NORDITA-2018-063, MPP-2018-237

Axion Miniclusters in Modified Cosmological Histories

1, 2, 3, 4, 5, Luca Visinelli ∗ and Javier Redondo † 1Department of Physics and Astronomy, Uppsala University, L¨agerhyddsv¨agen1, 75120 Uppsala, Sweden 2Nordita, KTH Royal Institute of Technology and Stockholm University, Roslagstullsbacken 23, 10691 Stockholm, Sweden 3Gravitation Astroparticle Physics Amsterdam (GRAPPA), Institute for Theoretical Physics Amsterdam and Delta Institute for Theoretical Physics, University of Amsterdam, Science Park 904, 1098 XH Amsterdam, The Netherlands 4Departamento de F´ısica Te´orica, Universidad de Zaragoza. c/ Pedro Cerbuna 12, 50009 Zaragoza, Spain 5Max Planck Institute for Physics, Foehringer Ring 6, 80805 Munich, Germany (Dated: December 30, 2019) If the symmetry breaking leading to the origin of the axion dark matter field occurs after the end of inflation and is never restored, then overdensities in the axion field collapse to form dense objects known in the literature as axion miniclusters. The estimates of the typical minicluster mass and radius strongly depend on the details of the cosmology at which the onset of axion oscillations begin. In this work we study the properties and phenomenology of miniclusters in alternative cosmological histories and find that they can change by many orders of magnitude. Our findings have direct implications on current and future experimental searches and, in the case of discovery, could be used to learn something about the universe expansion prior to Big-Bang-Nucleosynthesis.

I. INTRODUCTION The history and the properties of the present axion field strongly depend on the moment at which the break- The nature of the cold dark matter (CDM) remains ing of the PQ symmetry occurs with respect to infla- unknown to date despite the growth of evidence in sup- tion [44–56]. If the PQ symmetry breaking occurs after port of its existence coming, on top of the original mo- inflation, a fraction of the total axions component is ex- tivations [1,2], from gravitational lensing [3], the cos- pected to organize into gravitationally bound structures mic microwave background radiation (CMBR) [4,5], also known as axion “miniclusters” [57–61], prompted by the in combination with Lyman-α and weak lensing [6], the inhomogeneities of the axion field in this scenario. Axion hierarchical structure formation of the observable uni- miniclusters are compact objects with a density of various verse [7], the formation and evolution of galaxies [8–10], orders of magnitude higher than the present local CDM galactic collisions [11, 12], and a plethora of other obser- density. Inside axion miniclusters another type of ex- vational techniques. otic structure, an axion star [62–84], could possibly form. Among the many hypothetical particles that could It has been argued that the first miniclusters that ever compose the CDM is the quantum chromodynamics come into place have a characteristic size of the order of 12 10− solar masses. This scale is much smaller than the (QCD) axion [13, 14]. The axion is the pseudo-Goldstone ∼ boson arising in the spontaneous breaking of a U(1) sym- smallest clump that WIMPs would assemble into in the metry first introduced by Peccei and Quinn (PQ [15, 16]) standard cosmology, because of their much longer free- 1 to address the strong-CP problem [17–20]. The fact that streaming length . Therefore, detecting these clumps axions could solve two distinct problems in physics makes provides a unique discrimination signature among CDM its search particularly appealing. If the axion field ex- candidates. As structure formation evolves, axion mini- ists, it has a very small mass and faint couplings to or- clusters are expected to hierarchically assemble into dark dinary particles. Both happen to be suppressed by a matter halos of galactic size, forming minicluster halos. new energy scale, the axion decay constant fA, which This claim has yet to be addressed numerically, as well as corresponds approximately to the scale of PQ symmetry the possibility that miniclusters might have not survived breaking and which is constrained by axion phenomenol- tidal disruption. Some studies suggest that miniclusters > 7 1 survive hierarchical structure formation to date [86–89], ogy to be fA 10 GeV. In particular, the scale fA− sets arXiv:1808.01879v3 [astro-ph.CO] 26 Dec 2019 the axion coupling∼ to two photons, which opens the pos- claiming that it is possible to constrain the fraction of sibility for axion electrodynamics [21–26] and promising dark matter in halos using micro-lensing data [90, 91] laboratory detection methods [27–32]. In the literature, and femto-lensing in the future [92] (see also [93]). Semi- mixed dark matter models in which the axion makes up a analytic results on the mass function of axion miniclus- fraction of the dark matter while the rest is in the form of ters are available today [94], with refined numerical work weakly interacting massive particles (WIMPs) have also recently reported [95, 96]. been considered [33–35]. See Refs. [36–43] for reviews of Today, the standard axion miniclusters would be grav- the QCD axion. itationally bound clumps of axions with masses of the

∗ Electronic address: [email protected] 1 See Ref. [85] for the effect of a non-standard cosmology on the † Electronic address: [email protected] free-streaming length of WIMPs. 2 order of the largest asteroids like Vesta or Pallas, and dominant contribution to the CDM they become CDM at of size comparable to an astronomical unit. Miniclusters temperatures GeV, precisely when we cannot ascertain could suffer tidal disruption by stars, with the value of the crucial assumption∼ of radiation domination. bound axions diminishing with respect to its early uni- In this paper we want to drop entirely this assumption verse estimate. It is usually expected that a sizeable frac- and study the properties of axion miniclusters in differ- tion of the galactic DM axions is bound into minicluster ent non-standard cosmologies (NSCs) before BBN. The structures, the remaining part forming a relatively ho- mass and the radius of a minicluster depend crucially on mogeneous halo. Despite the large number of clumps the size of the causal horizon at the time when axions expected, the Earth would encounter only a few such become non-relativistic, which in alternative cosmologies objects every galactic year (!). On the other hand, if might differ by various orders of magnitude with respect the fraction of axions bound into miniclusters is close to to the standard scenario. In order to focus the discus- unity, the direct detection of axions in microwave cavity sion on the novel aspects, we assume that axions make searches could be severely affected. A negative search by up the totality of the cold dark matter observed and that a cavity experiment could be an indication that axions essentially all axions fall into miniclusters. Our results, are mostly organised into clumped structures like mini- summarised in TableI are quite spectacular! In fact, clusters. Since axion detection is sensitive to the local the typical minicluster mass and radius can change by CDM energy density, a clumpy axion distribution would many orders of magnitude with respect to the predictions lead to spikes in the axion detection spectrum and be rel- achieved in the standard cosmology. Most importantly, evant for a direct detection technique. The interest in all the time and duration of encounters with the Earth can of the upcoming axion detectors then lies in the present be largely enhanced or suppressed, opening many possi- phase-space distribution of the axion CDM, which is not bilities for the direct detection of axion dark matter. expected to be homogeneous even at the interstellar scale. Previous works have already studied how the axion A reliable detection must take into account the possibil- mass for which the axion explains the totality of the ity of a inhomogeneous CDM distribution either in space observed CDM is modified in several NSCs [107–109] (axion miniclusters and stars) or in momentum (low- but the changes to the properties of miniclusters are dispersion filaments from tidal stripping). Moreover, if presented here for the first time. Moreover, regarding the axion is discovered by a haloscope experiment like the treatment of NSCs and the axion DM mass, we ADMX, HAYSTAC or CULTASK, the energy spectrum improve over previous results by computing the non- will be immediately measured and could be used to do standard cosmologies numerically including the recent galactic astronomy [97]. The spectrum and its variation detailed input from lattice QCD (equation of state and in time (daily and anual modulations) can be used to axion mass) [110] and presenting simple analytical com- identify substructure in the axion DM distribution like parisons with the standard radiation dominated case. tidal streams from dwarf-galaxies [98] or even from axion Note that, in addition to the cold axion population, a miniclusters themselves, obtaining their main properties modified cosmology would alter the yield of thermally with precision [97]. With these techniques, astronomical produced axions [111], as well as of any other light par- quantities like the solar peculiar velocity could be mea- ticle such as neutrinos [112, 113]; however, since for the sured even better than with ordinary astronomy. Several range of masses considered in this work, mA < 10 meV, variations of the haloscope concept allow to measure the the thermal population is a sub-dominant component∼ to axion velocity distribution by enabling directional detec- the total energy density, we do not discuss this contribu- tion [99–101], increasing the precision and decreasing the tion further. required measurement time. Understanding the fraction This paper is organised as follows. Sec.II is a brief re- of axions bound in miniclusters and of the axion phase- view of the production of cosmological axions. In Sec.III space will maximise the outcome of the various experi- we set the stage for the non-standard cosmological mod- ments that will start looking for axion CDM in the near els of interest and we review their impact on the DM future. axion mass. Sec.IV presents our results on the proper- The rough properties of axion miniclusters described ties of axion miniclusters and their phenomenology. Final above have been studied only under the assumption remarks and conclusions are drawn in Sec.V. that the universe is radiation-dominated when axions become non-relativistic. This is certainly a simple and minimal assumption but also one that does not need II. AXION COLD DARK MATTER to be necessarily correct. We know with some cer- tainty that the universe expansion must be radiation- In the post-inflationary scenario, axion cold dark mat- dominated after and around neutrino decoupling (tem- ter is produced from the misalignment mechanism and peratures of T < 5 MeV) not to alter the successful pre- the decay of topological defects (strings and walls). The dictions of radiation-dominated∼ Big-Bang Nucleosynthe- axion field behaves like a massless field, becoming ho- sis (BBN) [102–106], but we have no direct evidence of mogenous on scales of the order of the causal horizon, the expansion rate of the universe at earlier times (higher as long as the Hubble expansion rate (H = d log a/dt, temperatures). Most importantly, if axions are to be a with a = a(t) the scale factor of a Friedmann-Robertson- 3

