PHYSICAL REVIEW D 103, 123507 (2021)

Dark radiation from inflationary fluctuations

*
Gordan Krnjaic
Theoretical Physics Department, Fermi National Accelerator Laboratory, Batavia, Illinois 60510, USA
and Kavli Institute for Cosmological Physics, University of Chicago, Chicago, Illinois 60637, USA

(Received 14 July 2020; accepted 14 May 2021; published 4 June 2021)

Light new vector bosons can be produced gravitationally through quantum fluctuations during inflation;
if these particles are feebly coupled and cosmologically metastable, they can account for the observed dark
matter abundance. However, in minimal anomaly-free Uð1Þ extensions to the Standard Model, these
vectors generically decay to neutrinos if at least one neutrino mass eigenstate is sufficiently light. If these
decays occur between neutrino decoupling and cosmic microwave background (CMB) freeze-out, the
resulting radiation energy density can contribute to ΔN eff at levels that can ameliorate the Hubble tension
and be discovered with future CMB and relic neutrino detection experiments. Since the additional neutrinos
are produced from vector decays after Big Bang Nucleosynthesis (BBN), this scenario predicts ΔN eff > 0
at recombination, but ΔN eff ¼ 0 during BBN. Furthermore, due to a fortuitous cancellation, the
contribution to ΔN eff is approximately mass independent.

DOI: 10.1103/PhysRevD.103.123507

I. INTRODUCTION when the Hubble scale equals the vector’s mass, H ¼ m

[15]. Such scales are typically much smaller than those
Cosmological inflation elegantly accounts for the
probed by CMB experiments, so the isocurvature bounds
observed flatness, isotropy, and homogeneity of the
on this scenario are negligible and this mechanism yields a
Universe. Additionally, the quantum mechanical fluctua-
viable dark matter candidate for
tions in the inflaton field during inflation generate a nearly
14 
scale invariant spectrum of density perturbations that seed 10 GeV 4
the growth of structure and imprint temperature anisotro- m ∼ μeV : ð1Þ
HI
pies onto the cosmic microwave background (CMB)—see
Ref. [1] for a review. Thus, if the vector is decoupled from the Standard Model
It is well known that new, feebly coupled particles are (SM) fields or is sufficiently light (m ≪ 2me ) and interacts
produced gravitationally through quantum fluctuations only through a small kinetic mixing, its cosmological
during inflation if their masses are small compared to metastability is generically realized.1
the inflationary Hubble scale HI [2]; heavier particles can However, if the vector is the gauge boson of a minimal
also be produced if the inflaton undergoes rapid oscillations Uð1Þ gauge extension, couplings to neutrinos are required
[3–8] or nontrivially affects the particle’s mass during for anomaly cancellation [19]; the only anomaly-free
inflation [9]. For light spin-0 particles, these fluctuations groups with no additional SM charged fermions are
yield isocurvature perturbations on large scales, which are
tightly constrained by CMB observations [10,11] and for Uð1ÞB−L ; Uð1ÞLi −Lj ; Uð1ÞB−3Li ; ð2Þ
spin-1=2 fermions, inflationary fluctuations are generically
suppressed unless they have nonconformal interactions where B=L is baryon/lepton number, i; j ¼ e, μ, τ are
through higher dimension operators [12–14]. lepton flavor indices, and the corresponding gauge bosons
It has recently been shown that the gravitational pro- in these models couple to at least one neutrino flavor. Thus,
duction of spin-1 particles during inflation is sharply unlike kinetically mixed dark photon scenarios, the vector
peaked at modes that reenter the horizon after inflation decays in these models can be relatively prompt and have
observable cosmological consequences.
In this paper, we consider the fate of light gauge bosons
*
[email protected] V produced during inflation. We assume these vectors
Published by the American Physical Society under the terms of
the Creative Commons Attribution 4.0 International license.
1
For a kinetically mixed V, allowed decays V → 3γ are highly
Further distribution of this work must maintain attribution to suppressed [16,17] and if the vector kinetically mixes with SM
the author(s) and the published article’s title, journal citation, hypercharge before electroweak symmetry breaking, decays to
and DOI. Funded by SCOAP3. V → ν̄ν are further suppressed by powers of ∼ðm=mZ Þ4 [18].

