Extending MGCAMB Tests of Gravity To Nonlinear Scales

Download as pdf or txt
Download as pdf or txt
You are on page 1of 10

Extending MGCAMB tests of gravity to nonlinear scales

Zhuangfei Wang,1 Daniela Saadeh,2 Kazuya Koyama,2 Levon


Pogosian,1 Benjamin Bose,3 Lanyang Yi,4, 5 and Gong-Bo Zhao5, 4, 6
1
Department of Physics, Simon Fraser University, Burnaby, BC, V5A 1S6, Canada
2
Institute of Cosmology & Gravitation, University of Portsmouth, Portsmouth, PO1 3FX, UK
3
Institute for Astronomy, University of Edinburgh,
Royal Observatory, Blackford Hill, Edinburgh, EH9 3HJ, U.K.
4
University of Chinese Academy of Sciences, Beijing, 100049, P.R.China
5
National Astronomical Observatories, Chinese Academy of Sciences, Beijing, 100101, P.R.China
6
Institute for Frontiers in Astronomy and Astrophysics,
Beijing Normal University, Beijing, 102206, P.R.China
arXiv:2406.09204v1 [astro-ph.CO] 13 Jun 2024

Modified Growth with CAMB (MGCAMB) is a patch for the Einstein-Boltzmann solver CAMB for cosmo-
logical tests of gravity. Until now, MGCAMB was limited to scales well-described by linear perturbation
theory. In this work, we extend the framework with a phenomenological model that can capture
nonlinear corrections in a broad range of modified gravity theories. The extension employs the
publicly available halo model reaction code ReACT, developed for modeling the nonlinear corrections
to cosmological observables in extensions of the ΛCDM model. The nonlinear extension makes it
possible to use a wider range of data from large scale structure surveys, without applying a linear
scale cut. We demonstrate that, with the 3×2pt Dark Energy Survey data, we achieve a stronger
constraint on the linear phenomenological functions µ and Σ, after marginalizing over the additional
nonlinear parameter p1 , compared to the case without the nonlinear extension and using a linear
cut. The new version of MGCAMB is now forked with CAMB on GitHub allowing for compatibility with
future upgrades.

I. INTRODUCTION Einstein equations relating the Newtonian gauge gravi-


tational potentials Ψ and Φ to the perturbations in the
The discovery of the accelerated expansion of the Uni- matter stress-energy:
verse [1, 2], the unknown nature of dark matter and the k 2 Ψ = −4πG µ(a, k) a2 [ρ∆ + 3(ρ + P )σ] , (1)
apparent fine-tuning needed to reconcile the observed k (Φ + Ψ) = −4πG Σ(a, k) a2 [2ρ∆ + 3(ρ + P )σ] , (2)
2
value of the cosmological constant, Λ, against predictions P P
from quantum field theory [3, 4] all stimulated explo- where ρ∆ ≡ i ρi ∆i and (ρ + P )σ ≡ i (ρi + Pi )σi ,
ration of alternative theories of gravity. The interest in ∆i , σi , ρi and Pi are the comoving matter density con-
extensions of General Relativity (GR) is further fuelled trast, the anisotropic stress, background density and
by the opportunities to test gravity at cosmological scales pressure with i ∈ {b, c, γ, ν} standing for baryons, cold
offered by the existing and forthcoming cosmological sur- dark matter, photons and neutrinos, respectively. The
veys. phenomenological functions µ ≡ µ(k, a) and Σ ≡ Σ(k, a)
Modified Growth with CAMB (MGCAMB) is a patch for the in Eqs. 1-2 regulate the response of non-relativistic and
widely used Einstein-Boltzmann solver CAMB [5, 6] that relativistic matter to gravity, respectively: in fact, in this
enables tests of modified gravity (MG) and dark energy formalism, photons experience an effective gravitational
(DE) models. It was first introduced in 2008 [7] and up- constant Geff,rel (k, a) = G × Σ(k, a), whereas matter is
graded in 2011 [8], 2019 [9] and 2023 [10]. MGCAMB com- subject to Geff,nonrel (k, a) = G × µ(k, a), where G is
putes predictions for the standard cosmological observ- Newton’s constant. Similarly to parameterizations of the
ables based on a phenomenological framework that allows equation of state, which encode deviations of the expan-
for modified relations between the gravitational poten- sion history from a pure ΛCDM model, µ and Σ charac-
tials and the matter density perturbation, and, like CAMB, terise linear deviations from standard growth. They are
can be used with Monte-Carlo Markov Chain (MCMC) particularly useful in that µ ̸= Σ is a key prediction of
samplers CosmoMC [11, 12] and Cobaya [11, 13, 14]. several modified gravity models [22]. To make the com-
MGCAMB can be used to explore a wide range of scalar- parisons with previously published constraints simple, we
tensor theories of gravity under the quasi-static approx- will focus on the DES parameterization [21]:
imation (QSA), both in terms of generic parameteriza- ΩDE (a)
tions [7] and specific models such as the Hu-Sawicki [15] µ = 1 + µ0 (3)
ΩDE,0
f (R), the symmetron [16] and the dilaton [17–19] models,
ΩDE (a)
as well as the purely phenomenological parameterizations Σ = 1 + Σ0 , (4)
of departures from GR introduced by the Planck [20] and ΩDE,0
Dark Energy Survey (DES) [21] collaborations. where µ0 and Σ0 are the present values of µ and Σ, re-
The MGCAMB framework is based on linear perturbation spectively, and ΩDE (a) = ρDE /ρtot is the dark energy
theory, with the following modifications of the perturbed density.
2

In this paper, we follow the conventions of [23] and II. MODELLING THE EFFECT OF
express the perturbed Friedman-Lemaı̂tre-Robertson- NONLINEARITIES ON THE MATTER POWER
Walker (FLRW) metric in the Newtonian gauge: SPECTRUM

In this section, we introduce the HMR method and


the code used to implement it, ReACT1 . We refer the in-
ds2 = a2 (τ ) −(1 + 2Ψ)dτ 2 + (1 − 2Φ)dxi dxi .
 