Walker universe) is much larger than the axion mass, contribution from the topological defects to the axion H mA. However, around the time t (or equivalently cold dark matter density as an additional factor multi-  1 the temperature T1) when the condition plying the misalignment contribution,

wall str mis mis 3H(T1) mA(T1), (1) ρ¯A = ρ + ρ + ρ α ρ . (4) ≈ A A ≡ tot A is met, the relevant long-wavelength modes begin to re- This parametrisation allows for the case in which the spond to the QCD potential by evolving towards the min- axion CDM yield is even smaller than the misalignment- imum of the field configuration A = 0 (where A(x) is only contribution αtot < 1, a scenario supported by re- the axion field), oscillating around it as a non-relativistic cent simulations [119–121]. (NR) field with equation of state wA 0, i.e. equiva- lent to CDM. Shortly after that time, the→ energy density distribution, which corresponds to the axion field oscil- III. AXION POPULATION IN MODIFIED lating with different amplitudes at different positions, be- COSMOLOGICAL SCENARIOS comes essentially frozen. The calculation of the spatially- averaged CDM yield from the misalignment mechanism A. Parametrising a cosmological model is reviewed in AppendixA and it is given by  3 In this paper we consider a non-standard cosmology mis 1 2 2 a1 ρA = mA(T0)mA(T1)fA θi = (NSC), a scenario in which the early universe, close to 2 h i a0 the relevant temperature T1 < (1) GeV, contains the  3 O 1p 2 a1 standard thermal bath of SM particles∼ and an additional = χ(T0)χ(T1) θi . (2) 2 h i a0 component labelled φ, whose energy density is so large during the temperatures considered that it affects the ex- In the last expression, we have introduced the pansion rate of the universe. The evolution of the energy temperature-dependent axion mass m (T ), which is re- A density of radiation (ρR) and of the extra substance (ρφ) lated to the axion decay constant fA by is governed by a set of coupled Boltzmann equations, 2 2 mA(T ) fA = χ(T ), (3) ρ˙φ + 3(1 + wφ)Hρφ = Γφρφ, (5) − where χ(T ) is the topological susceptibility of QCD. For ρ˙R + 3(1 + wR)HρR = Γφρφ, (6) practical purposes it is useful to define the energy scale 1/4 where a dot denotes a derivative with respect to cosmic ΛA = (χ(T = 0)) 75.5 MeV in terms of the sus- time t, the parameter wi = pi/ρi describes the equa- ceptibility at zero temperature≈ [110, 114]. In Eq. (2) tion of state of substance i with pressure pi and energy we introduced the standard notation, in which we have density ρi (with i φ, R ), and Γφ is a possible decay normalised the axion field with f by defining the angle ∈ { } A rate. The Friedmann equation H2 = (8π/3m2 )(ρ +ρ ) θ = A/f ( π, π), and we wrote θ2 for the average Pl R φ A i gives the evolution of the scale factor (and thus H) as a of the initial∈ misalignment− angle squared,h i including the function of time. In order to connect with a late cos- contribution from the non-harmonic terms in the axion mology in agreement with observations we will consider potential in Eq. (A2)[40, 52, 115, 116]. Finally, the last substances that decay and/or redshift faster than radia- factor in Eq. (2) involving the ratio of the scale factor tion, so that the description of the history of the universe at t and today t , is the dilution of the axion number 1 0 transitions from the NSC to the standard scenario after a density due to the expansion of the universe from the certain reheat time (t ) or equivalently below a certain moment at which the axions become NR and behave as RH temperature, which we indicate as the reheating temper- CDM until present time. Note that the energy density in ature T . We define the reheat time implicitly by the Eq. (2) is a quantity uniquely related to QCD scales: the RH condition ρ (t ) = ρ (t ). The Hubble rate at time axion decay constant enters the computation through the φ RH R RH tRH is then determination of the temperature T1 at the onset of the s field oscillations in Eq. (1), or equivalently the scale fac- r 3 2 8π ρφ(tRH)+ρR(tRH) 8π T tor a1 = a(T1). The quantity T1 is sensitive to the energy RH HRH = 2 g (TRH) . content of the universe, therefore any deviation from the 3 mPl ≈ 45 ∗ mPl standard cosmological model affects the axion DM yield (7) and the size of the horizon ( 1/H(t1) 1/H1) when In the last step we have assumed that the extra energy in- axions start to behave as CDM.∼ ≡ jected reaches the thermal equilibrium on a much smaller DM axions are also produced from the decay of strings time-scale than the Hubble expansion rate, so that the str and domain walls, with energy density contributions ρA contribution from the radiation to the energy density is wall 2 4 and ρA respectively that scale with the same exponent ρR = (π /30)g (T )T . ∗ in the decay constant fA [38, 117, 118], at least when the Once the extra substance has decayed, its energy den- details of the production and the scaling with the power sity ρφ can be neglected in Eq. (6), which can be inte- spectra are neglected. This allows us to parametrise the grated to relate temperature and scale factor by the usual 4 conservation of comoving entropy, B. Examples of modified cosmologies