2470-0010=2021=103(12)=123507(6) 123507-1 Published by the American Physical Society

GORDAN KRNJAIC PHYS. REV. D 103, 123507 (2021)

constitutes a present day dark matter fraction f 0V ≡

ΩV =ΩDM [15]
pffiffiffiffi 2 rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2
mH I m HI
f 0V ≈ ≈ 10 −2
; ð4Þ
4π 2 M 3=2
Pl T eq
10 μeV 1013 GeV

where MPl ¼ 1.22 × 1019 GeV is the Planck mass and

T eq ¼ 0.75 eV is the temperature of matter-radiation equal-
ity, so the energy density at earlier times is


aðt0 Þ 3
ρV ðtÞ ¼ ρ0V ; ρ0V ≡ f 0V ΩDM ρcr ; ð5Þ
aðtÞ

where ΩDM ¼ 0.24 is the fractional dark matter abundance,

ρcr ¼ 4.1 × 10−47 GeV4 is the critical density, a is the
FRW scale factor, t0 ¼ 13.8 Gyr, and a 0 label represents a
FIG. 1. Time-dependent energy fractions ρi =ρtot for a bench- present day quantity [25,26]. For stable vectors, Eq. (5) is
mark choice of model inputs. Here ρtot ¼ 3M2Pl H2 =8π is the total valid for t > t⋆ ¼ ð2mÞ−1, the horizon reentry time corre-
energy density of the Universe and we show ρV , the density of sponding to H ¼ m and temperature
vectors from inflationary production, δρν the additional neutrino
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
1=4
1=2
density from V → ν̄ν decays assuming a single neutrino flavor. mMPl 100 m
From left to right, the vertical dashed lines mark neutrino T⋆ ¼ pffiffiffiffiffi ≈ 85 GeV ; ð6Þ
decoupling, matter-radiation equality (MRE), and CMB decou- 1.66 g⋆ g⋆ 10 μeV
pling. Note that in Eq. (18), the number density of the neutrino
population from V decays might exceed that of the relic neutrino where g⋆ is the effective number of relativistic SM species
background, but as seen here, the energy density remains small in equilibrium. Note that because the V power spectrum is
for empirically viable values of ΔN eff .
dominated by momentum modes that reenter the horizon
when H ∼ m, the V population is nonrelativistic for all
couple feebly to neutrinos and that at least one neutrino times t > t⋆ .
mass eigenstate is sufficiently light to allow V → ν̄ν
decays. If such decays occur after neutrino decoupling, III. ADDING DECAYS TO NEUTRINOS
but before CMB photon decoupling, there is an irreducible
Since Abelian gauge extensions to the SM generically
contribution to ΔN eff that is potentially observable with
feature neutrino couplings, we add the representative
future CMB-S4 experiments [20] and a modified relic
interaction
neutrino spectrum observable at PTOLEMY [21,22].
Furthermore, such a contribution of ΔN eff can alleviate
the discrepancy between early and late time measurements L ⊃ gV μ ν̄i γ μ νi ; ð7Þ
of the Hubble constant (for recent reviews see [23,24]).
to Eq. (3), where g ≪ 1 is a gauge coupling and i is a lepton
family index. In the massless neutrino limit, the partial
II. STABLE VECTOR ABUNDANCE
width to a single flavor is [27]
The general Lagrangian during inflation contains
g2 m
ΓðV → ν̄i νi Þ ¼ ; ð8Þ
L 1 m2 μν 24π
pffiffiffi ⊃ − gμκ gνλ Fμν Fκλ þ g V μV ν; ð3Þ
g̃ 4 2 the total width ΓV is the sum of all allowed channels and
τV ¼ Γ−1 V is the V lifetime. We note that a single massless
where V is a gauge boson in a Friedmann-Robertson- neutrino eigenstate is empirically viable [28,29], so, in
Walker (FRW) metric, Fμν is the corresponding field principle, at least one decay channel is allowed for all
strength tensor, and g̃ is the metric determinant. If the vector masses.
mass satisfies 0 < m ≪ HI and V is stable, the longitudinal Unlike in Ref [15], here the vector is unstable and
mode2 is gravitationally produced during inflation and V → ν̄ν decays deplete the initial population, so Eq. (5) is
only useful for establishing the initial condition for ρV at
2
The transverse mode is conformally coupled to gravity, so its t ¼ t⋆ . Accounting for decays to neutrinos, the V popula-
production is greatly suppressed by comparison [15]. tion can now be written