(5) terested reader to Refs. [33–35] for further details.

A. The halo model reaction method

The µ − Σ parameterization was extensively used in To extract the information contained in the nonlinear
tests of gravity on cosmological scales (see e.g. Refs. [24, scales, and use it to constrain models beyond ΛCDM,
25]). Albeit powerful, this formalism can only be used we need to modify the standard halo model. We can
within the range of validity of linear perturbation the- introduce a reaction function R(k, z), defined as [33]:
ory. On the other hand, a significant portion of the in-
formation contained in the data from large scale struc- PNL (k, z)
R(k, z) ≡ , (6)
ture (LSS) surveys is in the correlations over scales ≲ 10 PLpseudo (k, z) + P1h
pseudo
(k, z)
Mpc, where growth is highly nonlinear [26–29]. While in
the Λ Cold Dark Matter (ΛCDM) model the nonlinear Here, PNL (k, z) is the nonlinear power spectrum for the
corrections can be accurately predicted using the Halofit target cosmology, whereas the denominator contains the
model [30, 31] (with the latest version of CAMB adopting nonlinear ‘pseudo’ spectrum, which is defined as the
HMcode 2020 [32]), this approach cannot be used for MG ΛCDM power spectrum whose linear clustering matches
models. Because of this limitation, it was previously nec- the modified cosmology at the target redshift z, i.e.:
essary to apply cuts to the data where they could not be
reliably modelled by linear theory [9]. In this study, we PLpseudo (k, z) = PLMG (k, z) . (7)
overcome this limitation by equipping MGCAMB with the In Eq. (6) we have implicitly assumed the halo model
capability to compute observables at the nonlinear scales. to construct the pseudo power spectrum, where we have
separated the 2-halo, or linear, term in the power spec-
Refs. [33–35] introduced an approach, based on the trum PLpseudo (k, z), and the 1-halo term, P1h pseudo
(k, z).
halo model, to characterise the effects of nonlinearities The reaction function represents all the corrections to the
in modified gravity theories: the halo model reaction pseudo power spectrum coming from nonlinear beyond-
(HMR). In the same work, that formalism was also used ΛCDM physics.
to test the DGP and f (R) theories at percent-level ac- Based on the approach in Refs. [34, 35, 42], the reaction
curacy. The HMR approach was also compared to N - function can be computed as:
body simulations in phenomenological extensions of GR
in [36, 37], showing good agreement. Recently, the HMR R(k) =
was applied to the forecasts of cosmic shear for Stage-IV 2 (cb) (cbv) (ν)
surveys, [38] which adopted the linear parametrization of (1 − fν ) PHM (k) + 2fν (1 − fν ) PHM (k) + fν2 PL (k)
pseudo
,
MG proposed in [39–41]. PHM (k)
(8)
However, until now, the HMR approach has not been
applied to cosmological tests of gravity using the µ-Σ where ‘HM’ represents the halo model, ‘cb’ denotes cold
parametrization used, for example, in the DES analy- dark matter and baryons, ‘ν’ denotes massive neutrinos,
sis [21]. In this work, we extend MGCAMB to work with fν = Ων,0 /Ωm,0 is the energy density fraction of mas-
the HMR at nonlinear scales in this parameterization. sive neutrinos to the total matter density today, and we
As we will show, this addition results in stronger con- have omitted the z dependence for compactness of no-
straints on µ and Σ, even after marginalizing over the tation. In Eq. (8), the total matter power spectrum is
additional HMR parameters. Although we focus on the constructed from the weighted sum of individual compo-
(cbν)
DES parametrization of µ and Σ in this paper, the non- nents in the presence of massive neutrinos [43]. PHM (k)
linear extension works for all the models of µ and Σ im- stands for the cross power spectrum between neutrinos
plemented in MGCAMB. and the other two matter components, defined as
q
(cbν) (cb) (ν)
In what follows, we introduce the HMR method in PHM (k) ≈ PHM (k)PL (k) , (9)
Sec. II. Sec. III describes the implementation of the HMR
in MGCAMB, whereas we present the cosmological likeli-
hoods used in our demonstration in Sec. IV. We conclude
1 https://github.com/nebblu/ACTio-ReACTio
with a summary in Sec. V.
3