3 3 gS(T ) T a = constant, (8) We consider explicitly three modified scenarios which are justified in known extensions of the Standard Model. where gS(T ) and g (T ) are the number of effective de- ∗ grees of freedom respectively in entropy and energy at Early matter decay domination (MD) temperature T , for which we use the parametrisation in • In this scenario, we take the early universe en- Ref. [110]. ergy density to be dominated by a gas of NR mas- However, Eq. (8) is not guaranteed to hold “during” sive particles (or a coherently oscillating NR field, the NSC period if the non-standard “substance” decays which amounts to the same) for which we have into SM particles. The decay products will thermalise w = 0, β = 2/3. These particles have to de- and release additional entropy. In such a modified sce- φ cay in order for the standard cosmology to take nario, the decrease in temperature with the scale fac- place. This scenario has been used extensively tor would be slower than the standard T 1/a law in to study low-reheat temperature (LRT) cosmolo- Eq. (8). Therefore, we write down the relation∝ for en- gies [127–132], and it is famously problematic in tropy conservation in a NSC as the case of string moduli. One such tractable sce- 3 3α nario embed in string theory is the large volume gS(T ) T a = constant, (9) scenario [133, 134], where a unique modulus field where 0 < α 1 is a quantity that we use to parametrise appears. The decay leads to a non-conservation of ≤ the entropy density α = 1, in particular the model the entropy injection and the subsequent decrease in the 6 temperature decrease with the scale factor. Note that predicts α = 3/8 [123]. entropy conservation relates to α = 1, for which Eq. (9) We present the evolution of H and a in this sce- returns the standard case in Eq. (8), while values of α > 1 nario in Fig.1 as red lines, obtained by solving would require a substance that efficiently sucks up energy numerically Eqs. (6) and (5) with wφ = 0. As long from the SM bath, not contemplated in this study. as ρφ dominates over ρR at some point, the decay It proves useful to introduce a generic relation between rate Γφ fixes to a good extent the reheating tem- time and scale factor, perature. In Fig.1 we have adjusted Γ φ to obtain TRH 0.5 GeV and we have solved for different a tβ, (10) ' ∝ initial matter abundances (thicker lines have larger ρφ/ρR ratio). We distinguish three regimes for H: where we assume the range 0 < β < 1, corresponding 1) At early times (high T ) the radiation from the to a cosmology in which the expansion is decelerating. decaying matter field is subdominant with respect Formally, β = 2/(3(1+w)) where w is the mean equation to a pre-existing radiation so effectively α = 1 and of state of the various fluids that are present at a certain 3/2 3/2 H a− T ; 2) The radiation from the de- time. During the NSC phase, the equation of state is well caying∝ field∝ dominates over the pre-existing one. In approximated with that of the substance dominating the MD cosmology, the Hubble rate H decreases much energy density for times t < tRH, so w wφ. Eq. (10) faster with T because T is decreasing slower in time does not include the important case w =≈ 1, β in − → ∞ due to the radiation generated during the decay which a(t) exp(H t), which we do not treat explicitly. 3/2 3/2α 4 ∝ (H a− T T ); 3) The radiation- The Hubble rate during the modified cosmological epoch ∝ ∝ ∼ 2 dominated period after the decay H T − . If the immediately follows from Eq. (10) as H a/a˙ = β/t. initial matter abundance is much larger∝ than the It is a simple exercise to solve Eqs.≡ (5) and (6) at < pre-existing radiation, only the second and third early times t 1/Γφ and for a non-zero Γφ, assuming period will be relevant for us. We will mainly focus ∼ that the radiation produced via the decay of the φ field on this simple scenario (thick red line), for which dominates over any pre-existing radiation, see for exam- indeed we reproduce α = 3/8, β = 2/3 before re- ple Refs. [112, 122–126]. Under these conditions, using 4 heating. Eq. (10) and the definition ρR T (ignoring g (T )), we find an interesting relation between∝ α and β [123∗ ], Early Kination domination • In the Kination cosmology [135–140], the early uni- 3 1 α = (1 + wφ) = . (11) verse expansion is driven by the kinetic energy of a 8 4β scalar field. The field φ is a “fast-rolling” field with To find the expression for H(T ) at early times, we com- an equation of state wφ = 1 (pressure pφ equals αβ the energy density ρφ). The energy density of the bine Eqs. (9) and (10) to obtain t 1/T , so that the 6 expression for H(T ) valid for T T ∝ reads field φ scales as ρφ a− , so that it redshifts faster RH ∼ 4 ≥ than the radiation energy density ρR a− and be- 1 ∼ 1 ! αβ comes subdominant below TRH. Therefore, during   3 gS(T ) T Kination domination we have β = 1/3. The ther- H(T ) HRH . (12) ≈ gS(TRH) TRH mal production of a WIMP during Kination has 5

100 104 STD STD Matter Decay Kination decay 10 2 − 101 Kination

4 10− 2

[eV] [eV] 10−

H 6 H 10− 5 10− 8 10−

8 10− 10 10− Hubble rate Hubble rate

11 12 10− 10−

14 14 10− 10− 3 2 1 0 1 2 3 2 1 0 1 2 10 10 10 10 10− 10− 10 10 10 10 10− 10− 100 100

1 10− 2 10− 2 10− MeV MeV 3 4 10− 10− a/a a/a

4 10−

6 10− 5 10−

6 scale factor scale factor 10− 8 10− 7 10−

10 8 10− 10− 3 2 1 0 1 2 3 2 1 0 1 2 10 10 10 10 10− 10− 10 10 10 10 10− 10− T [GeV] T [GeV]

FIG. 1. Evolution of Hubble rate and scale factor (normalised FIG. 2. Evolution of Hubble rate and scale factor (nor- to the value at T = 1 MeV) in Early matter domination (MD) malised to the value at T = 1 MeV) in Kination cosmologies cosmologies (red lines) compared to standard radiation dom- (blue line) and decaying Kination (KD) (green lines) com- ination (black). The red lines differ only in the initial mat- pared to standard radiation domination (black). The decay- ter/radiation fraction, which increases with increasing line- ing Kination lines differ only in the initial Kination/radiation thickness. All the models share the same decay rate Γφ, here fraction, which increases with increasing line-thickness. All set to obtain TRH ∼ 0.5 GeV. the models share the same decay rate Γφ, here set to ob- tain TRH ∼ 0.5 GeV. In the non-decaying scenario, the initial abundance has been set to obtain the same TRH. been discussed in Refs. [139–150]. Since the Kina- tion field φ does not decay but it redshifts away, we 2 we show cosmological histories for different ini- have entropy conservation with α = 1. tial Kination densities with respect to pre-existing The solution of Eqs. (5) and (6) with wφ = 1 is radiation (increasing abundance follows the line 3 3 shown in Fig.2 as a blue line ( H a− T ). thickness). As in the MD case, when the ther- The reheating temperature can be adjusted∝ ∝ by the mal bath is dominated by the pre-existing radia- initial value of ρφ for a given temperature (equiv- tion, the Hubble rate H follows a softer power law 3 3 alently ρR). In Fig.2 we have adjusted the initial (H a− T ) than when the radiation from ∝ ∝ 2 3 3/α 4 value of ρφ to obtain TRH 0.5 GeV. the φ-decay dominates (H a T = T ). ' − Deep inside the KD period∝ (thick∝ green line in Kination decay (KD) Fig.2), one finds α = 3/4 and β = 1/3. • We also consider a decaying φ field as in Ref. [149]. In this case, the reheating temperature is set by the abundance if the decay rate is very small and by the 2 The fact that the latter coincides with the MD case is not a co- decay rate if the abundance is very large. The sec- incidence. For any decaying substance when its radiation domi- −1/β 1/αβ 4 ond makes the limiting case of interest here. In Fig. nates the temperature H ∝ a ∝ T ∼ T . 6

