all metric perturbations transform
some point p, while xμ ðqÞ is evaluated at the point q under a general gauge transformation, with the notable
located along the flow of the gauge generator field ξ exception of hij (at linear order).
from p. Geometrically, one can think of this as a trans- Similarly, we can calculate how the components of the
formation in the map used to identify a point in the stress-energy tensor transform under a gauge transforma-
background space-time with a point in the perturbed tion with ξμ ∼ ϵ. For a perfect fluid, these components can
space-time. Let us now focus on the set of active gauge be written
transformations that are possible in both cosmological
perturbation theory and post-Newtonian theory. T 00 ¼ −ðρ̄ þ δρÞ ð38Þ

A. Cosmological perturbation theory

T 0i ¼ðρ̄ þ p̄Þðvi þ B;i − Si Þ ð39Þ
If we apply the transformation in Eq. (30) to the metric
we obtain:
T ij ¼ðp̄ þ δpÞδi j ð40Þ
1
g̃μν ¼ gμν þ Lξ gμν þ L2ξ gμν þ    ; ð32Þ
2 and under the transformation (30) therefore give
which gives the various components of the metric trans-
forming as ρ̄_
δ → δ þ ξ0
ρ̄
g00 → g00 þ ξμ ∂ μ g00 þ 2g0μ ξ_μ þ    ð33Þ δp → δp þ ξ0 p̄_

g0i → g0i þ ξμ ∂ μ g0i þ g0μ ξμ;i þ giμ ξ_μ þ    ð34Þ vi → vi − ξ_i : ð41Þ

gij → gij þ ξμ ∂ μ gij þ 2gμði ξμ ;jÞ þ    ð35Þ It is again apparent that all perturbed quantities transform
under a general gauge transformation, in the matter sector
where the ellipses in these expression denote terms that as well as the gravitational sector.
are quadratic or higher-order in the gauge generator, ξμ . It The gauge transformations given above are very well
can be seen immediately from these equations that every known in perturbation theory. What is less well known in
component of the gauge generators must be of the same cosmology are the transformation properties of post-
order-of-magnitude in the perturbative expansion as the Newtonian variables under the most general possible gauge
metric perturbations, i.e., that transformation. We will spell this out below.

063530-7
CLIFTON, GALLAGHER, GOLDBERG, and MALIK PHYS. REV. D 101, 063530 (2020)

B. Post-Newtonian theory standard approach to perturbations in cosmology, as the

Let us now consider linear gauge transformations of the different orders of magnitude of the different components
metric, as given to linear order in ξμ by Eqs. (32)–(35). of the gauge generators significantly alter the possible
These equations take exactly the same form in post- transformations of the metric.
Newtonian theory as they do in cosmological perturbation For example, if we investigate the leading-order
theory, as so far they have only assumed that ξμ is small, terms that are generated in the transformation of the
and that terms quadratic or higher can therefore be 00-component of the metric we now find
neglected to leading order.
Let us now consider the size of each of the terms that g00 → g00 þ Oðη2 ξi Þ þ Oðηξ0 Þ; ð45Þ
results from the gauge transformation in Eq. (32). Starting
with the ij-component of the metric, we can see that the which, using Eq. (44), can be seen to be equivalent to
transformation (35) has terms of magnitude
g00 → g00 þ Oðη4 Þ: ð46Þ
gij → gij þ Oðξi Þ þ Oðηξ0 Þ; ð42Þ
This means that the leading-order perturbation to the
where we use OðxÞ to mean terms of order x or smaller in 00-component of the metric, which exists at order η2, is
the post-Newtonian expansion. In deriving this expression entirely unchanged by the gauge transformations that this
we have used the rules for the order-of-magnitude of each theory admits, and that only subleading terms are affected.
of the components of the metric, and the relative size of This result severely limits what can be done with gauge
their derivatives, as outlined in Sec. II B. We have also transformations in the presence of nonlinear structures
taken the components of the gauge generator ξμ to obey the when using post-Newtonian expansions.
same rules with respect to derivative operators (i.e., that Having identified the orders of magnitude of the leading-
time derivatives of these objects are small compared to order parts of the gauge generators, we can now find the
space derivatives). leading-order parts of the gauge transformations of each of
Performing the same analysis for the 0i-component of the degrees of freedom in the metric. These are given by
the metric we find
g00 → g00 þ ξ0 g_00 þ ξi g00;i þ 2g00 ξ_0 þ Oðη5 Þ ð47Þ
0
g0i → g0i þ Oðηξ Þ þ Oðξ Þ:
i
ð43Þ
g0i → g0i þ g00 ξ0;i þ gij ξ_j þ Oðη4 Þ ð48Þ
In order for the gauge transformed ij and 0i-components of
the metric to be no larger than η2 and η3 , respectively, we gij → gij þ 2gkði ξk ;jÞ þ Oðη3 Þ; ð49Þ
can see that we must have
which gives
ξi ∼ η 2 and ξ0 ∼ η3 : ð44Þ
ϕ → ϕ þ Hξ0 þ ξ_0 þ ϕ;i ξi ð50Þ
If the former of these conditions was violated, and the
magnitude of ξi were allowed to be larger than η2 , then it
B → B þ ζ_ − ξ0 ð51Þ
can be seen that the gauge transformed ij-components of
the metric would have terms larger than η2 . This would
mean that they would be larger than allowed in the post- Si → Si − ζ_i ð52Þ
Newtonian expansion of the metric, and the transformation
would not be part of the gauge group of the theory. ψ →ψ ð53Þ
Similarly, if the magnitude of ξ0 were allowed to be any
larger than η3 then the gauge transformed 0i-components of E→Eþζ ð54Þ
the metric would contain parts that were larger than η3 ,
which is also forbidden for the same reason. Fi → Fi þ ζ i ð55Þ
The appearance of different components of the gauge
generators with different orders of magnitude is entirely hij → hij ; ð56Þ
absent from the approach to cosmological perturbation
theory, but is entirely consistent with the post-Newtonian where we have kept terms up to order η4 in ϕ, as this is the
approach to gravity [11,12]. Indeed, we have already seen order required to obtain the post-Newtonian equations of
that the perturbations to different components of the metric motion for massive test particles. The reader will note that
can have leading-order parts with different orders of as well as the leading-order part of ϕ (at order η2 ) being
magnitude. The consequences of this, however, do produce gauge invariant, the same can also be said of the leading-
results that are unexpected from the perspective of the order parts of ψ and hij . Only the last of these was invariant