(cb)
where PHM is given by [35, 44] function F(k, a) in the second equation captures the fully
h i nonlinear modification to the Poisson equation.
(cb) (cb) (cb)
PHM (k) = (1 − E)e−k/k⋆ + E PL (k) + P1h (k) . In [35], two different approaches were proposed to de-
rive quantitative prediction for δNL , and the relation
(10) between the function µ and the nonlinear modification
In these equations, we have defined F(k, a): the parameterized post-Friedmannian frame-
(1−f )2 P
(cb)
(k,z)
work [45] and a simpler phenomenological parameteri-
ν
E(z) = limk→0 P pseudo 1h
(k,z)
, (11) zation that we will adopt below. The latter approach
1h
  −1 has the benefit of being simple, having only a few free
k⋆ (z) = −k̄ ln A1 (k̄,z)±A2 (k̄,z)  , (12) parameters, while still being able to reproduce the non-
2 (cb)
(1−fν ) PL (k̄,z) 1−E(z) linear effects in representative modified gravity theories.
In the phenomenological parameterization, the func-
with A1 (k, z) and A2 (k, z) expressed as tion F(k, a) is taken to be the error function (Erf), which
was shown to reproduce well the general profile of the
(ν) pseudo
A1 (k, z) =fν2 PL (k, z) + PHM (k, z)RSPT (k, z) effective gravitational constant in a variety of modified
2
h
(cb) (cb)
i gravity theories [35]. It also allows for a smooth tran-
− (1 − fν ) E(z)PL (k, z) + P1h (k, z) , sition from the unscreened to the screened regime. The
(13) specific form is taken to be
q h i
¯
pseudo
A2 (k, z) =2 fν2 PHM
(ν)
(k, z)PL (k, z)RSPT (k, z) , FErf = Erf ayh 10J × (1 − µ(k̂, a)), (17)
(14)
where
respectively, and the default scale where we calculate k⋆ 10p4
k̂ = , (18)
is set to k̄ = 0.06 h Mpc−1 . a2 yh Rth
RSPT (k, z) is the 1-loop standard perturbation theory J¯ = p1 − p2 log (Rth ) + p3 log (ayenv ) , (19)
(SPT) prediction for the reaction function, which charac-
and where yh = (Rth /a)/(Ri /ai ), Rth being the comoving
terizes the quasi-nonlinear scales well, and is used to cali-
halo top-hat radius and the subscript ‘i’ indicating the
brate the halo model reaction to better predict this tran-
initial time. In Eq. (19), yenv is the normalised radius of
sitory regime in the presence of modifications to grav-
the environment. Note that, if µ ∼ 1, as in the GR limit,
ity. Typically SPT has a good accuracy in the range
then 1 + FErf ∼ 1, and the nonlinear correction to the
0.01 < k Mpc h−1 < 0.1, whereas the nonlinear regime
modified perturbations also disappears.
is typically considered to be given by k > 0.1 h Mpc−1
There are four additional parameters in Eqs. 17-19,
for late time structure formation. This latter regime can
(cb) characterising the nonlinear regime, p1 - p4 , represent-
be more accurately predicted by P1h (k). ing MG phenomenological effects. The first one, p1 , pa-
rameterizes the strength of screening, p2 and p3 quantify
the mass and environmental dependencies, respectively,
B. The nonlinear phenomenological whereas p4 sets the scale of Yukawa suppression [46]. The
parameterization
QNL correction, described by S(k, a), was shown in [35]
to have a negligible effect for scale-independent MG the-
Since we need to compute the matter power spectrum, ories, while it can have up to a 2% effect on the power
a natural starting point is the Poisson equation, describ- spectrum for the Hu-Sawicki f (R) model. We will ignore
ing the relation between gravitational metric potentials the QNL correction in the current implementation.
and matter density contrast. In linear perturbation the-
ory, which is the default framework of MGCAMB, we have
the modified equations given by Eqs. (1) and (2). These III. THE NONLINEAR EXTENSION AND
can be extended to describe perturbations on smaller cos- OTHER UPGRADES OF MGCAMB
mological scales, which can be generally separated into
quasi-nonlinear scales and fully nonlinear scales as [35]: In this Section, we describe how we interfaced
ReACT [47], a publicly available code for computing the
k 2 ΨQNL = −4πGµ(a, k)a2 ρm δQNL (k, a) + S(k, a) , reaction function, within MGCAMB. We also describe the
(15) other upgrades implemented in the latest version of
k 2 ΨNL (k, a) = −4πG[1 + F(k, a)]a2 ρm δNL (k, a), (16) MGCAMB and MGCobaya [10, 48].

where the labels QNL and NL denote ‘quasi-nonlinear’


and ‘nonlinear’, respectively. Here, S(k, a) is a source A. MGCAMB with ReACT
term containing the 2nd-order and 3rd-order perturba-
tive contributions characterizing the modifications to the Since both ReACT and MGCAMB come with Python wrap-
Poisson equation on quasi-nonlinear scales. Instead, the pers, making them work together amounts to establish-
4

ing an appropriate Python interface. As mentioned in


Sec. II B, the nonlinear modification function F(k, a) de-
pends on the linear modification µ(k, a), which is natively p1=-3 p1=0 p1=2
0.10 p1=-2 p1=1 p1=3
computed within MGCAMB for several models. Therefore, p1=-1
the same models need to be implemented in the functions
that compute µ within ReACT. 0.05

(p1 = 0.5)(k) - 1
To achieve this, we add a new “case”, labeled 14, to
the 13 cases available within ReACT [47]. Then, we use
a wrapper function to pass the parameters of the µ(k, a) 0.00