C. The dark matter axion mass where in the last equality we have used the parametri- sation for the scale factor a = a(T ) in Eq. (9) when the For a given NSC, there is one specific value of the ax- production of entropy is considered. Note that if the ef- ion mass (or equivalently of the axion decay constant fect of a NSC is that of decreasing the scale factor a1 at fA) for which axions account for all of the observed the onset of axion oscillations, then T1 should also de- 48 4 crease in order to attain the observed CDM abundance CDM (ρCDM 9.72 10− GeV [4]). Such a value ∼ × ρ using Eq. (15). of mA(T 0), here denoted as mCDM and called the CDM “DM axion→ mass”, is in general a function of the param- We first focus on comparing different NSCs with the same T . Later we will discuss the effects of T for a eters that define the NSC model (TRH, α, β), as well as RH RH given cosmology. In Fig.3 (top panel), we plot a scheme the αtot factor. We refer to the value in the standard cos- std of the evolution of ρ0(a1) as a function of a1 = a in the mological scenario as mCDM. Generally we denote by mA the value of the axion mass at zero temperature and we standard radiation dominated cosmology (RD), Kination note explicitly the temperature as mA(T ) when relevant. (K), Kination decay (KD) and matter decay (MD) sce- In the remainder of the paper we want to assume for narios. We also show a horizontal line denoting the ob- simplicity that axions account for all the CDM so we need served value that we are interested to reproduce. The to compute it for the different NSCs of interest. Tech- point at which ρ0(a) intersects ρCDM corresponds to a1 nically, we compute these quantities as follows. Given a in the different cosmologies and yields the valueρ ¯A in Eq. (4). By defining the ratio of ρ (a ) to ρ as NSC model, equatingρ ¯A = ρCDM, withρ ¯A as in Eq. (4), 0 RH CDM Q ρ (a )/ρ and using Eq. (15), we find imposes a direct constraint on the value of the tempera- ≡ 0 RH CDM ture T , so that using the definition in Eq. (1) we obtain a 1 a = RH , (16) the value of f as a function of α for which the axion is 1 1 A tot Q 3+γα the CDM particle. Since this calculations have been pre- sented elsewhere [127, 151, 152], here we only outline the so that the comparison with the case in which entropy is key aspects that differentiate the computation in NSC conserved α = 1, like in the standard cosmology, gives models. For simplicity in the exposition we neglect O(1) a 1 1 1 3+γ 3+γα factors involving the slowly varying numbers of degrees std = Q − . (17) a1 of freedom, g (T ), g S(T ). A technical∗ point∗ is that the topological susceptibility We obtain that the scale factor at which axions become χ(T ), and likewise the axion mass, depends on temper- CDM depends only on α but not on β. The dependence ature. In the numerics of this paper we use the recent on TRH is hidden in the definition of Q and will be dealt lattice calculation in the lattice from Ref. [110], but when with later. considering analytical calculations a truncated power-law If entropy is conserved in the NSC (α = 1), oscillations expression is useful [153], as commence at the same value of the scale factor as in std standard radiation dominated history, a1 = a1 . In this std  case, we should also have T1 = T in order to respect  2 1 for T TΛ, 1 mA(T ) χ(T )  2γ ≤ the CDM constraint. In our small collection of NSCs, = = TΛ (13) mA(0) χ(0)  , for T TΛ. this is the case of red-shifting Kination cosmology. T ≥ By contrast, if during the NSC period there has been an extra injection of entropy, the temperature redshifts where T 160 MeV and γ 4. Λ slower than 1/a (α < 1), the topological susceptibility In general,' the relevant sequence∼ of events in the his- increases more slowly with a, therefore we have to get tory of the axion field in a NSC reads an earlier onset of the axion oscillations to get the same i.e. std a < a < a , (14) final CDM, a1 < a1 . Note that the value of a1 in 1 RH Λ Eq. (17) decreases with decreasing α, therefore we have i.e., first the axion field begins to oscillate during the MD KD K std a1 < a1 < a1 = a1 , (18) NSC part of the early universe (a1), then the universe converges to the standard radiation-dominated descrip- because α = 3/8 during MD, α = 3/4 during KD, and tion (aRH), then the topological susceptibility saturates α = 1 during both Kination and the standard cosmology. with aΛ = a(TΛ). The latter turns out to be true for the Once the value for a1 has been obtained, the temper- ranges of parameters which are not yet excluded. ature at which axion oscillations commence, T1, is also The first important thing to note is that the present computed. For this, just consider the ratio of Eq. (15) for axion CDM energy density depends on the NSC param- the NSC and standard cosmology giving the same ρCDM, eters only through T1, or equivalently a1. For a given  3/γ T1 a1 value a1 of the scale factor at the onset of oscillations, std = std , (19) the present energy density ρ0(a1) is T1 a1

3 which implies the relation between temperatures a1 3+γα ρ0(a1) γ a1 , (a1 < aRH), (15) MD KD K std ∝ T1 ∝ T1 < T1 < T1 = T1 . (20) 7

γ 4 3 Writing Eq. (1) in the form 3H1 = mCDM (TΛ/T1) and 10− 10− 105 taking ratios of such an expression when evaluated for a standard RD, Kination NSC and the standard scenarios gives 104 Kination decay KD Matter decay MD γ 3 m H  T  H  a  103 CDM = 1 1 = 1 1 , (21) mstd Hstd T std Hstd astd CDM 1 1 1 1 102

where in the last step we used Eq. (19). Therefore, the CDM 101 3 /ρ

DM axion mass depends on the particular product H1a1. 0 To proceed further we note that the Hubble expansion ρ 100 rate as a function of a during the NSC period is3, 1 10− 1/β aRH  < 2 H(a) HRH , (a aRH) . (22) 10− ' a ∼ 3 In standard RD cosmology with β = 1/2 we get 10− 14 2 10− H1a1 =constant, if we neglect changes due to g (T ). To ∗ standard RD compare NSCs with the standard RD case we show the Kination 2 combination H1a1 for the different cosmologies consid- Kination decay KD ered in the bottom panel of Fig.3. Matter decay MD K std In the case of Kination cosmology, for which a1 = a1 , 2 std ) the value of H1 is larger than H1 due to the Kination field whose energy dominates the expansion. We con- MeV 15