123507-2
DARK RADIATION FROM INFLATIONARY FLUCTUATIONS PHYS. REV. D 103, 123507 (2021)


aðt0 Þ 3 −ΓV ðt−t⋆ Þ
ρV ðtÞ ¼ ρ0V e ; ð9Þ
aðtÞ

and the energy density of the modified neutrino population

δρν evolves according to

δ_ρν þ 4Hδρν ¼ ΓV ρV ; ð10Þ

which can be integrated to yield

Z
Γ t
δρν ðtÞ ¼ V 4 dt0 aðt0 Þ4 ρV ðt0 Þ; ð11Þ
aðtÞ tν

where a is the FRW scale factor and tν ∼ 1 sec is the time

of neutrino decoupling; we only keep contributions for
t > tν because neutrinos injected before tν thermalize with
the radiation bath and do not contribute to dark radiation.
Similarly, V that decay after CMB decoupling will not FIG. 2. Parameter space that yields observable levels of dark
contribute to ΔN eff , but will increase the dark matter radiation from a population of gravitationally produced vectors that
density during recombination. In Fig. 1 we show a decay via → ν̄ν after neutrino decoupling but before recombination.
representative solution of Eq. (10) plotted as a fraction Horizontal blue shaded bands represent regions where 10−2 <
ΔN eff < 0.5 for representative choices of the inflationary Hubble
of the total energy density.
scale H I ; for each choice, the parameter space below the bottom
In terms of the equivalent number of SM neutrinos boundary predicts ΔN eff > 0.5, which is excluded assuming other-
ΔN eff , this additional radiation from δρν predicts wise standard cosmological assumptions [25,30,31]. Above the

  horizontal dotted lines, V thermalizes with the SM, yielding ΔN eff ≈
8 11 4=3 δρν  2.5 [30], which is excluded if V couples to e or μ. The vertical dotted
ΔN eff ≡ ; ð12Þ
7 4 ργ T CMB lines mark m ¼ 2me;μ where V → eþ e− and V → μþ μ− decays are
kinematically allowed. Most models in Eq. (2) feature V − e
couplings, so for m > 2me the ΔN eff ≈ 0 as V → eþ e− decays heat
where ργ ¼ π 2 T 4 =15 and the contribution is evaluated at photons to compensate for V → ν̄ν decays, which heat neutrinos.
the temperature of photon decoupling, T CMB ≈ 0.2 eV; this
sets the upper integration range in Eq. (11) since V decays In Fig. 2 we show ΔN eff predictions for the inflationary
after last scattering do not contribute to dark radiation in the vector population where we compute δρν numerically using
CMB dataset. Eq. (11). The blue horizontal bands represent the currently
For the full parameter space, ΔN eff in Eq. (12) must be viable 10−2 ≤ ΔN eff < 0.5 range that is within the reach of
computed numerically by solving Eq. (11). However, if V CMB-S4 predictions [20]. Note that current Big Bang
decays between T ν;dec and T eq , the decay temperature can Nucleosynthesis (BBN) bound ΔN eff < 0.5 [30] is less
be written stringent than the CMB and large scale structure bound
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ΔN eff < 0.28 [25], but the BBN limit is less model


1=2 dependent because it is not as sensitive to the choice of
g2 mMPl g m
T decay ≈ pffiffiffiffiffi ≈ 100 eV ; ð13Þ cosmological model. However, despite the nominal choice
40π g⋆ 10−8 10−5 eV of ΔN eff < 0.5 as our conservative exclusion benchmark,
this scenario is not directly constrained by the BBN
where g⋆ ≈ 3.36 in our temperature range of interest measurement of ΔN eff since the additional neutrinos from
between decoupling and recombination. Assuming instan- V decays do not appear until after BBN.
taneous V → ν̄ν decay and approximating δρν ≈ ρV ðT decay Þ The area in between the dashed diagonal bands represent
using Eq. (5), Eq. (12) becomes parameter space for which V → ν̄ν decays occur between

 pffiffiffiffi neutrino and CMB decoupling; decays outside this band do
30 11 4=3 ΩDM ρcr mH 2I not contribute to ΔN eff . The vertical lines at m ¼ 2me ; 2mμ
ΔN eff ≈ 4
7π 4 M 3=2 3 represent regions where the ΔN eff prediction here does not
Pl T 0 T eq T decay