063530-8
VIABLE GAUGE CHOICES IN COSMOLOGIES WITH … PHYS. REV. D 101, 063530 (2020)

under general gauge transformations in cosmological post-Newtonian theory ψ is gauge-invariant, and so this
perturbation theory. gauge is impossible to realize (though it is possible to set E
Expanding the stress-energy tensor in the parameter η and Fi to zero).
we find
B. Synchronous gauge
T 00 ¼ −ρð1 þ v2 Þ þ Oðη5 Þ ð57Þ
Synchronous gauge is defined by setting
T 0i ¼ ρvi þ Oðη4 Þ ð58Þ ϕ ¼ B ¼ Si ¼ 0: ð65Þ

T ij ¼ δi j p þ ρvi vj þ Oðη5 Þ; ð59Þ This gauge is popular for numerical studies, but does not
uniquely define the time-slicing (this can be fixed by
which under the gauge transformation (30) gives choosing an additional gauge condition, for example that
the perturbed dark matter 3-velocity vanishes). In this
μ→μ ð60Þ gauge it can be seen that the time coordinate corresponds
to the proper time of comoving observers at fixed spatial
Π → Π þ ξi ðln μÞ;i ð61Þ coordinates. Synchronous gauge is routinely used in a wide
variety of cosmological calculations, and is the default
p→p ð62Þ gauge for CMBFAST [20] and CAMB [21].
This gauge is obtained within cosmological perturbation
vi → vi ; ð63Þ theory by solving the differential equations ξ_0 þ Hξ0 ¼
−ϕ and ξ_i − ξ0 ¼ −B;i þ Si . However, it cannot be
where we have written ρ ¼ μð1 þ ΠÞ, such that μ ∼ η2 is achieved in post-Newtonian theory as in this case ϕ is
the rest-mass density and Π ∼ η2 is the specific energy gauge invariant at leading order (though B and Si are not).
density. All lowest-order parts of the matter variables can
be seen to transform trivially, with an additional term at C. Comoving orthogonal gauge
order η4 appearing in the transformation of Π.
The comoving orthogonal gauge is defined by the gauge
conditions
IV. STANDARD GAUGE CHOICES
IN COSMOLOGY vi ¼ 0 and B;i ¼ Si ; ð66Þ
Choosing a gauge is often essential in cosmology.
which states that the fluid 3-velocity and 3-momentum
However, the majority of gauges that are frequently used
vanish. In this gauge the constant time hypersurfaces are
in the literature are not viable choices in the presence of
orthogonal to the fluid 4-velocity. In cosmological pertur-
nonlinear structures modeled by post-Newtonian theory.
In this section we will review some of the “popular” bation theory this gauge choice requires ξ_i ¼ vi and
gauges used in cosmological perturbation theory (see, ξ0;i ¼ ξ_i . Once more, this gauge choice cannot be realized
e.g., Ref. [19] for details). These gauges are usually in post-Newtonian theory, this time because vi is gauge
specified by assigning a particular set of variables to zero, invariant at leading order (though B;i and Si are not, and
either in the gravitational sector or the matter sector (or a could be set equal).
mixture of both). In each case we will also comment on
the whether such a gauge can be achieved in the post- D. Total matter gauge
Newtonian expansion. The total matter gauge is related to the comoving
orthogonal gauge. It has the gauge conditions
A. Spatially flat gauge
The spatially flat gauge is defined by the choice vþB¼0 and E ¼ 0 ¼ Fi : ð67Þ

ψ ¼ E ¼ Fi ¼ 0; ð64Þ Evaluating the density contrast in the total matter gauge,

and the metric potential ϕ in the longitudinal gauge, allows
which leaves the induced 3-metric on spatial hypersurfaces one to write the cosmological perturbation equations in the
unperturbed (in the absence of tensor perturbations). This form of a Poisson equation, equivalent to its Newtonian
gauge is often used for the calculation of observables counterpart [22,23]. This gauge can be realized in cosmo-
during inflation. logical perturbation theory by choosing ξ0 ¼ v þ B and
It can be seen from Eq. (37) that this gauge can be readily ξi ¼ −E;i − Fi . It cannot be realized in post-Newtonian
achieved in cosmological perturbation theory by choosing theory as the condition v þ B ¼ 0 has parts at order η and
ξ0 ¼ ψ=H, and ξi ¼ −E;i − Fi . On the other hand, in the η3 , the former of which cannot be satisfied as it corresponds