K std K std /a 10− clude from Eq. (21) that mCDM/mCDM = H1 /H1 > 1, 1 a i.e. the DM axion mass in the Kination cosmology is ( 1

generally larger than in standard scenario. H KD std In the case of the KD scenario, a1 < a1 and obvi- KD std ously H1 > H1 so the DM axion mass is not imme- diately clear. Indeed, a very interesting situation hap- 3 pens. Since H1 1/a1 during Kination cosmology, both 16 ∝ 10− 4 3 factors of a1 in Eq. (21) cancel out and the result be- 10− 10− comes independent of the particular value of a1 and equal a/aMeV KD std std to mCDM/mCDM = aRH/a1 , which turns out to be HK/Hstd, similarly to what we discussed for the case 1 1 FIG. 3. Top: Evolution of ρ0(a) in different non-standard of the Kination cosmology without the decay term. cosmologies considered in the text for the same “reheating” In the case of matter decay (MD) we have again temperature TRH = 0.5 GeV. Imposing ρ0 = ρCDM at scale MD std std a1 < a1 so again the DM axion mass depends on factor a1 we find a1 ≤ a1 . Bottom: The Hubble expansion 3 2 the balance between the H1 and a1 factors. In this case, rate, ×a , captures the main trend of the DM axion mass, MD MD 2 std std 2 (21). note that we have H1 (a1 ) < H1 (a1 ) because matter density redshifts slower than radiation. There- fore we conclude that the DM axion mass in the early decaying matter cosmology is generally smaller than in of reheating, it is larger or equal than the standard value the standard cosmology, mMD /mstd < 1. (because a /astd 1). Also, since the main dependence CDM CDM RH 1 ≥ Combining Eqs. (21) and (22), we obtain the depen- on a1 cancels out, the DM mass can only depend mildly dence of the DM axion mass on the scale factor, on whether the Kination field is decaying or not. Let us now study the dependence of the value of the 1  3 β mCDM aRH a1 − DM axion mass on TRH. First, let us note that using std std , (23) standard cosmology (for instance) we have mCDM ' a1 aRH 3+γ which makes very clear that for Kination cosmology (β = T std  Q = 1 , (24) 1/3), the DM axion mass only depends on the scale factor TRH

so that we replace Eq. (16) into Eq. (23) to write

3 3β−1 3+γ In this formula there is a small correction due to our definition   1+ β 3+γα mCDM 1 3−1/β TRH − of tRH, Eq. (7), which results in a difference in the two Hubble 3+γ 3+γα std Q − = std . (25) rates of the order of ≈ 10% when a  aRH. This factor does not mCDM ' T1 affect our qualitative discussion, but appears in Fig.3. For this reason we use ' instead of the equality in the rest of the Section. Therefore, for the NSC scenarios here considered, we ob- 8 tain difference is a minimal shift in T1, less severe in the MD  std  case. K KD std T1 mCDM mCDM mCDM , (26) ' ' TRH 100 101 102 103 104 0 4+3γ 10   8+γ MD std TRH m m , (27) 1 CDM CDM std 10− Astrophysical bounds ' T1

2 where the last exponent is 4/3 for γ 4. The chosen 10− ∼ ∼ NSCs have the virtue of representing scenarios in which 3

10− RH

the DM axion mass differs from the standard case in op- T K KD 4 posite ways, since m and m are generally larger 10− CDM CDM [eV] std MD std than mCDM while mCDM is generally smaller than mCDM. 5 10− Eq. (26) shows that the mass of the axion CDM in CDM Too low both Kination and Kination Decay models is approxi- m 10 6 mately the same. This might seem surprising because of −

7 the different relation relating the scale factor and temper- 10− Matter decay ature in the two cosmologies. However, the smaller value Kination 8 of T1(or a1) in the KD scenario (with respect to the K 10− Kination decay Standard RD scenario) means that the axion mass at T1 is larger (than 9 10− in K), but H1 is also larger by the same factor, so that γ the equality 3H1 = mCDM(TΛ/T1) is satisfied by the same value of mCDM. As explained, this is a bit coin- cidental and happens only for Kination where 1/β = 3, essentially because H during Kination redshifts with the scale factor a as CDM density. This coincidence is ob- 0

10 RH tained by modeling the evolution of a1,T1,H by broken power-laws and will be corrected by small O(1) factors T when the continuous evolution is taken into account, fac-

tors of g , gS are incorporated and by the fact that H(a) [GeV] ∗ 1 T

differs by O(10%) between K and KD scenarios due to Too low our definition of tRH. We have performed full numerical calculations, ade- quately including all the relevant temperature-dependent functions like g (T ), g S(T ), the susceptibility χ(T ) from Ref. [110] and the∗ numerical∗ results for the cosmological evolution of H(a),T (a), etc., computed in the previous 10 1 −100 101 102 103 104 section. In particular, we have solved numerically for T1 TRH[MeV] and mCDM as a function of the temperature TRH in the different NSCs. Our results are shown in Fig.4. The bottom panel of Fig.4 shows the value of the tem- FIG. 4. Top panel: The QCD DM axion mass as a function perature T1 required for having the totality of the CDM of the reheat temperature for different cosmological models. in axions. The black thin line represents the value of the Black line: (standard) radiation-dominated cosmology. Red DM axion mass in the standard cosmology. As expected, line: matter decay (MD) cosmology. Blue line: Kination. results in the standard and in the Kination cosmologies Green line: Kination with a decaying field (KD). Bottom superimpose because entropy is conserved in both scenar- panel: the temperature T1 at which the axion field begins ios and a 1/T is the same. The cosmologies with a coherent oscillations, as a function of the reheat temperature. 1 1 Solid and dashed lines assume α = 10 or α = 1, respec- decaying field,∝ KD (green) and MD (red) require smaller tot tot tively. values of T1 (and of a1) to obtain the same CDM abun- dance despite the larger entropy production and subse- quent CDM dilution. We have coloured in pink the region The upper panel of Fig.4 shows the DM axion mass. where the reheating temperature is smaller than 5 MeV The black thin line sets the value in the standard sce- ∼ and is thus excluded by BBN considerations [104]. It is an nario. Again, we show results for two values of αtot = 10 interesting fact that this constraint limits the lowest value and αtot = 1 in solid and dashed lines, respectively. We of T to be T (in the MD scenario) so that our previ- note as usual that larger α requires a larger m 1 ∼ Λ tot CDM ous analytical considerations where we assumed T1 > TΛ following a roughly linear dependence. As expected, Ki- are justified. For illustration, we show results for two nation and KD cosmologies require a similar DM axion values of the topological defects contributions αtot = 10 mass because of their particular expansion rate, β = 1/3 and αtot = 1 in solid and dashed lines, respectively. The that cancels the dependence on a1. This does not cancel 9 the dependence with the reheating temperature 1/TRH IV. AXION MINICLUSTERS predicted in Eq. (26). In the Kination scenario,∼ the DM mass becomes so large that it can conflict with the astro- Around the temperature T of the onset of oscillations, > 1 physical bounds at mA 30 meV [154–156] (orange re- the axion field is highly inhomogeneous. In the standard gion) so we have interrupted∼ the line at this approximate radiation dominated scenario, the density field becomes mass. This range of values is particularly interesting as frozen in comoving coordinates around that time [57]. axions could explain several astrophysical anomalies [157] At the same time, axions are rapidly acquiring a non- and be detected by IAXO [158]. However, this statement zero mass, faster than the redshift of their momentum depends on the assumed value of αtot. Using αtot = 1, with the expansion of the universe so that they become one hits the BBN constraint in Eq. (A1) before the as- non-relativistic and cannot further free-stream. Because trophysical bounds. The value of mCDM in a Kination of this, the correlation length of the axion DM density model can change up to two orders of magnitude and field is of the order of the causal horizon at T1, i.e. is only limited indirectly by the astrophysical bounds or 1/H1, and freezes in comoving coordinates. This pic- BBN. ture,∼ sketched by the seminal studies [58–60] has been The matter decay case is shown in red and is well de- recently confirmed in large scale numerical computations 4/3 scribed by mCDM TRH . The axion CDM mass can including cosmic strings [95]. Although the results show be almost three orders∝ of magnitude smaller than the a rich structure of even larger inhomogeneities at sub- standard value before conflicting BBN. In general, the horizon scales due to the dynamics of cosmic-strings, do- full numerical result agrees very well with the analyti- main walls and axitons, these substructures encapsulate cal trends reviewed there. The only apparent differences a relatively small amount of the DM [95] so we will ignore are attributable mostly to the matching of cosmologies it henceforth. In this section we derive the properties of around TRH and to the changes in g (T ) and gS(T ) to a the miniclusters in non-standard cosmologies. We begin lesser extent. ∗ by a brief description of the standard case.