PNL(k)/PNL
functions from MGCAMB to ReACT. Since ReACT can also
output the modified linear power spectrum, we check
that the linear predictions of MGCAMB and ReACT agree, 0.05
so as to validate our implementation. As a representa-
tive case, we compare the linear power spectra for the
0.10
DES parameterization with µ0 = 0.4, Σ0 = 0.1, finding
0.1% agreement between the two codes. 10 3 10 2 10 1 100 101
We also implement the dynamical dark energy op-
k [h Mpc 1]
tions, including the cubic-spline parameterization of the
dark energy density [10, 24, 49], to make it possible to FIG. 1. The effect of varying the reaction parameter p1 on
use ReACT with the full range of models implemented in the nonlinear matter power spectrum PNL (k). As reference,
we use p1 = 0.5 with the DES parametrization and µ0 = 0.4
MGCAMB.
and Σ0 = 0.1.
In MGCAMB, we use the function call
compute reaction nu ext, which is an option in
the Python interface of ReACT, to compute the reaction ear power spectrum at z = 0, for µ0 = 0.4 and Σ0 = 0.1,
function with the contribution of massive neutrinos. As as shown by Fig. 1. The limit p1 → ∞ corresponds to
input, this function requires three linear power spectra: no screening, whereas the limit p1 → −∞ corresponds to
total cb cb
PMG (k), PMG (k) and PΛCDM (k), under the assumption strong screening, where the power spectrum goes back to
of a ΛCDM background. They can be easily computed the ΛCDM prediction.
by MGCAMB. Separately, we add a new function, named The current version of ReACT only makes predictions
get react function, into the MGCAMB nonlinear Python up to redshift z = 2.5, which is sufficient to account
module, to return the result of the reaction function for nonlinear modifications at late times. On the other
and get MGCAMB to compute the pseudo power spectrum hand, the redshift bins in the DES likelihood extend to
pseudo
PHM (k, z) in Eq. (8). We then adopt the commonly- z = 3.5. To maintain a relatively smooth transition from
used interpolator function in CAMB and MGCAMB to get low to high redshifts, we use P pseudo (k, z) at z > 2.5,
the fully nonlinear power spectrum PNL (k, z). without losing accuracy, given that most of the modified
ReACT only provides nonlinear corrections for the mat- nonlinear corrections take place on redshifts below z =
ter power spectrum, not the Weyl potential W = (Φ + 2.5. Fig. 2 shows we do have a smooth transition in the
Ψ)/2. Thus, PW W (k) and PW δ (k), needed for interpret- output power spectrum from low to high redshifts, which
ing data from weak lensing surveys, must be calculated we display for several representative k modes.
separately. We assume that the relation between W and
δ is the same as the one on linear scales and is given by
Eq. (2). With this assumption, we can compute PW W (k) B. Other improvements
and PW δ (k) from the matter power spectrum Pδδ (k).
In addition to the parameters of the functions µ and
Σ, we have four parameters, p1 - p4 , specifying the non- In addition to the nonlinear extension, we added a few
linear correction. Although we interface all of them other features to MGCAMB and MGCobaya, to aid cosmolog-
within MGCAMB, in what follows, we take p1 as the only ical tests of gravity.
parameter for the nonlinear regime, for simplicity, set-
ting p2 = p3 = p4 = 0 as in Ref. [38]. This is justified
when using the DES parameterization (where µ only de- 1. Galaxy-Weyl correlation in the DES Year-1 likelihood
pends on time), since for scale-independent MG models
the screening has no environmental or halo mass depen- The DES Year-1 likelihood, as implemented in Cobaya,
dence, meaning it also has no dependence on p2 or p3 . In computes the galaxy-Weyl correlation from the galaxy-
models with no scale dependence, such as those the DES galaxy correlation, using the standard equations of GR
parameterization belongs to, there is also no Yukawa sup- relating δ and W . In order to use the likelihood to
pression, and hence no dependence on p4 . constrain modifications of Einstein’s equations, includ-
We have observed that varying p1 in the DES parame- ing the possibility of Σ ̸= 1, we have modified the DES
terization results in up to ∼ 10% difference in the nonlin- Year-1 likelihood implementation in MGCobaya to com-
5

separation θ are computed as:


X 2ℓ + 1
wi (θ) = Pℓ (cos θ)Cδiiobs δobs (ℓ) ,
104 ℓ

X 2ℓ + 1 P 2 (cos θ)
PNL(z) [h 3Mpc3]

γtij (θ) = ℓ
Cδijobs E (ℓ) ,
103 ℓ
4π ℓ(ℓ + 1)
(20)
X 2ℓ + 1 2(G+ −
ij ℓ,2 (cos θ) ± Gℓ,2 (cos θ))
102 ξ± (θ) =
4π ℓ2 (ℓ + 1)2
ℓ⩾2
k = 0.01 hMpc 1
ij ij
k = 0.1 hMpc 1 × [CEE (ℓ) ± CBB (ℓ)] .
101 k = 0.3 hMpc 1
k = 0.5 hMpc 1
k = 1.0 hMpc 1
k = 2.0 hMpc 1 Here, i, j denote two different redshift slices, wi is the
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 two-point function between lens galaxy positions in red-
z shift bin i, γtij is the two-point function between lens
galaxy positions and source galaxy tangential shear in
ij
FIG. 2. The modified matter power spectrum P (k, z) as a redshift bins i and j, and ξ± is the correlation be-
function of redshift, for several representative values of k and tween source galaxy shears in redshift bins i and j. E
the DES parametrization, with µ0 = 0.4, Σ0 = 0.1, and p1 = and B stand for the electric and magnetic parts of the
0.5. Note that we have a smooth transition around z = 2.5, shear field, respectively and δobs is the observed projected
which is the redshift at which ReACT corrections are activated. galaxy density contrast. Furthermore, Pℓ is the Legen-
dre polynomials of order ℓ, Pℓm is the associated Legendre
+/−
polynomial, and functions Gℓ,m (x) are combinations of
the associated Legendre polynomials Pℓm (x) and Pℓ−1 m
(x),
given explicitly in Eq. (4.19) of [53].
pute PW δ (k) within MGCAMB, together with PW W (k) and
Pδδ (k). The angular power spectra C(ℓ) in Eqs. (20) receive
contributions from the galaxy density perturbation (δg ),
gravitational shear (κ), intrinsic alignments (I), cos-
mic magnification (mag) and redshift space distortions
(RSD). Specifically,