063530-9
CLIFTON, GALLAGHER, GOLDBERG, and MALIK PHYS. REV. D 101, 063530 (2020)

to v ¼ 0, and v is gauge invariant (though the other anisotropic stress, the field equations in this gauge give
conditions are again possible). ψ ¼ ϕ, which allows one to write the governing field
equations from cosmological perturbation theory in a form
E. Uniform density gauge that is very close to the Newtonian equation.
This is the only standard gauge choice we have found in
In the uniform density gauge we use the density
the cosmology literature that can be fully specified in both
perturbation, or equivalently the density contrast, to specify
cosmological perturbation theory and post-Newtonian
the temporal gauge condition
theory. This is achieved in both cases by taking ξ0 ¼
δρ ¼ 0: ð68Þ B þ E_ and ζ ¼ −E. This gauge choice therefore appears to
be particularly valuable if one wishes to perform calcu-
To fix the spatial gauge we can choose, for example, lations in both the linear and nonlinear regimes of cosmol-
E ¼ 0 ¼ Fi . In cosmological perturbation theory this ogy, and to find results in each case that can be consistently
choice of specification of the temporal gauge can be written related to one another.
as ξ0 ¼ −δρ̄=ρ̄_, but such a condition is impossible to
implement in the post-Newtonian approach as μ is gauge V. NEWTONIAN MOTION GAUGE
invariant in this setup. The Newtonian motion gauge was recently introduced
by Fidler et al in Ref. [27], and further developed in
F. N-body gauge Ref. [28]. It is based on the idea of fixing a gauge such that
The N-body gauge is formulated in a situation where the gravitational field equation and equations of motion of
test particles take the same form that they do in the
v þ B ¼ 0; ð69Þ Newtonian problem, i.e., such that

as in the total matter gauge, above. The remaining gauge μ̃_ þ 3Hμ̃ þ ∂ i ðμ̃ṽi Þ ¼ 0 ð72Þ
freedoms are then used to set the so-called “counting
density” associated with N bodies equal to the leading- μ̃ṽ_j þ μ̃ṽi ∂ i ṽj þ μ̃Hṽj ¼ −μ̃∂ j Ũ − ∂ j p̃; ð73Þ
order part of the energy density. This condition requires that
the scalar deformation of the spatial volume is set to zero, where Ũ must satisfy an equation of the form
which can be written as [24]
∇2 Ũ ¼ 4πδμ̃a2 : ð74Þ
1
ψ þ ∇2 E ¼ 0: ð70Þ
3 The variables ṽi , μ̃, Ũ, and p̃ can be seen to satisfy
equations of exactly the same form as the Newtonian
This can be achieved in cosmological perturbation theory equations (28) and (29), but are not themselves the
by taking ξ0 ¼ v þ B and setting the spatial gauge using Newtonian variables. Instead, they should be thought of
the solution of ∇2 ζ ¼ 3Hðv þ BÞ − ∇2 E − 3ψ. Now, as variables that are constructed from objects that are
while v þ B ¼ 0 still cannot be realized in post- defined in the corresponding relativistic problem.
Newtonian gravity, the condition given in Eq. (70) is This is a very interesting idea, as almost all N-body
achieved by taking ∇2 ζ ¼ −∇2 E − 3ψ. It may therefore simulations are based on the equations that result from
be possible to develop new variants of the N-body gauge considering Newtonian physics on an expanding back-
with alternative specification of the temporal gauge con- ground. The Newtonian motion gauge therefore allows
dition, such as the N-boisson gauge [25,26]. Newtonian N-body simulations to be interpreted in a
relativistic context, and therefore for relativistic gravita-
G. Longitudinal gauge tional effects to be extracted from nonrelativistic simula-
tions. This is achieved by deforming the coordinate system
Longitudinal gauge (also referred to as conformal
(using gauge transformations) such that the coordinate
Newtonian, or as part of Poisson gauge) is defined by
positions of particles are the same as those that would
the scalar gauge conditions
appear in the Newtonian problem. Here we will investigate
this idea in the context of cosmological perturbation theory
B ¼ E ¼ 0: ð71Þ
and post-Newtonian theory.
As the scalar shear is given by σ ¼ E_ − B, this gauge is
also known as “zero-shear” gauge (the spatial hyper- A. Cosmological perturbation theory
surfaces have vanishing shear). These gauge conditions It is clear that the nonlinear equations (72) and (73) will
give a diagonal metric tensor for the scalar perturbations, not be able to be satisfied by the linearized equations of
which considerably simplifies calculations. If there is no first-order cosmological perturbation theory. In order to

063530-10
VIABLE GAUGE CHOICES IN COSMOLOGIES WITH … PHYS. REV. D 101, 063530 (2020)

establish whether or not this gauge is viable in such an This derivation of this equation has used the Euler
approach, we therefore propose to expand Eqs. (72)–(73) equation (15) from cosmological perturbation theory in
perturbatively, and see whether or not the equations of order to eliminate v,_ and can be seen to be equivalent to
cosmological perturbation theory can be manipulated into Eq. (4.5) of Ref. [28] (though without specifying any
the form of the equations that result. restriction on the time gauge).
We start by writing Further manipulation, using the linear equations from
cosmological perturbation theory with δp ¼ 0, we find that
μ̃ ¼ μ̄˜ þ δμ̃ þ Oðϵ2 Þ ð75Þ Eq. (82) can be rewritten as

ṽi ¼ δṽi þ Oðϵ2 Þ: ð76Þ ̈ þ HE_ − 4πa2 ρ̄E ¼ 3ρ̄ΦR ;