A. Standard cosmology 1. Effects of pre-existing radiation in MD and KD Density of the minicluster.- The density inhomo- In principle, it is straightforward to extend our results geneities separate out from the cosmic expansion as grav- to the case in which the pre-existing radiation before itationally bound miniclusters around matter-radiation the decay of the matter or Kination field φ dominates equality [58], happening around a redshift is zeq = the temperature during the period when axions become a /a 1 3360 and plasma temperature T = 0 eq − ' eq CDM. This amounts to consider a period of α = 1 dur- T0a0/aeq 0.8 eV, much after our NSC has converged ing the NSC period. The physical idea is that the pre- to a standard' cosmology. We parametrise an overdensity 4 existing radiation redshifts as a− so T 1/a and in the axion CDM density as [60] ∝ ∝ α = 1. Let us denote by Tr, ar the temperature and scale ρc ρ¯A factor at which the radiation resulting from the decay of Φ = − , (29) the φ field catches up with the pre-existing amount. A ρ¯A 3+γ first modification is that ρ0 = ρ0(a) increases as a where ρc is the local density of the minicluster andρ ¯A 3+γα before ar, which is steeper than the a behaviour is the mean axion energy density. Since the energy den- shown in Fig.3 (top) This makes the∝ required value of sity in axion CDM divided by that of radiation scales as a1 to lie between the value of a1 calculated in a NSC ρ¯A/ρR a, an overdense region with Φ > 0 enters into rd std ∝ model (here labelled a1 ) and a1 = a1(α = 1). Let- matter-domination when the scale factor is aleq = aeq/Φ. rd ting ar vary within the range (a1 , aRH) yields a1 in the The overdensity grows linearly with the scale factor un- rd std range (a1 , a1 ). At this stage, Eq. (19) is still valid, so til it becomes of order unity, when collapses and forms rd std that the temperature T1 lies in the interval (T1 ,T1 ), a gravitationally bound virialised object. Since Φ is corresponding to the area between a given NSC line and typically of order unity [95], miniclusters collapse typi- the STD line in Fig.4 (bottom). The derivation of the cally around matter radiation equality. Note that values DM axion mass m is a bit more involved for a sim- Φ 1 are also possible, although these regions tend to CDM  ilarly simple and not surprising outcome: depending on be smaller than the causal horizon at T1 and thus cor- the values of the parameters ar and Tr, the DM axion respond to smaller, much less massive miniclusters, see mass can take any value between that of a given NSC below. The energy density of the minicluster at the mo- std line shown in Fig.4 (top) and the standard case mCDM. ment of collapse is For completeness, the expression equivalent to Eq. (25) ρc(T ) = (1 + Φ)ρa(T ) = reads collapse collapse  3 aeq eq 1 γ 1−α 3β−1 = (1 + Φ) ρA =  2 β   α 3+γ β mCDM TRH − TRH acollapse std = std . (28) 3 eq mCDM T1 Tr = (1 + Φ)Φ ρA , (30) 10

eq where ρA is the axion energy density at matter-radiation which corresponds to about one astronomical unit. equality. A detailed calculation that follows from the dy- namics of the spherical collapse and further virialisation of the minicluster obtains an extra factor of 140 [58, 60], B. Miniclusters in non-standard cosmologies so the expression we use in place of Eq. (30) for the en- ergy density of the minicluster is The properties of axion miniclusters like its mass, its 3 eq size, and the velocity dispersion, are modified in a num- ρc = 140 (1 + Φ) Φ ρA . (31) ber of ways by the presence of a NSC period. The most Recall that we will be assuming that axions account for important and obvious way is that the correlation length eq eq the entirety of the CDM, so that we can write ρR = ρm = of the axion DM field 1/H when axions become NR eq ∼ 1 ρA (1 + Ωb/ΩCDM), where Ωb/ΩCDM 0.19 is the ratio and thus DM can be totally different from the standard eq∼ 3 of baryonic matter to CDM. Using ρA = ρCDM(1+zeq) , radiation-dominated scenario. this gives A second potentially relevant effect is a non-trivial evo- lution of axion overdensities during the NSC. Indeed, it is 6 3 GeV ρc 6.7 10 (1 + Φ)Φ . (32) ∼ × cm3 well known that during radiation domination, CDM per- turbations only suffer a negligible logarithmical growth The cores of miniclusters have the typical energy den- with the scale factor, but during a period of matter dom- sity of either matter or radiation components at matter- ination perturbations can grow linearly and produce al- radiation equality. ready some collapses. Mass of the minicluster.- The mass of the axion Another potential issue is that the relation H minicluster is given by the dark matter mass enclosed 1 m(T ) defining the moment when axions become DM and' within the overdense region at about the onset of oscil- 1 the correlation length which gives us the size of the typ- lations [57, 58]. We consider a spherical region of physi- ical minicluster would suffer O(1) corrections in a NSC. cal radius R at temperature T . According to the sim- 1 1 A similar effect is the expected different values of α in ulations in Ref. [95], the typical value of R is of the tot 1 different NSCs, due for instance to the O(1) differences of order of 1/H , although the temperature T is bet- 1 1 the string-network density, see Ref. [159] for an example ter defined∼ slightly differently than in Eq. (1), that is in matter domination and the generalisation in Ref. [107]. without the factor of three which is useful to discuss the Since the two effects discussed before vary the miniclus- DM abundance, as H = H(T ) = m (T ). Due to the 1 1 A 1 ter properties by several orders of magnitude, while these large value of the exponent γ describing the temperature- latter ones impact the energy density only by a factor of dependence of the axion mass, this modification amounts order one, we will not discuss this issue further here. to only a 10% differences in the values of T1 and a1. Here, we parametrise∼ deviations from the canonical size as R1 = r/H1, with r = (1). In terms of the present 1. Impact of the horizon size at T CDM density, we have O 1  3 3 std 4π 3 4π r a0 The implications of the change of H1 and a1 in NSCs Mc = (1 + Φ)ρ1R1 = (1 + Φ)ρCDM , 3 3 H1 a1 to the properties of miniclusters are relatively trivial to (33) study. First of all, we have already prepared the stage mis mis where ρ1 αtotρA (T1) = αtot mA(T0) nA (T1), with in the previous section when we studied the DM axion mis ≡ mis ρA defined in Eq. (4) and nA (T1) defined in Eq. (B21) mass. Indeed, we showed in Eq. (21) how the DM mass 3 below. Using the numerical calculations in Ref. [95] to fix in a NSC is proportional to H1a1 (with an extra a1) and 17 0.1712 the quantity a0/a1H1 1.1 10 (50µeV/mA) cm, how this quantity compares in the different NSCs. More- we find ' × over, we have performed the numerical computations and  0.500 analytical estimates of H1, a1 as a function of the NSC std 12 3 50µeV Mc = 6.6 10− (1 + Φ)r M , (34) parameters and TRH. Therefore, assuming that the over- × mA densities become frozen when H1 = mA(T1) we can use where again we are assuming that αtot is chosen so that the same rationale as in the standard case to compute the cold axions explain all of the CDM observed. density, mass and radius of the minicluster. We will drop Radius of the minicluster.- Given the results for the the assumption on the next subsection but we advance mass and the density of the minicluster, one estimates that it is actually quite reasonable. the radius of a minicluster as The typical minicluster density derived in  1/3 Eqs. (30), (31), and (32) does not depend at all on std 3Mc r aeq the NSC. These equations only depend on the as- Rc = = 1/3 , (35) 4πρc (140) Φ H1a1 sumption that axions account for all the CDM and which for our estimates in Eqs. (32) and (34) gives parameterise the effects of an early/late collapse with an overdensity parameter Φ.  0.1668 std 12 r 50µeV The typical minicluster mass computed in Eq. (33) also Rc = 6.4 10 cm , (36) × Φ mA depends on the axion CDM assumption but involves the 11