ij ij ji
CEE ij
(ℓ) =Cκκ (ℓ) + CκI E
(ℓ) + CκI E
(ℓ) + CIijE IE (ℓ) ,
ij
CBB (ℓ) =CIijB IB ,
2. Implementation of the DES Year-3 likelihood in Cobaya
Cδijobs E (ℓ) =Cδijg κ (ℓ) + Cδijg IE (ℓ) + Cδijmag κ (ℓ) + Cδijmag IE (ℓ) ,
Cδiiobs δobs (ℓ) =Cδiig δg (ℓ) + Cδiimag δmag (ℓ) + CδiiRSD δRSD (ℓ)
We implement the DES Year-3 likelihood in Cobaya + 2Cδiig δmag (ℓ) + 2Cδiig δRSD (ℓ) + 2CδiiRSD δmag (ℓ) .
by following the formulation for calculating the weak (21)
lensing observables presented in [50]. For the lens sam-
ple, we also follow the treatment in [50]. Specifically,
where IE and IB are the electric and magnetic parts of
we choose the MagLim sample [51], which contains six
the intrinsic alignment, respectively.
tomographic redshift slices with nominal edges at z =
[0.20, 0.40, 0.55, 0.70, 0.85, 0.95, 1.05], as the lens galaxy The exact expression for the angular clustering power
sample. We remove the two highest redshift MagLim spectrum between two galaxy fields A, B is
bins from our analysis, as [52] revealed issues with the Z Z
ij 2
sample at z > 0.85. CAB (ℓ) = dχ1 WAi (χ1 ) dχ2 WBj (χ2 )
π
Z
Regarding the intrinsic alignments (IA) model used in dk 3
k PAB (k, χ1 , χ2 )jℓ (kχ1 )jℓ (kχ2 ) , (22)
the likelihood, we use the nonlinear alignment (NLA) k
model, which is consistent with the DES Collaboration’s
study of modified gravity constraints from their Year-3 with PAB being the corresponding three-dimensional
ij
data [50], which is not the same as the model (tidal align- power spectrum, and the kernels WA,B contain the rele-
ment and tidal torquing, TATT) used in their constraints vant contributions mentioned above.
on the ΛCDM model [52]. For the level of sensitivity of the DES measurements
on the shear-shear (CEE , CBB ) and galaxy-shear (Cδobs E )
The two-point angular correlation functions for the spectra, one can evaluate them efficiently using the Lim-
6

ber approximation [54], namely, and without the nonlinear extension. In the latter case,
we only vary the linear parameters µ0 and Σ0 , whereas
WAi (χ)WBj (χ) ℓ + 12
Z  
ij we include p1 when the nonlinear extension is applied.
CAB (ℓ) = dχ P AB k = , z(χ) , In the analyses we run, we employ the priors reported in
χ2 χ
(23) Tab. I.
As mentioned earlier, the DES 3×2pt data includes in-
for which PW δ , PW W (k) and Pδδ (k) are computed within formation from the nonlinear scales. In their Year-1 con-
MGCAMB, as in the case of the DES Year-1 likelihood de- straints on ΛCDM [21], the DES collaboration applied a
scribed in Sec. III B 1. conservative cut to keep only scales that could be reliably
However, the galaxy-galaxy spectrum (Cδg δg ) is mea- modelled with Halofit. We will refer to the DES 3×2pt
sured better, and the Limber approximation fails to pro- data with this conservative cut as “DES baseline”. Con-
vide the theory prediction with the required precision. versely, in earlier MGCAMB studies that used DES data, a
Therefore we evaluate Cδg δg using the exact formula in more stringent cut was applied [9] (denoted as “aggres-
Eq. (22). To be efficient, we follow the method de- sive cut” in the following) that eliminated all the nonlin-
scribed in Ref. [55] to calculate the double-Bessel integral ear scales – as MGCAMB was unable to model them prior
in Eq. (22) with the FFTLog algorithm. to this work. Finally, when using DES Year-3 data, we
will also consider a third type of cut: that used by the
DES collaboration in the DES Year-3 constraints [50] on
3. Further improvements and synchronization with CAMB modified gravity models, which also aims to restrict data
to the linear scales.
We have created wrapper functions for µ, γ, and Σ To test the nonlinear extension to MGCAMB, we first
in the Python interface of MGCAMB, making it easy to test check that it does not impact constraints derived from
the time evolution of these phenomenological functions at the linear scales alone. To do so, we analyse the DES-
a given Fourier number k for all implemented modified like data with and without the nonlinear extension, while
gravity models. applying the aggressive cut. We show the results of this
The current version of MGCAMB2 is now forked with test in Fig. 3 and the first two columns of Tab. II: the
CAMB3 on GitHub, making it convenient to keep consis- recovered constraints on the parameters µ0 and Σ0 are
tent with future upgrades of CAMB. consistent with each other and with the fiducial model
in both cases, demonstrating that indeed the nonlinear
corrections does not impact the linear scales in unwanted
IV. DEMONSTRATIONS ways.
As a second test, we assess the improvement in the
To demonstrate the utility of the nonlinear extension constraints when we include the nonlinear scales. For
of MGCAMB, we consider the DES parametrization, to make this, we analyse the baseline DES-like synthetic data with
it easier to compare against previous results in the liter- the nonlinear extension. We separately analyse the same
ature. data with the nonlinear extension excluded, this time
First, we perform a test on mock data, generat- without varying p1 . The joint constraints on µ0 , Σ0 and
ing synthetic DES-like galaxy-galaxy, galaxy-lensing and p1 are shown in Fig. 4. It is clear that we obtain better
lensing-lensing (3×2pt) correlations for a fiducial model constraints on µ0 and Σ0 when the nonlinear extension
with µ0 = 0.4, Σ0 = 0.1 and p1 = 0.5, and assuming a is used, even after marginalizing over p1 . We report the
ΛCDM background expansion. The values of the other parameter values we recover in Tab. II, showing that the
(cosmological) parameters are reported in Tab. I, whereas 1σ uncertainty on µ0 is reduced by a factor of ∼ 2 when
the DES likelihood parameters are fixed at the standard the nonlinear extension is used, which agrees with Fig.3.
values of DES Year-1 [21, 56], to help us isolate the im- of Ref. [38].
pact of including the nonlinear extension. The mock data Having tested our setup, we finally apply the ex-
are generated using MGCAMB. As our “data” covariance, we tended MGCAMB to current cosmological datasets, deriv-
adopt the covariance matrix from the DES Year-1 like- ing constraints on µ0 and Σ0 . For this, we use the
lihood, rescaled down by a factor of 25, as the actual DES Year-3 3×2pt data [50], Planck 2018 CMB temper-
DES Year-1 covariance would not allow us to constrain ature, polarization and lensing power spectra [57], joint
the nonlinear parameter p1 . We then run MCMC chains measurements of baryon acoustic oscillations (BAO) and
using MGCobaya, to see if we can recover the input model. redshift-space distortions (RSD) from eBOSS DR16, the
To test our implementation, and to assess the added con- SDSS DR7 MGS data [58], the BAO measurement from
straints brought in by the nonlinear scales, we run with 6dF [59], and the Pantheon sample of uncalibrated super-
novae [60]. To assess, again, the impact of including non-
linear scales, we perform two analyses, with and without
the nonlinear extension, applying the DES Year-3 linear
2 https://github.com/sfu-cosmo/MGCAMB cut in the latter. We show the recovered posteriors in
3 https://github.com/cmbant/CAMB Fig. 5: it is clear that we obtain stronger constraints on
7