E ð84Þ
To background order we find that Eq. (72) can be written as
where
μ̄˜_ þ 3H μ̄˜ ¼ 0; ð77Þ Z
R0
ΦR ¼ −a2 ðτÞ d3 x0 ð85Þ
which is clearly of the same form as the energy conserva- jx − x0 j
tion equation (11), as long as p̄ ¼ 0, whilst the momentum
conservation equation (73) is automatically satisfied. We and where R ¼ ψ − Hðv þ BÞ is the curvature perturba-
therefore have μ̄˜ ¼ ρ̄, and the requirement p̄ ¼ 0 (i.e., that tion in comoving orthogonal gauge (a well-known gauge
we consider dust, at the level of the background). invariant quantity, frequently used in cosmology). The
Next, we can study the perturbed equations at first order. boxed equation (84) needs to be satisfied if the
For Eqs. (72)–(73) this gives Newtonian motion gauge is to be realized in cosmological
perturbation theory.
δμ̃_ þ 3Hδμ̃ þ μ̄˜δṽi;i ¼ 0 ð78Þ At this point it seems worth pointing out that the choices
for the effective Newtonian variables made in Eqs. (80) and
μ̄˜δṽ_j þ μ̄˜Hδṽj ¼ − μ̄˜Ũ ;j − δp̃;j : ð79Þ (81) are not unique, though they did lead to a viable
application of the idea of a Newtonian motion gauge. We
If we now consider the equation of energy conservation at explore an alternative choice in Appendix A. There also
first order in cosmological perturbation theory (14), then remain gauge freedoms in the generators ξ0 and ζ i , which
we see that if we choose δμ̃ ¼ δρ − 3ρ̄ψ þ ρ̄∇2 E and δṽi ¼ can be set to any convenient values without affecting the
vi then we can write this equation in the form of the property that the equations of motion can be written in a
linearized Newtonian equation (78). This gives us Newtonian form.

μ̃ ¼ ρ̄ þ δρ − 3ρ̄ψ þ ρ̄∇2 E þ Oðϵ2 Þ ð80Þ B. Post-Newtonian theory

and The lowest-order parts of T μν ;ν ¼ 0 very obviously give
equations that are in the form of the Newtonian equations of
ṽ ¼ v þ Oðϵ2 Þ; ð81Þ motion in post-Newtonian theory, as this is exactly how the
Newtonian limit is derived in the context of relativistic
where ṽ and v are the scalar parts of ṽi and vi , respectively. gravity. The challenge in this case is therefore to put the
For this correspondence to follow we also require δp ¼ 0 equations of motion at first post-Newtonian order into the
(i.e., that the requirement to consider dust is extended to form of the Newtonian equations.
linear order). The relativistic field equations and equations of motion,
The combination of variables used to construct μ̃ and ṽ in to the required orders, are given in Appendices B and C,
Eqs. (80) and (81) have not yet required any choice of respectively. If we consider the time component of
gauge. Let us now consider the linearized momentum T μν ;ν ¼ 0 to order η5 we see that see that we can write
conservation equation (79) that these variables must satisfy. the equation of relativistic energy conservation in the form
Substituting in from Eq. (81), and taking δp̃ ¼ 0, we find of the Newtonian equation of mass conservation (72), as
that the following equation must be satisfied: long as we have p ¼ 0 (i.e., dust). In this case the effective
Newtonian variables are as follows:
B_ þ HB ¼ Ũ − ϕ; ð82Þ

1
μ̃ ¼ μ 1 þ v2 − 3U þ Π þ ∇2 E þ Oðη5 Þ ð86Þ
where Ũ must now satisfy 2

∇2 Ũ ¼ 4πa2 ρ̄ðδ − 3ψ þ ∇2 EÞ: ð83Þ and

063530-11
CLIFTON, GALLAGHER, GOLDBERG, and MALIK PHYS. REV. D 101, 063530 (2020)


1 2 All of the potentials in this expression can be determined
ṽ ¼ v 1 − v þ U þ Oðη4 Þ;
j j
ð87Þ
2 from post-processing Newtonian N-body simulations, and
in writing f in this way we have chosen to eliminate the
where U is the potential defined in Eq. (B5). It is notable vector gravitational potential W i using the identities in
that no choice of gauge is yet required in order to put the Appendix B.
relativistic energy conservation equation into the form of Putting the metric into Newtonian motion gauge, to first
Eq. (72), and that the variables μ̃ and ṽi therefore exist in all post-Newtonian order, requires choosing a gauge such that
possible gauges. Eq. (88) is true. Solving this equation will almost certainly
The space component of T μν ;ν ¼ 0 to order η6 is more have to be done numerically, but once numerical solutions
complicated, but we find that it can be written in the form in have been obtained then it is clear from Sec. III B that the
Eq. (29) if the following is true: gauge can be fixed by a suitable choice of ξi . This can be
seen from Eqs. (54)–(55). This leaves total gauge freedom
0 ¼ −3vj U_ − Hvj v2 þ v2 U ;j − 4vj vk U ;k þ 2UU;j in the time component of ξ0 , which can be set to any
convenient value while still maintaining the required
− 2E;ij U ;i þ 2vk E_ ;jk þ vk vn E;jkn − 2Fði;jÞ U;i
property that the equations of motion of test particles obey
ð4Þ
þ 2vk F_ ðj;kÞ þ vk vn Fj ;nk þ ϕ;j − ðŨ − UÞ;j equations of the same form as they do in Newtonian
physics.
þ B_ ;j þ HB;j − S_j − HSj − 2vk S½j;k ; Once in this gauge, all relativistic gravitational degrees
of freedom can be derived by inverting Eqs. (86) and (87),
where we have divided through by a common factor of μ so and then by using the solutions given in Appendix B for the
that this equation is order η4, and where it has been assumed metric perturbations, together with the numerical solutions
that hij ¼ 0 ¼ p. The expression above represents three for E and Fi , which can be obtained from Γi . This gives
separate equations, with four degrees of freedom in the enough information to calculate all relativistic gravitational
choice of gauge. It is expected that all of these equations effects up to first post-Newtonian order, by post-processing
should be able to be satisfied in many ways (probably a Newtonian N-body simulation. It is remarkable that this is
infinitely many ways), with one degree of gauge freedom possible, and that one can in principle obtain a relativistic
remaining. simulation in this way. We have made no approximations in
Manipulating the above expression, using the solutions obtaining this result other than the fluid being dust, which
to the field equations given in Appendix B, as well as the includes the particle interpretation by simply taking the
P
identities in Appendix C, allows us to write this as the mass density to be μðxÞ ¼ i mi δðx − xi Þ, for i particles
following differential equation: with masses mi and positions xi .