2
−8  apply if V couples to electrons or muons; in such models, V
HI 10 decays to charged particles after neutrino decoupling will
≈ 10−2 ; ð14Þ
1014 GeV g heat photons and thereby reduce ΔN eff relative to Eq. (14).
We note for completeness that there is also a possible
where the vector mass has canceled. contribution to ΔN eff from the V population itself if an

123507-3
GORDAN KRNJAIC PHYS. REV. D 103, 123507 (2021)
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
pffiffiffiffiffi ffi 
appreciable fraction of the ρV redshifts like radiation at 4π g⋆ mf 5 × 10−10 ; f ¼ e
recombination. Since inflationary V production is sharply g≲ ¼ ; ð16Þ
peaked around modes that enter the horizon at H ∼ m, from αMPl 7 × 10−9 ; f ¼ μ
Eq. (6) only masses below m ≲ 10−30 eV will be quasir-
elativistic around T CMB . However, from Eq. (4) such small where g⋆ is evaluated at T ¼ me ; mμ , respectively.
masses yield negligible inflationary production for all The stronger electron-based bound here applies to
HI ≲ 1014 GeV allowed by CMB limits on tensor modes most anomaly-free Uð1Þ extensions—including
[25,32], so we can safely neglect this contribution. gauged B − L, B − 3Le , Le − Lμ , Le − Lτ —as they
all require V to couple to electrons for anomaly
cancellation [19]; the main outlier is gauged Lμ − Lτ
IV. INTERACTIONS WITH THE SM PLASMA for which muon-induced thermalization is the dom-
The above discussion assumes that the early Universe V inant process at low temperatures [27], so the bound
population arises entirely to inflationary production and is is somewhat weaker. Both of the requirements in
unaffected by the SM radiation bath. However, for any value Eq. (16) are presented as dotted horizontal black
of the gauge coupling, there is irreducible sub-Hubble curves in Fig. 2 and the parameter space above these
“freeze-in” production of additional V [27,33–35] and, if regions is excluded if the model in question features
the coupling is sufficiently large, the V population can the corresponding e or μ coupling.
thermalize with the SM plasma; which yields additional We emphasize that the parameter space shown in Fig. 2
contributions to ΔN eff . is extremely weakly coupled, such that there is no danger of
(i) Inverse decays: Independent of any other assump- the inflationary V population thermalizing with the SM
tions about ultralight V particles beyond their plasma or of any appreciable contributions from SM
coupling to neutrinos, there is a bound on thermal- processes that produce additional V particles via inverse
izing with the SM plasma via population via ν̄ν ↔ V decays or SM scattering reactions. For a careful study of
decays and inverse decays. If thermalization occurs such processes in the context of the models studied here,
before neutrino decoupling, this scenario predicts see [27], which identifies the parameter space where freeze-
ΔN eff ≈ 2.5, so avoiding this fate requires in production via inverse decays contributes to cosmologi-

 cal observables including ΔN eff .
Γν̄ν→V g2 m2 MPl −5 eV
∼ ≪ 1 ⇒ g ≲ 10 ; ð15Þ
H T 3ν;dec m
V. PRESENT DAY NEUTRINO FLUX
where T ν;dec ∼ MeV is the temperature of neutrino
decoupling via the SM weak interactions. If, instead, In this section we review the results of Ref. [21], adapted
thermalization occurs between T ν;dec and T CMB as in to the case of inflationary vector production. The neutrinos
Ref. [36], then ΔN eff ∼ 0.2 independent of mass and in our scenario arise from V decays and if the entire
coupling [27].3 Since this contribution is fixed only population decays in the early Universe, the present day
by the neutrino coupling, it must be added to the number density is
component from the inflationary population.
0 