Parameters Flat Prior Fiducial


µ0 (-0.2,1.2) 0.4
Σ0 (-0.2,0.4) 0.1
p1 (-2,2) 0.5
Ω b h2 - 0.112
Ωc h2 - 0.0226
ns - 0.969
ln(1010 As ) - 3.06
100 θMC - 1.0410
τ - 0.067
Σ mν - 0.06

TABLE I. Fiducial values and priors on the MG and cosmological parameters used in tests on synthetic DES-like data in
Sec. IV. Parameters other than µ0 , Σ0 and p1 are fixed to their fiducial values, to allow us to isolate the impact of including
the nonlinear extension to MGCAMB.

no NL ext + aggressive cut NL ext + aggressive cut NL ext + DES baseline


µ0 0.392 ± 0.097 0.393 ± 0.098 0.393 ± 0.046
Σ0 0.110 ± 0.015 0.102 ± 0.016 0.100 ± 0.012

TABLE II. Tests on the nonlinear extension (NL ext). Parameter inferences on µ0 and Σ0 obtained from synthetic data without
the nonlinear extension and using the aggressive linear cut (first column), with the nonlinear extension while still using the
aggressive linear cut (second column), and with the nonlinear extension and the DES baseline data, which includes nonlinear
scales (third column). Results show that adding the nonlinear extension does not impact constraints derived from the linear
scales alone, as expected (also see Fig. 3). Note that constraints on µ0 are much tighter when nonlinear data is included (also
see Fig. 4).

NL ext + aggressive cut NL ext + DES baseline


no NL ext + aggressive cut no NL ext + aggressive cut

0.14
0.12
0

0.10
0.15 0.08
0.06

0.10
0

1
0
p1

0.05 1
0.2 0.4 0.6 0.06 0.10 0.14 0.2 0.4 0.6 0.06 0.10 0.14 1 0 1
0 0 0 0 p1
FIG. 3. Test on the MGCAMB extension. In this plot, we show FIG. 4. Comparison of the constraints on {µ0 , Σ0 } obtained
the joint 68% and 95% CL constraints on µ0 and Σ0 obtained with and without the nonlinear extension. In the latter case,
from synthetic data with and without the nonlinear exten- the aggressive cut is applied. The constraints are derived
sion (NL ext), while applying the aggressive linear cut in both from synthetic data generated from the fiducial parameters
cases. Results show that the nonlinear extension does not im- reported in Tab. I. Constraints are tighter when the nonlinear
pact constraints derived from linear scales alone, as expected. extension and the baseline data are used.

µ0 and Σ0 with the nonlinear extension. In fact, whilst on the Σ0 constraint is apparent. We can understand this
the posterior on µ0 is largely unaffected, the improvement by noting that DES is primarily a weak lensing survey,
8

DES Y3 baseline + PBRS DES Y3 with linear cut + PBRS


µ0 0.003 ± 0.201 -0.012 ± 0.216
Σ0 0.016 ± 0.039 0.043 ± 0.045

TABLE III. Parameter inferences on µ0 and Σ0 (68% CL) from DES Year-3 baseline data, with and without the nonlinear
extension (also see Fig. 5). When excluding the nonlinear extension, we use the same linear cut as the DES collaboration [50].
PBRS denotes the combination of Planck 2018, BAO, RSD, and SN Ia data, as detailed in Sec. IV. Note the improvement on
the constraints when the nonlinear extension is included.