d2 Γj dΓj VI. DISCUSSION

2
þH þ U ;ij Γj − Φ7i;ij ¼ f ; ð88Þ
dτ dτ
We have considered the structure of gauge transforma-
where we have define Γj ¼ E;j þ Fj , and where we have tions in both cosmological perturbation theory (applicable
introduced the material derivative on large scales) and post-Newtonian perturbation theory
(applicable on small scales). While both treatments of
d ∂ ∂ gravitational fields have their own well defined gauge
¼ þ vi i ð89Þ
dτ ∂τ ∂x problems, we find that most of the particular gauge choices
that are used in cosmology are not valid using post-
and the potential Φ7i , which is defined by Newtonian theory in the presence of nonlinear structures.
Z In particular, the spatially flat gauge, the synchronous
μ0 Γ0i 3 0
Φ7i ¼ −a2 ðτÞ d x: ð90Þ gauge, the comoving orthogonal gauge, the total matter
jx − x0 j gauge, the N-body gauge, and the uniform density gauge are
all beyond the limits of what it is possible to achieve by
The source function in Eq. (88) is a function of the applying an infinitesimal coordinate transformation in the
potentials given in Appendix B, such that f ¼ fðU; vi ; V i ; post-Newtonian sector.
Φ1 ; Φ2 ; δΦ2 ; A; BÞ, and is given explicitly by In contrast, the Newtonian motion gauge appears to be
1 1 well-defined in both the post-Newtonian and cosmological
f ¼ −2Φ1;j − 6Φ2;j þ 5δΦ2;j þ A;j þ B;j perturbation theory treatments of gravitational fields, but
2 2
requires numerical integration of a nonlocal differential
_
− 4V j − 4HV j − 8v V ½j;i − 3v V i;i
i j
equations (84) and (88) in order to be applied in practice. If
− 2ðU 2 Þ;j − 3Hvj U þ Hvj v2 − v2 U;j þ 4vj vi U;i : this is possible, then it should allow one to post-process the
results of existing cosmological Newtonian N-body sim-
ð91Þ ulations in order to derive relativistic corrections to

063530-12
VIABLE GAUGE CHOICES IN COSMOLOGIES WITH … PHYS. REV. D 101, 063530 (2020)

gravitational fields on all scales, and to determine the μ̃ ¼ ρ̄ þ δρ − 3ρ̄ψ þ Oðϵ2 Þ ðA1Þ
effects of these fields on observables without having to
perform additional simulations. This is an intriguing possi- and
bility, which we hope to explore further in future studies.
The one standard gauge choice that remains viable, in
both of the weak-field treatments that we have considered, ṽ ¼ v þ E_ þ Oðϵ2 Þ; ðA2Þ
is the longitudinal gauge. The fact that the cosmological
perturbation theory equations give sensible results in this which would have also satisfied the linearized Newtonian
gauge, even when the density contrast of matter becomes equation of energy conservation (78). Substituting into the
nonlinear, is well known in the cosmology community. linearized momentum conservation equation (79) from
Here we formalize this result, and explain its veracity by Eq. (A2), and taking δp̃ ¼ 0, we find that in this case
showing that this gauge is the only commonly used cosmo- the following equation must be satisfied:
logical gauge that can be realized in post-Newtonian
expansions (which are purposefully constructed to model ̈ þ HE_ ¼ −Ũ;
v_ þ Hv þ E ðA3Þ
weak-field gravity in such situations). This provides sup-
port for the use of longitudinal gauge in studies that attempt
to simultaneously model both small-scale nonlinear struc- where Û is
tures as well as linear structures on large scales, see, e.g.,
the numerical code GEVOLUTION [29] or the 2-parameter ∇2 Û ¼ 4πa2 ρ̄ðδ − 3ψÞ: ðA4Þ
perturbative approach [14].
The fact that one cannot use gauge transformations to This can be equivalently written as
change coordinates from a coordinate system that is
perturbatively close to FLRW to a synchronous coordinate
system in post-Newtonian theory has interesting conse- σ_ þ Hσ ¼ ϕ − Û; ðA5Þ
quences, but must be interpreted with some care. In
particular, this result does not imply that it is impossible where σ ¼ E_ − B. This equation needs to be satisfied if the
in general to find a coordinate system where the time Newtonian momentum conservation equation is to be true
coordinate corresponds to the proper time of observers for the variables in Eqs. (A1)–(A2).
comoving with matter (in fact, this is always possible when We can now use the evolution equation for σ, given
the matter content is dust [30]). Instead, it means that the below Eq. (15), to find that the condition in Eq. (A5) is
difference between a synchronous coordinate system, and equivalent to requiring
the coordinates of a perturbed FLRW space-time, cannot be
related by an infinitesimal gauge generator. That is, the
Hσ ¼ Û − ψ: ðA6Þ
difference between these two different notions of time is
large, in the sense defined by the perturbative expansion,
and is therefore unattainable by gauge transformations. In order to evaluate this equation, we can use Eqs. (12) and
Such a result would appear to have significance for a (A4) to write
number of studies that use proper time in the presence of
nonlinear structures, such as the calculation of galaxy bias ∇2 ðŨ − ψÞ ¼ H∇2 σ − 12πa2 ρ̄ðψ − Hðv þ BÞÞ: ðA7Þ
on hypersurfaces of constant proper time [31]. It may also
go someway to explaining the vastly different expectations This equation makes it clear that Eq. (A6) is satisfied for
that different groups of cosmologists appear to have when ρ̄ ≠ 0 if and only if
considering the problem of cosmological backreaction (see,
e.g., [32] and [7]).
ψ − Hðv þ BÞ ¼ 0; ðA8Þ
ACKNOWLEDGMENTS
where sensible boundary conditions have been assumed.
We are grateful to Christian Fidler and Marco Bruni
One may now note that the combination of variables on
for fruitful conversations, and very helpful comments. We
the left-hand side of Eq. (A8) is equal to the curvature
acknowledge financial support from the STFC under Grant
perturbation in comoving orthogonal gauge, R ¼ ψ −
No. ST/P000592/1.
Hðv þ BÞ. It is therefore impossible to satisfy Eq. (A8),
and hence Eq. (79), by a choice of gauge using the variables
APPENDIX A: NONVIABLE CHOICE FOR
in Eqs. (A1) and (A2). This shows that the choice of
NEWTONIAN MOTION GAUGE IN CPT
effective variables is extremely important in the implemen-
Instead of Eqs. (80) and (81), we could have equally well tation of this gauge, and that Newtonian motion gauge
chosen our effective Newtonian variables to be cannot be achieved in every case.