(ii) Production from charged particles: If V also couples f 0V ΩDM ρcr −3 fV eV
δnν ðt0 Þ ¼ ≈ 130 cm : ð17Þ
to charged fermion f, dangerous f̄f → γV and mV 0.05 m
fγ → fV processes can thermalize V with the SM
radiation bath, thereby yielding ΔN eff ≈ 2.5, which If these decays occur between neutrino decoupling and
is excluded by both BBN and CMB observables recombination, Eq. (17) can be rewritten [21]
[25,27,30,31].4 The V production rate can be esti-


sffiffiffiffiffiffiffiffiffiffiffiffi
ffi
mated as Γf̄f→Vγ ∼ Γfγ→fV ∼ αg2 T=4π, so these ΔN eV 10 3
yr
δnν ðt0 Þ ≈ 103 cm−3 eff
: ð18Þ
processes grow relative to Hubble until T ∼ mf , 0.28 m τV
when they become Boltzmann suppressed. Ensuring
that the maximum rate not exceed Hubble expansion Although the number density of additional neutrinos in
requires Eq. (18) can exceed the ∼300 cm−3 number density of the
CνB as predicted in the Standard Model, as long as the
3
Although Ref. [27] specifically considered the gauged corresponding value of ΔN eff satisfies observational
Lμ − Lτ scenario, this conclusion holds for any ultralight vector bounds, the energy density of this population is always
m ≪ me with a coupling to neutrinos, which includes all subdominant and remains empirically viable.
anomaly-free Uð1Þ extensions that gauge global SM quantum For some parameter choices, this additional neutrino
numbers [19].
4
This ΔN eff ≈ 2.5 prediction assumes that the thermalized V population may be observable with the PTOLEMY experi-
population does not decay before neutrino decoupling, which is ment using inverse beta decay reactions from captured relic
true for the entire parameter space we consider here. neutrinos [22]. Assuming a detector target mass of MT ,

123507-4
DARK RADIATION FROM INFLATIONARY FLUCTUATIONS PHYS. REV. D 103, 123507 (2021)

To see this, note that the late time flux of neutrinos from
V decays is

dϕ f 0 Ω ρ Γ e−ΓV ðt−t⋆ Þ
¼ V dm cr V Θðt − tν;dec Þ; ð22Þ
dΩdEν 2πmV Eν Hðzc Þ
where Eν is the observed energy of a present day neutrino
emitted atpredshift z with energy Eν ð1 ffi þ zÞ ¼ mV =2,
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
HðzÞ ¼ H0 ΩΛ þ Ωm ð1 þ zÞ3 þ Ωr ð1 þ zÞ4 is the Hubble
rate, H0 ¼ 67 km=sec=Mpc [25], and zc ¼ ½mV =ð2Eν Þ − 1.
The theta function ensures that decays before neutrino
decoupling do not contribute to the flux; this population
will thermalize with CνB. In Fig. 3 we show representative
flux spectra for both early (tν;dec < τV < tCMB ) and late
time ðτV > tCMB Þ decaying populations. From Eq. (21),
early decaying V with low ∼ few eV masses can yield
appreciable PTOLEMY signal rates, but as we see in these
FIG. 3. Present day neutrino flux spectra from V → ν̄ν decays
spectra, the fluxes similar to the CνB unless mV much
for representative benchmark points (dashed). Also shown are
spectra from the cosmic neutrino background (CνB), primordial
greater, which trades off against the overall rate as
neutron decays during BBN (n → ν), tritium decays during BBN R ∝ m−1V . It is possible to get an appreciable PTOLEMY
(T → ν), and solar neutrinos [21]. From Eq. (21), is clear that the flux for a low mass particle, but obtaining a distinctive
early decaying parameter points (green and red) only yield spectral shape requires late time decays, which do not
appreciable (≳ few) events at PTOLEMY for lower values of affect ΔN eff .
mV , which are difficult to distinguish from the CνB spectrum, but
might be detected as an enhancement over the Standard Model VI. CONCLUSION
signal rate. As the mass is increased, the spectrum gets harder, but
the rate becomes unobservable with a feasible exposure; for the In this paper we have studied the fate of massive vector
50 keV benchmark, we find R ∼ 10−3 events=yr at PTOLEMY. particles produced gravitationally from inflationary fluctu-
For later decaying particles (blue dashed curve) the rate and ations. If these vectors only interact with the SM via kinetic
spectrum can be favorable, but there is no contribution to ΔN eff . mixing, for m < 2me, the only allowed decay is V → 3γ
which is sharply suppressed, so V is generically metastable
electron neutrino fraction f νe , neutrino capture cross section can serve as a dark matter candidate [15]. However, if the
on tritium σ ¼ 3.83 × 10−45 cm2 , and the excess neutrino vector arises in well-motivated, minimal Uð1Þ gauge
density from Eq. (18), the signal rate is estimated to be [21] extensions from Eq. (2), it must couple to neutrinos, so
if at least one neutrino mass eigenstate is sufficiently light,