nation models, while the factor (3 + γ)/(3 + αγ) re- 10 10 mains reasonably close to unity. The origin of this − 1013 complicated exponent is nevertheless clear. Recall that 1/β 10 12 H1 (aRH/a1) during the NSC epoch and there- − ∝ (1 β)/β fore 1/(H1a1) (a1/aRH) − besides other less im- RH 1012 ∝ 14 T portant factors of aRH. Also, using Eq. (16) we ob-

] 10−

tain the first part of the exponent since a1/aRH = std (3+γ)/(3+γα) M [cm]

[ (T /T ) . Since 1/β > 1 for the NSCs 16 RH 1 c c 10 − 1011 considered here (and indeed for all additional fluids with R M Too low a non-tachyonic equation of state wφ > 1), the effect of 18 − 10− increasing H1 (reducing the coherence length or patch of Standard RD 1010 the typical minicluster) is stronger than the mere larger 20 Kination 10− density from projecting ρCDM earlier in time. Matter decay The fact that the quantities 1/β 1 and H are the Kination decay − 1 22 largest among the cosmologies considered explain why in 10− 100 101 102 103 104 the Kination cosmologies the minicluster mass and radius TRH[MeV] can attain such low values shown in Fig.5. The small difference between a red-shifting, non- FIG. 5. The mass (left vertical axis) of a typical axion mini- decaying Kination field and a decaying one is due to cluster (r = 1, Φ = 1) in units of solar masses as a function the slower decrease of the temperature with scale fac- of the reheat temperature for different cosmological models tor, due to the entropy production. Indeed, entropy KD K before nucleosynthesis. We also report the radius of the mini- production is the reason why a1 < a1 (see discus- K KD cluster in cm (vertical right axis). We have assumed that sion around (18)) and thus (H1a1) < (H1a1) because 100% of the CDM is in axions and Φ = 1. Each value of the 1/β 1 2 H a 1/a − = 1/a for both kination types. This reheating temperature requires a different axion mass from 1 1 1 1 effect is∝ captured by the small difference in the exponent Fig.4. Solid and dashed lines assume αtot = 10 or αtot = 1, respectively. in Eq. (37) between the two cases, which contains the factor (3 + γ)/(3 + γα) which is 7/6 for KD cosmology while it is equal to one for Kination∼ cosmology. typical size of an inhomogeneity at T1, given by a radius The most relevant parameter in determining the mini- R H . We note that the mass and the radius of the cluster mass turns out to be the equation of state of the 1 ∼ 1 minicluster depend on the quantity 1/ (H1a1). We have component dominating the expansion at T1, rather than whether entropy is conserved or not. Since 1/β 1 > 0 seen that, insisting on the axion CDM hypothesis, the − value of a1 in the NSC is smaller than the standard value for all the NSCs, this seems to suggest that the mini- unless no extra entropy is generated (α = 1), see Eq. (18). cluster mass increases with the reheating temperature. This effect points towards more massive miniclusters due Indeed, this happens for 3 to a larger DM density ρA ρ /a . However, in our ∝ CDM 1 1 γ(1 α) NSCs, the causal horizon at a1 is also smaller than the β < + − , (38) standard results, so that H1 is larger, due to the presence 2 6 + γ(1 + α) of extra energy and the value of a1 computed at earlier times. It is not straightforward to anticipate which of the which is satisfied in the case of Kination cosmology two effects is the prominent one, due to the complexity in (β = 1/3), but not for the MD cosmology (β = 2/3), the expressions brought in by assuming that the axion is which is an exception to the main trend. In MD, the value 1/β 1 = 1/2 is so small that the other term the CDM particle. Using the numerical results developed − in the previous section we have computed the values of ( 1) in the exponent in Eq. (37) dominates and changes the− trend: the minicluster mass decreases with increas- the typical minicluster mass Mc and radius Rc in the NSCs considered, as shown in Fig.5. ing TRH, although the maximal value of the minicluster 3 mass in MD is only one order of magnitude larger than Notice that Mc 1/(H a ) , so that we use the ex- ∝ 1 1 the standard results. pressions for H in Eq. (22) and for a1 in Eq. (16) to derive a simple analytical formula for the mass of a minicluster. Neglecting the variation in the number of degrees of free- dom as in the previous section, we obtain 2. Structure formation during the NSC period

3 1−β 3+γ 1   ( β 3+αγ ) In this subsection we consider whether some std TRH − Mc = M . (37) c T std gravitationally-driven structure formation could already 1 happen at very early times, during the NSC period. If the In the NSCs, the most important contributions in the axion is the CDM particle, the energy density in the axion exponent turn out to come from the term (1 β)/β, field at the onset of oscillations is extremely small com- which varies from 1/2 to 2 respectively for MD− or Ki- pared to the background radiation, typically by a factor 12

10 aeq/a1 10 , see also AppendixC. Therefore, the evo- The growth of axion CDM perturbations at sub- lution of∼ axion density perturbations is mostly sourced by horizon scales sourced by the background perturbations perturbations of the background fields (φ and radiation), in the Newtonian limit can be described by [171] rather than by axion perturbations themselves.   m˙ 3 2 Let us first briefly discuss the standard case of radia- δ¨A + 2H + δ˙A H δφ = 0, (40) tion domination together with the non-standard Kination m − 2 domination. We will adhere to a simple Newtonian pic- in which the termm/m ˙ is required from momentum con- ture valid for sub-horizon perturbations, as super-horizon servation in the event of a time-varying mass. Since the perturbations change at most by factors of (1). Let us O axion mass depends on temperature due to the instanton consider the Fourier modes of the density fluctuations effects around the QCD phase transition, see Eq. (13), of the background density δφ = ρφ/ρ¯φ 1 (a comoving − the termm/m ˙ αγH turns out to be extremely impor- wavenumber k labelling modes is implicit). Small fluctu- tant for correctly≈ describing perturbations in the axion ations in the linear regime evolve according to field, since it is of the same order as the term 2H. In