063530-13
CLIFTON, GALLAGHER, GOLDBERG, and MALIK PHYS. REV. D 101, 063530 (2020)

APPENDIX B: SOLVING POST-NEWTONIAN Let us now consider the 00 field equation. To order η2,
EQUATIONS IN ARBITRARY GAUGE and using the results above, this equation gives
If we write the line-element in as a weak-field perturba-
tion of FLRW, as in Eqs. (4)–(6), then the leading-order −3H_ ¼ 4π μ̄a2 − Λa2 ; ðB10Þ
part of the ij field equation can be written as
which can clearly be seen to correspond to the second
_ i j − ∇2 hij
a2 Rð2Þi j ¼ ∇2 ψ − ðϕ − ψÞ;ij þ ð2H2 þ HÞδ Friedmann equation in (10), and which together with
¼ 4πμa2 δi j þ Λa2 δi j ; ðB1Þ Eq. (B3) gives the first Friedmann equation (9). The same
field equation to order η4 gives
where the superscript in Rð2Þi j indicates that this is the part
of this tensor at order η2 in the v=c expansion, in appropri- ̈
−a2 Rð4Þ0 0 ¼ ∇2 ϕð4Þ þ ∇2 B_ þ H∇2 B þ 3U
ately chosen units. This equation immediately tells us that − U ;j ∇2 Fj − 2Fðj;kÞ U;jk − 2U ;j U ;j
ð2Þ ð2Þ
∇2 hij ¼ 0 ⇒ hij ¼ 0 ; ðB2Þ ̈ − H∇2 E_ − U;j ∇2 E;j
− ∇2 E
ð2Þ _
− 2U;jk E;jk þ 6HU_ þ 6HU:
where appropriate boundary conditions have been used to
infer the result on the right. The same equation also gives us
This result can now be used with the relevant field equation,
H_ þ 2H2 ¼ 4π μ̄a2 þ Λa2 ðB3Þ