DES Y3 baseline + PBRS parameter phenomenological model, recovering the input


DES Y3 with linear cut + PBRS fiducial parameters µ0 and Σ0 . Further tests showed that
constraints on µ0 and Σ0 improve when the new exten-
sion is included.
Having tested our method, we then used it to obtain
constraints on µ0 and Σ0 from a combination of current
0.2 datasets that included the DES Year-3 data, and found
an enhancement in the constraints compared to results
0.1 obtained from the linear scales only.
0

This nonlinear extension of MGCAMB will enable us to


0.0 use a wider range of data from current and forthcoming
cosmological surveys, including weak lensing surveys such
0.1
as the Vera C. Rubin Observatory4 and Euclid5 .
1
0
p1

0.5 0.0 0.5 0.00 0.15 1 0 1


0 0 p1
FIG. 5. Constraints on µ0 and Σ0 obtained from the baseline
DES Year-3 data, with and without the nonlinear extension. ACKNOWLEDGMENTS
In the latter case, we apply the same linear cut as the DES
collaboration [50]. PBRS denotes the combination of Planck
2018, BAO, RSD, and SN Ia as detailed in Sec. IV. Con- The work of ZW and LP is supported by the National
straints on Σ0 improve when the nonlinear extension is used. Sciences and Engineering Research Council (NSERC) of
Canada. LY and GBZ are supported by the National
Key Basic Research and Development Program of China
which are especially constraining for Σ. Our parameter (No. 2018YFA0404503), NSFC (11925303, 11890691),
estimates are provided in Tab. III. The constraints we ob- and science research grants from the China Manned
tain with the DES Year-3 linear cut are consistent with Space Project with No. CMS-CSST-2021-B01. DS and
those reported by the DES collaboration [50]. KK are supported by STFC grant ST/W001225/1. BB
is supported by a UKRI Stephen Hawking Fellowship
(EP/W005654/2). For the purpose of open access, we
V. SUMMARY have applied a Creative Commons Attribution (CC BY)
licence to any Author Accepted Manuscript version aris-
We have extended MGCAMB to predict cosmological ob- ing. Supporting research data are available on reasonable
servables in modified gravity at the nonlinear scales. request from the corresponding author.
The extension employed the halo model reaction method,
with the reaction function computed using ReACT and in-
terfaced in the Python wrapper of MGCAMB.
We have tested the nonlinear extension on synthetic
data generated using the DES parameterization, a two-

4 https://www.lsst.org/
5
Euclid
https://www.esa.int/Science_Exploration/Space_Science/
9