0 

5 MT f νe fV eV V → ν̄ν decays can efficiently deplete this inflationary
R≈ ; ð19Þ population and increase the relic neutrino density, thereby
yr 100 g 0.5 0.05 mV
predicting ΔN eff ≠ 0. For certain regions of parameter
which only assumes that the V decay after decoupling. space, the same neutrino population may be observable
However, for V that also decay before recombination, the at late times with the PTOLEMY experiment; for long-lived
fraction satisfies [21] vectors that decay after recombination, it is also possible to
obtain an appreciable PTOLEMY signal even


3 
0 ΔN eff 10 yr though ΔN eff ¼ 0.
f V ≈ 0.42 ; ð20Þ Intriguingly. due to a cancellation, this contribution
0.3 τV
depends only on H I and g as long as the V lifetime falls
so the rate for early decaying V can be written within this time window. For a wide range of model




 parameters, the ΔN eff prediction in these scenarios is
10 M T f νe ΔN eff 10 eV within reach of CMB-S4 projections [20]. We note that,
R≈ ; ð21Þ outside of the narrow parameter region where
yr 100 g 0.5 0.3 mV
50 keV ≲ T decay ≲ MeV, this scenario predicts ΔN eff ≠ 0
which may be detectable with a year of exposure at only in CMB data because nearly all of the V decays occur
PTOLEMY, whose projected CνB sensitivity is at the after BBN has completed; decays before BBN thermalize
∼10 event level. Note that there is general tension between with the SM, so T ν =T γ does not deviate from the SM
having an appreciable ΔN eff signal and having a prediction. However, for parameter space in which decay
distinguishable neutrino spectrum with a detectable occurs after recombination, the resulting neutrino popula-
PTOLEMY rate. tion may be observable directly at PTOLEMY [21,22].

123507-5
GORDAN KRNJAIC PHYS. REV. D 103, 123507 (2021)

Furthermore, since the mechanism studied here is intriguing that the contributions required to reduce its
sensitive to the Hubble scale during inflation, future statistical significance are readily accommodated in the
measurements of inflationary B-modes at CMB-S4 experi- parameter space studied in this class of models.
ments will have important implications for this class of
scenarios. The forecasted sensitivity to the scalar-to-tensor
ACKNOWLEDGMENTS
ratio r ∼ 10−3 implies a sensitivity to HI ∼ 1012 GeV [37],
which yields observable ΔN eff from V decays in the upper We thank Asher Berlin, Nikita Blinov, Dan Hooper,
half of Fig. 2. Kevin Kelly, Rocky Kolb, and David McKeen for helpful
Finally, it has been shown that contributions to ΔN eff ∼ conversations. This manuscript has been co-authored by
0.5 may play an important role in alleviating the discrep- employees of Fermi Research Alliance, LLC under
ancy between early and late time determinations of the Contract No. DE-AC02-07CH11359 with the U.S.
Hubble tension [23,24]. Although models with nonzero Department of Energy, Office of High Energy Physics.
ΔN eff do not completely eliminate the tension, it is