2 ð4Þ0 2 2 p
−a R 0 ¼ 4πμa 2v þ Π þ 3 ; ðB11Þ
and μ

ϕð2Þ ¼ ψ ð2Þ ¼ U ; ðB4Þ to find

where U is the Newtonian gravitation potential that satisfies ð2Þ
∇2 U ¼ 4πδμa2 , i.e., ϕð4Þ þ B_ ð3Þ þ HBð3Þ − U2 − U ;j Fj
Z ̈ ð2Þ − HE_ ð2Þ − U;j Eð2Þ
−E
δμðx0 ; τÞ 3 0 ;j
Uðx; τÞ ¼ −a2 ðτÞ d x; ðB5Þ
jx − x0 j ¼ 12 Φ1 þ 3Φ2 − 5δΦ2 þ Φ3 þ 3Φ4
where δμ ¼ μ − μ̄ is the mass density contrast, and appro- − δΦ5j;j − δΦ6 þ 32 A þ 32 B;
priate boundary conditions have again been applied.
Next, if we consider the leading-order part of the 0j field where
equation, we find Z
μ0 ½v0 · ðx − x0 Þ2 3 0
1 1 A¼ −a2 ðτÞ dx
−a2 Rð3Þ0 j ¼ ∇2 Sj þ ∇2 F_ j þ 2U_ ;j þ 2HU ;j jx − x0 j3
2 2 Z
μ0 dv0 3 0
2
¼ −8πμvj a ; ðB6Þ B ¼ −a2 ðτÞ 0 ðx − x0 Þ · dx
jx − x j dτ
Z
where we have used the results in Eq. (B4). Solving this μ0 v02 3 0
Φ1 ¼ −a2 ðτÞ dx
equation we find jx − x0 j
Z
ð3Þ μ0 U0 3 0
Sj þ F_ j ð2Þ ¼ −2ðV j þ W j Þ ; ðB7Þ 2
Φ2 ¼ −a ðτÞ dx
jx − x0 j
Z
where the potentials on the right-hand side are given by δμ0 U 0 3 0
δΦ2 ¼ −a2 ðτÞ dx
Z jx − x0 j
2
μðx0 ; τÞv0j 3 0 Z
V j ¼ −a ðτÞ dx ðB8Þ μ0 Π0 3 0
jx − x0 j Φ3 ¼ −a2 ðτÞ dx
jx − x0 j
and Z
2 p0
Z Φ4 ¼ −a ðτÞ d 3 x0
2
μðx0 ; τÞv0 · ðx − x0 Þðx − x0 Þj 3 0 jx − x0 j
W j ¼ −a ðτÞ d x; ðB9Þ Z
jx − x0 j3 δμ0 Fj 3 0
δΦ5j ¼ −a2 ðτÞ dx
jx − x0 j
and where we have used the result U_ ;j þ HU ;j ¼ Z
1 2 δμ0;j E;j 3 0
2 ∇ ðW j − V j Þ, which can be proven using the continuity δΦ6 ¼ −a2 ðτÞ d x:
equation. jx − x0 j

063530-14
VIABLE GAUGE CHOICES IN COSMOLOGIES WITH … PHYS. REV. D 101, 063530 (2020)

These are all standard potentials used in post-Newtonian 1

T 00 ¼ μð1 þ v2 − 2U þ ΠÞ þ Oðη5 Þ
gravity, with the exceptions of δΦ2 , δΦ5j and δΦ6 , which a2


we have introduced here. Primed quantities in these 1 i 1 2 1
equations should be taken to mean they are functions of T ¼ 2 μv 1 þ v − U þ Π þ 2 pvi þ Oðη6 Þ
0i
a 2 a
the primed coordinate positions, such that μ0 ¼ μðx0 ; τÞ, for
1
example. T ij ¼ 2 ðμvi vj þ pδij Þ þ Oðη7 Þ
In deriving this last result we have used the following a
identities: 1
þ 2 ½ðμΠ þ pÞvi vj þ 2Upδij − 2ðEi; þ Fi Þp:
a
2U ;i U ;i − 8π ρ̄Ua2 ¼ ∇2 ðU 2 − 2Φ2 Þ
Likewise, the connection coefficients, up to the required
−U ;j ∇2 Fj − 2Fðj;kÞ U ;jk ¼ ∇2 ðδΦ5j;j − U ;j Fj Þ order, are given by
−U;j ∇2 E;j − 2U;jk E;jk ¼ −∇2 ðU ;j E;j − δΦ6 Þ
Γ0 00 ¼ H þ U_ þ Oðη4 Þ
_
̈ þ 2HU_ þ ðH2 þ HÞU 1
U ¼ − ∇2 ðA þ B − Φ1 Þ; Γ0 0i ¼ U ;i þ Oðη3 Þ
2
Γ0 ij ¼ δij H þ Oðη2 Þ
the last of which is proven using the continuity equation.
ð4Þ
We have also used the following identities in Sec. V of the Γj 00 ¼ U ;j þ ϕ;j þ B_ ;j þ HB;j − S_ j − HSj
paper:
þ 2UU;j − 2E;ij U;i − 2Fði;jÞ U;i þ Oðη5 Þ
U_ þ HU ¼ −V j;j ¼ W j;j _ − S½j;k þ E_ ;jk þ F_ ðj;kÞ þ Oðη4 Þ
Γj 0k ¼ δjk ðH − UÞ
V ½j;k ¼ W ½j;k Γj kn ¼ −δjn U;k − δjk U;n þ δkn U;j þ E;jkn þ Fj ;nk þ Oðη3 Þ;
V_ j − W_ j ¼ −HðV j − W j Þ þ A;j þ B;j − Φ1;j : and the square root of the determinant of the metric is
pffiffiffiffiffiffi
All identities can be proven under the assumption that −g ¼ a4 ð1 − 2U þ ∇2 EÞ þ Oðη3 Þ:
boundary terms vanish, as would occur (for example) in a
In deriving all of these equations we have used the results
space with periodic boundary conditions. ð2Þ
from Eqs. (B2) and (B4) to eliminate hij , and to write ϕð2Þ
and ψ ð2Þ in terms of U.
APPENDIX C: POST-NEWTONIAN EQUATIONS The order η3 part of the time component of Eq. (C2) can
OF MOTION immediately be seen to reproduce the Newtonian equation of
The equations of motion of post-Newtonian gra- mass conservation on an expanding background, as given in
vity can be obtained by expanding the conservation Eq. (28). Likewise, the order η4 of the spatial components of
equations: Eq. (C2) gives the momentum conservation equation from
(29), once we set ϕ ¼ U. The next nonvanishing contribu-
T μν ;ν ¼ 0; ðC1Þ tions to the conservation equations (C2) come at order η5 in
the time component, and order η6 in the space components.
which can be conveniently written as These correspond to first post-Newtonian order, in the
normal language of this type of weak-field expansion
pffiffiffiffiffiffi pffiffiffiffiffiffi
∂ ν ð −gT μν Þ þ Γμ ρν −gT ρν ¼ 0; ðC2Þ (though on a cosmological background here). We will
now consider each of these in turn.
where g is the determinant of the metric. The metric in Calculating the order η5 part of the time component of
Eqs. (4)–(6) gives the components of the stress-energy Eq. (C2), and simplifying using the momentum conserva-
tensor to the required order as tion equation (29), gives