[1] S. Perlmutter et al. (Supernova Cosmology Project), [29] A. Spurio Mancini and B. Bose, (2023), 10.21105/as-
Astrophys. J. 517, 565 (1999), arXiv:astro-ph/9812133 tro.2305.06350, arXiv:2305.06350 [astro-ph.CO].
[astro-ph]. [30] R. E. Smith, J. A. Peacock, A. Jenkins, S. D. M. White,
[2] A. G. Riess et al. (Supernova Search Team), Astron. J. C. S. Frenk, F. R. Pearce, P. A. Thomas, G. Efstathiou,
116, 1009 (1998), arXiv:astro-ph/9805201 [astro-ph]. and H. M. P. Couchmann (VIRGO Consortium), Mon.
[3] S. Weinberg, Rev. Mod. Phys. 61, 1 (1989). Not. Roy. Astron. Soc. 341, 1311 (2003), arXiv:astro-
[4] C. P. Burgess, in 100e Ecole d’Ete de Physique: Post- ph/0207664.
Planck Cosmology Les Houches, France, July 8-August [31] R. Takahashi, M. Sato, T. Nishimichi, A. Taruya, and
2, 2013 (2015) pp. 149–197, arXiv:1309.4133 [hep-th]. M. Oguri, Astrophys. J. 761, 152 (2012), arXiv:1208.2701
[5] A. Lewis, A. Challinor, and A. Lasenby, Astrophys. J. [astro-ph.CO].
538, 473 (2000), astro-ph/9911177. [32] A. Mead, S. Brieden, T. Tröster, and C. Heymans,
[6] http://camb.info. (2020), 10.1093/mnras/stab082, arXiv:2009.01858 [astro-
[7] G.-B. Zhao, L. Pogosian, A. Silvestri, and J. Zylberberg, ph.CO].
Phys. Rev. D79, 083513 (2009), arXiv:0809.3791 [astro- [33] M. Cataneo, L. Lombriser, C. Heymans, A. Mead,
ph]. A. Barreira, S. Bose, and B. Li, Mon. Not. Roy. Astron.
[8] A. Hojjati, L. Pogosian, and G.-B. Zhao, JCAP 1108, Soc. 488, 2121 (2019), arXiv:1812.05594 [astro-ph.CO].
005 (2011), arXiv:1106.4543 [astro-ph.CO]. [34] B. Bose, M. Cataneo, T. Tröster, Q. Xia, C. Heymans,
[9] A. Zucca, L. Pogosian, A. Silvestri, and G.-B. Zhao, and L. Lombriser, Mon. Not. Roy. Astron. Soc. 498, 4650
JCAP 05, 001 (2019), arXiv:1901.05956 [astro-ph.CO]. (2020), arXiv:2005.12184 [astro-ph.CO].
[10] Z. Wang, S. H. Mirpoorian, L. Pogosian, A. Silvestri, [35] B. Bose, M. Tsedrik, J. Kennedy, L. Lombriser, A. Pourt-
and G.-B. Zhao, JCAP 08, 038 (2023), arXiv:2305.05667 sidou, and A. Taylor, (2022), 10.1093/mnras/stac3783,
[astro-ph.CO]. arXiv:2210.01094 [astro-ph.CO].
[11] A. Lewis and S. Bridle, Phys. Rev. D66, 103511 (2002), [36] S. Srinivasan, D. B. Thomas, F. Pace, and R. Battye,
astro-ph/0205436. JCAP 06, 016 (2021), arXiv:2103.05051 [astro-ph.CO].
[12] http://cosmologist.info/cosmomc. [37] S. Srinivasan, D. B. Thomas, and R. Battye, JCAP 03,
[13] J. Torrado and A. Lewis, JCAP 05, 057 (2021), 039 (2024), arXiv:2306.17240 [astro-ph.CO].
arXiv:2005.05290 [astro-ph.IM]. [38] M. Tsedrik, B. Bose, P. Carrilho, A. Pourtsidou,
[14] https://cobaya.readthedocs.io. S. Pamuk, S. Casas, and J. Lesgourgues, (2024),
[15] W. Hu and I. Sawicki, Phys. Rev. D76, 064004 (2007), arXiv:2404.11508 [astro-ph.CO].
arXiv:0705.1158 [astro-ph]. [39] P. J. Peebles, The Large-Scale Structure of the Universe
[16] K. Hinterbichler and J. Khoury, Phys. Rev. Lett. 104, (Princeton University Press, 1980).
231301 (2010), arXiv:1001.4525 [hep-th]. [40] E. V. Linder, Phys. Rev. D72, 043529 (2005),
[17] T. Damour, G. W. Gibbons, and C. Gundlach, Phys. arXiv:astro-ph/0507263 [astro-ph].
Rev. Lett. 64, 123 (1990). [41] E. V. Linder and R. N. Cahn, Astropart. Phys. 28, 481
[18] T. Damour and A. M. Polyakov, Nucl. Phys. B423, 532 (2007), arXiv:astro-ph/0701317.
(1994), arXiv:hep-th/9401069 [hep-th]. [42] M. Cataneo, J. D. Emberson, D. Inman, J. Harnois-
[19] P. Brax, A.-C. Davis, B. Li, and H. A. Winther, Deraps, and C. Heymans, Mon. Not. Roy. Astron. Soc.
Phys. Rev. D86, 044015 (2012), arXiv:1203.4812 [astro- 491, 3101 (2020), arXiv:1909.02561 [astro-ph.CO].
ph.CO]. [43] S. Agarwal and H. A. Feldman, Monthly Notices of the
[20] P. A. R. Ade et al. (Planck), Astron. Astrophys. 594, Royal Astronomical Society , no (2010).
A14 (2016), arXiv:1502.01590 [astro-ph.CO]. [44] A. Cooray and R. K. Sheth, Phys. Rept. 372, 1 (2002),
[21] T. M. C. Abbott et al. (DES), Phys. Rev. D 99, 123505 arXiv:astro-ph/0206508.
(2019), arXiv:1810.02499 [astro-ph.CO]. [45] L. Lombriser, JCAP 11, 039 (2016), arXiv:1608.00522
[22] L. Pogosian and A. Silvestri, Phys. Rev. D94, 104014 [astro-ph.CO].
(2016), arXiv:1606.05339 [astro-ph.CO]. [46] H. Yukawa, Proc. Phys. Math. Soc. Jap. 17, 48 (1935).
[23] C.-P. Ma and E. Bertschinger, Astrophys. J. 455, 7 [47] https://github.com/nebblu/ACTio-ReACTio.
(1995), arXiv:astro-ph/9506072. [48] https://github.com/sfu-cosmo/MGCobaya.
[24] L. Pogosian, M. Raveri, K. Koyama, M. Martinelli, [49] M. Raveri, L. Pogosian, M. Martinelli, K. Koyama,
A. Silvestri, G.-B. Zhao, J. Li, S. Peirone, and A. Zucca, A. Silvestri, G.-B. Zhao, J. Li, S. Peirone, and A. Zucca,
Nature Astron. 6, 1484 (2022), arXiv:2107.12992 [astro- (2021), arXiv:2107.12990 [astro-ph.CO].
ph.CO]. [50] T. M. C. Abbott et al. (DES), Phys. Rev. D 107, 083504
[25] U. Andrade, A. J. S. Capistrano, E. Di Valentino, and (2023), arXiv:2207.05766 [astro-ph.CO].
R. C. Nunes, Monthly Notices of the Royal Astronomical [51] A. Porredon et al., Physical Review D 103 (2021),
Society 529, 831–838 (2024). 10.1103/physrevd.103.043503.
[26] v. Ivezić et al. (LSST), Astrophys. J. 873, 111 (2019), [52] T. M. C. Abbott et al., Physical Review D 105 (2022),
arXiv:0805.2366 [astro-ph]. 10.1103/physrevd.105.023520.
[27] A. Blanchard et al. (Euclid), Astron. Astrophys. 642, [53] A. Stebbins, “Weak lensing on the celestial sphere,”
A191 (2020), arXiv:1910.09273 [astro-ph.CO]. (1996), arXiv:astro-ph/9609149 [astro-ph].
[28] N. Frusciante et al. (Euclid), (2023), arXiv:2306.12368 [54] D. N. Limber, Astrophys. J. 117, 134 (1953).
[astro-ph.CO]. [55] X. Fang, E. Krause, T. Eifler, and N. MacCrann, Journal
of Cosmology and Astroparticle Physics 2020, 010–010
10

(2020). Soc. 449, 835 (2015), arXiv:1409.3242 [astro-ph.CO].


[56] E. Krause et al. (DES), (2017), arXiv:1706.09359 [astro- [59] F. Beutler, C. Blake, M. Colless, D. H. Jones, L. Staveley-
ph.CO]. Smith, L. Campbell, Q. Parker, W. Saunders, and
[57] N. Aghanim et al. (Planck), (2019), arXiv:1907.12875 F. Watson, Monthly Notices of the Royal Astronomical
[astro-ph.CO]. Society 416, 3017 (2011).
[58] A. J. Ross, L. Samushia, C. Howlett, W. J. Percival, [60] D. M. Scolnic et al. (Pan-STARRS1), Astrophys. J. 859,
A. Burden, and M. Manera, Mon. Not. Roy. Astron. 101 (2018), arXiv:1710.00845 [astro-ph.CO].

You might also like