[1] D. Baumann, Inflation, in Theoretical Advanced Study [19] M. Bauer, P. Foldenauer, and J. Jaeckel, J. High Energy
Institute in Elementary Particle Physics: Physics of the Phys. 07 (2018) 094.
Large and the Small (2009), https://doi.org/10.1142/ [20] K. N. Abazajian et al. (CMB-S4 Collaboration),
9789814327183_0010. arXiv:1610.02743.
[2] V. F. Mukhanov, H. Feldman, and R. H. Brandenberger, [21] D. McKeen, Phys. Rev. D 100, 015028 (2019).
Phys. Rep. 215, 203 (1992). [22] E. Baracchini et al. (PTOLEMY Collaboration),
[3] D. J. Chung, E. W. Kolb, and A. Riotto, Phys. Rev. D 59, arXiv:1808.01892.
023501 (1998). [23] E. D. Valentino, O. Mena, S. Pan, L. Visinelli, W. Yang, A.
[4] V. Kuzmin and I. Tkachev, Phys. Rev. D 59, 123006 Melchiorri, D. F. Mota, A. G. Riess, and J. Silk,
(1999). arXiv:2103.01183.
[5] Y. Ema, R. Jinno, K. Mukaida, and K. Nakayama, J. [24] L. Knox and M. Millea, Phys. Rev. D 101, 043533 (2020).
Cosmol. Astropart. Phys. 05 (2015) 038. [25] N. Aghanim et al. (Planck Collaboration), Astron. As-
[6] Y. Ema, R. Jinno, K. Mukaida, and K. Nakayama, Phys. trophys. 641, A6 (2020).
Rev. D 94, 063517 (2016). [26] E. W. Kolb and M. S. Turner, Front. Phys. 69, 1 (1990).
[7] Y. Ema, K. Nakayama, and Y. Tang, J. High Energy Phys. [27] M. Escudero, D. Hooper, G. Krnjaic, and M. Pierre, J. High
09 (2018) 135. Energy Phys. 03 (2019) 071.
[8] D. J. Chung, E. W. Kolb, and A. J. Long, J. High Energy [28] M. Tanabashi et al. (Particle Data Group), Phys. Rev. D 98,
Phys. 01 (2019) 189. 030001 (2018).
[9] M. A. Fedderke, E. W. Kolb, and M. Wyman, Phys. Rev. D [29] P. De Salas, S. Gariazzo, O. Mena, C. Ternes, and M. Trtola,
91, 063505 (2015). Front. Astron. Space Sci. 5, 36 (2018).
[10] D. Seckel and M. S. Turner, Phys. Rev. D 32, 3178 (1985). [30] N. Blinov, K. J. Kelly, G. Z. Krnjaic, and S. D. McDermott,
[11] A. D. Linde, Phys. Lett. 158B, 375 (1985). Phys. Rev. Lett. 123, 191102 (2019).
[12] L. Parker, Phys. Rev. Lett. 21, 562 (1968). [31] P. F. Depta, M. Hufnagel, K. Schmidt-Hoberg, and S. Wild,
[13] P. Adshead and E. I. Sfakianakis, J. Cosmol. Astropart. J. Cosmol. Astropart. Phys. 04 (2019) 029.
Phys. 11 (2015) 021. [32] D. J. Marsh, Phys. Rep. 643, 1 (2016).
[14] P. Adshead, L. Pearce, M. Peloso, M. A. Roberts, and L. [33] S. Dodelson and L. M. Widrow, Phys. Rev. Lett. 72, 17
Sorbo, J. Cosmol. Astropart. Phys. 06 (2018) 020. (1994).
[15] P. W. Graham, J. Mardon, and S. Rajendran, Phys. Rev. D [34] L. J. Hall, K. Jedamzik, J. March-Russell, and S. M. West, J.
93, 103520 (2016). High Energy Phys. 03 (2010) 080.
[16] M. Pospelov, A. Ritz, and M. B. Voloshin, Phys. Rev. D 78, [35] A. Fradette, M. Pospelov, J. Pradler, and A. Ritz, Phys. Rev.
115012 (2008). D 90, 035022 (2014).
[17] S. D. McDermott, H. H. Patel, and H. Ramani, Phys. Rev. D [36] A. Berlin and N. Blinov, Phys. Rev. Lett. 120, 021801
97, 073005 (2018). (2018).
[18] I. Hoenig, G. Samach, and D. Tucker-Smith, Phys. Rev. D [37] CMB-S4 Collaboration, arXiv:2008.12619.
90, 075016 (2014).

123507-6