1 2
0 ¼ ∂ τ a μ v − 3U þ Π þ ∇ E þ ∂ j ½a3 μvj ð−2U þ Π þ ∇2 EÞ þ a3 pðvj ;j þ 3HÞ:
3 2
ðC3Þ
2

Calculating the order η6 part of the spatial components of Eq. (C2), and taking p ¼ 0, gives

063530-15
CLIFTON, GALLAGHER, GOLDBERG, and MALIK PHYS. REV. D 101, 063530 (2020)


4 j 1 2
0 ¼ ∂ τ a μv v − 3U þ Π þ ∇ E þ ∂ k ½a4 μvj vk ð−2U þ Π þ ∇2 EÞ
2
2
ð4Þ
þ a4 μ½U ;j ð2v2 − 4U þ Π þ ∇2 EÞ þ ϕ;j þ HB;j þ B_ ;j − HSj − S_j þ 2UU;j − 2E;ij U;i − 2Fði;jÞ U;i 
 
4 _ _ _ 1 n 1 n j
− 2a μv S½j;k þ δjk U − E;jk − Fðj;kÞ þ v U;k − v E;jkn − v F ;kn :
k j
2 2

These are all of the equations that are required to calculate the trajectories of test particles to first post-Newtonian order.

[1] N. Aghanim et al., arXiv:1807.06209. [18] K. A. Malik and D. R. Matravers, Classical Quantum
[2] T. M. C. Abbott et al., Phys. Rev. D 98, 043526 (2018). Gravity 25, 193001 (2008).
[3] R. J. Cooke, M. Pettini, K. M. Nollett, and R. Jorgenson, [19] K. A. Malik and D. Wands, Phys. Rep. 475, 1 (2009).
Astrophys. J. 830, 148 (2016). [20] U. Seljak and M. Zaldarriaga, Astrophys. J. 469, 437
[4] M. Betoule et al., Astron. Astrophys. 568, A22 (2014). (1996).
[5] S. Alam et al., Mon. Not. R. Astron. Soc. 470, 2617 (2017). [21] A. Lewis, A. Challinor, and A. Lasenby, Astrophys. J. 538,
[6] A. G. Riess, S. Casertano, W. Yuan, L. M. Macri, and D. 473 (2000).
Scolnic Astrophys. J. 876, 85 (2019). [22] D. Wands and A. Slosar, Phys. Rev. D 79, 123507
[7] T. Buchert and S. Räsänen, Annu. Rev. Nucl. Part. Sci. 62, (2009).
57 (2012). [23] J. C. Hidalgo, A. J. Christopherson, and K. A. Malik,
[8] R. J. van den Hoogen, arXiv:1003.4020. J. Cosmol. Astropart. Phys. 08 (2013) 026.
[9] E. Villa and C. Rampf, J. Cosmol. Astropart. Phys. 01 [24] C. Fidler, C. Rampf, T. Tram, R. Crittenden, K. Koyama,
(2016) 030; 05 (2018) E01. and D. Wands, Phys. Rev. D 92, 123517 (2015).
[10] I. Milillo, D. Bertacca, M. Bruni, and A. Maselli, Phys. Rev. [25] C. Fidler, A. Kleinjohann, T. Tram, C. Rampf, and K.
D 92, 023519 (2015). Koyama, J. Cosmol. Astropart. Phys. 01 (2019) 025.
[11] S. R. Goldberg, T. Clifton, and K. Malik, Phys. Rev. D 95, [26] N. E. Chisari and M. Zaldarriago, Phys. Rev. D 83, 123505
043503 (2017). (2011).
[12] S. R. Goldberg, C. Gallagher, and T. Clifton, Phys. Rev. D [27] C. Fidler, T. Tram, C. Rampf, R. Crittenden, K. Koyama,
96, 103508 (2017). and D. Wands, J. Cosmol. Astropart. Phys. 09 (2016) 031.
[13] C. Gallagher and T. Clifton, Phys. Rev. D 98, 103516 [28] C. Fidler, T. Tram, C. Rampf, R. Crittenden, K. Koyama,
(2018). and D. Wands, J. Cosmol. Astropart. Phys. 12 (2017)
[14] C. Gallagher, T. Clifton, and C. Clarkson, arXiv: 022.
1910.04894. [29] J. Adamek, D. Daverio, R. Durrer, and M. Kunz, J. Cosmol.
[15] E. Poisson and C. M. Will, Gravity: Newtonian, Post- Astropart. Phys. 07 (2016) 053.
Newtonian, Relativistic (Cambridge University Press, [30] L. D. Landau and E. M. Lifschitz, The Classical Theory of
Cambridge, England, 2014). Fields (Pergamon Press, Oxford, 1971).
[16] T. Clifton, Classical Quantum Gravity 28, 164011 (2011). [31] J. Yoo, Phys. Rev. D 90, 123507 (2014).
[17] V. A. A. Sanghai and T. Clifton, Phys. Rev. D 91, 103532 [32] C. Clarkson, G. Ellis, J. Larena, and O. Umeh, Rep. Prog.
(2015); 93, 089903 (2016). Phys. 74, 112901 (2011).

063530-16