 2 2  particular, the time-dependence of the axion mass plays ¨ ˙ csk 4πρ¯φ a relatively significant in the case of the Kination cos- δφ + 2Hδφ + 2 2 δφ = 0. (39) a − mPl mology. If the background cosmology is described by either the where the sound-speed is essentially given by the equa- 2 radiation or the Kination scenarios, the density fluctua- tion of state (cs wφ). Perturbations with k above tions given by Eq. (39) are oscillating and decaying, so ≈ 2
1/2 the Jeans momentum kJ /a = 4πρ¯φ/cs oscillate with that their role in Eq. (40) can be neglected. For this, damped amplitude while those below it can suffer grav- setting δφ 0 in Eq. (40) gives a constant solution ' itational amplification leading to collapse. Using the δA const, plus a time-dependent solution ∼ Friedman equation, the latter term in Eq. (39) can be ( 1 β(2+αγ) written as 4πρ¯ /m2 = 3H2/2. In both the cases of t − , β(2 + αγ) = 1 φ Pl δA 6 (41) radiation- or Kination-dominated scenarios, the Jeans ∝ log t, β(2 + αγ) = 1. momentum decreases in time and once a perturbation en- ters the horizon, k/a > H, it begins oscillating with an In either the radiation-dominated or the Kination- amplitude 1/√a. The situation is completely differ- dominated scenarios, the argument of the power-law is ent in a matter-dominated∝ background. In the ultra-cold smaller than zero so that no growing mode exists. This is a direct implication of the inclusion of the termm/m ˙ , limit, for which wφ 0, perturbations start growing as → which is non-zero until the axion mass stops growing δφ a as soon as they enter the horizon. The extent ∝ around temperature TΛ, so that its contribution is sub- up to which our background perturbations can source > the evolution of axion perturbations is determined by: stantial whenever TRH TΛ in Kination scenarios. Had ∼ 1) their absolute size and 2) the speed of the universe we omitted the termm/m ˙ , which is equivalent to setting γ 0, we would have obtained the well-known loga- expansion. → The size of the background perturbations at the small rithmic growth of matter perturbations during radiation scales of our interest is relatively unconstrained. Obser- domination or the not-so well known linear growth of CDM perturbations during Kination δ t1/3 a, vations of the CMB anisotropies, Lyman-α forest, large- CDM ∝ ∼ scale structure and other cosmological probes measure which is valid in WIMP CDM scenarios [150]. the dimensionless power spectrum of curvature pertur- Indeed, the authors of Ref. [150] argue that the lin- 2 2 9 1 ear growth is essentially due to the free-streaming of bations ∆ δ 10− in the range 1 Gpc− < 1∝ | | ∼ CDM particles after the kick received by the gravitational k < 3 Mpc− [160–162]. The absence of spectral dis-∼ field upon horizon crossing. The comoving distance free- tortions∼ of the CMB Planck distribution can be used to 2 5 1 streamed by such particles has exactly the same time constrain ∆ < 10 up to k < (100 pc)− [163–166]. − dependence, The effects on∼ the yields of Deuterium∼ and Helium from Z Z αγ+1 BBN can also be used to extend this bound even further, dt pi ai  dt 1 β(2+αγ) 2 1 1 d = vi t − (42) ∆ < 0.007 for (100 pc)− < k < (10 pc)− [167]. The ti a ∼ ti mai a a ∼ absence of effects of primordial∼ black∼ holes has also been 2 < where we used that momentum redshifts as pi/a, the ax- used to constraint ∆ 0.01 0.1 above that range [168]. γ αγ β ∼ − ion mass grows as mA 1/T a and a t . In Additional, model-dependent results [169, 170] constrain ∝ ∝ ∝ 2 7 summary, as long as the axion mass continues to increase, the dimensionless power spectrum down to ∆ < 10− 1 1 axion CDM perturbations do not grow during radiation-, in the range 1 Mpc− < k < 0.1 pc− if the∼ DM is made of WIMPs which∼ would∼ form ultracompact mini- KD-, or Kination-dominated cosmologies. Once the ax- ion mass reaches a constant value, we recover the stan- haloes where annihilations and the subsequent emission 4 of gamma-rays could have been detected. Needless to dard log a growth valid during radiation domination . say, this latter case does not affect the scenario under discussion here, since we do not consider WIMP as the DM candidate in this work. 4 The fact that axion CDM perturbations do not grow during the 13

The situation can be entirely different in an early pe- be parametrised as [95] riod of matter domination. In the following, we re-  3 < fer to the non-relativistic matter dominating the expan- 3 2 0.03(k/k1) , k k1 k δA(k)  sion rate of the universe as Dominating Decaying Matter ∆2(k) = h| | i 1, k ∼ (10 100)k a 2π2 V ' ∼ ∈ − 1 (DDM). During such a period, the DDM background per- (k /k)ζ , k k , turbations grow linearly and undergoes gravitational col- 1  1 (44) lapse approximately when becoming non-linear. Solving where V is the volume and ζ 0.5 but probably decreas- Eq. (40) for the axion perturbations when the growing ing with k. The characteristic∼ comoving wavenumber k solution of Eq. (39) is considered gives 1 is defined as the wavenumber entering the horizon at a1, 2/3 δA δφ t . (43) ∝ ∝ k1 = H1a1. (45) Indeed, the situation is beautifully similar to the growth Although this spectrum was derived from simulations in a of baryon perturbations after recombination in standard radiation-dominated background, we take it as rough rep- ΛCDM cosmology. Before recombination, the baryon resentative of the axion CDM initial conditions around fluid is tightly coupled to the photon fluid. Because of the a1 also for NSC cosmologies, since numerical results in a large effective sound-speed, all relevant modes are above matter-dominated scenario are not yet available. the Jeans wavenumber and baryons cannot clump. How- The fact that standard axion miniclusters with over- ever, (sub-horizon) CDM perturbations grow linearly as densities Φ O(1) form with a typical mass of 3 ∼ ∼ soon as the universe becomes matter-dominated. When ρA/(k1/a1) is related to the power spectrum decreas- the temperature drops enough for most protons to form ing very sharply at scales k < k1, i.e. length scales neutral Hydrogen, baryons effectively decouple from the larger than 1/H1 at a1. At small length-scales, k > k1, CMB, feeling a sudden drop of pressure, which allows fluctuations are already non-linear and one thus expects them to fall rapidly into the already deep potential wells the formation of ensuing “small” miniclusters potentially dug by CDM. denser, but the largest physical objects are determined Analogously, axion fluctuations cannot efficiently grow by k k1, which is a consequence of the low-k cut-off 2 ∼ 3 at early times because the axion mass is so small that ∆A k . axions are relativistic or frozen due to the Hubble fric- In∝ a MD period, fluctuations in the axion field grow tion. Meanwhile, DDM field perturbations start to grow even larger than the estimates in Sec. IV B 1 if the pertur- as soon as they enter the horizon during matter domi- bations in the DDM are large enough, i.e. δφ > δA(t1). nation. When axions abruptly start to behave as CDM, In particular, in order to change the typical miniclus- they would immediately fall into the potential wells of the ter size, they should be able to shift down the typical 2 additional DDM. In the analogy depicted, axions play the wavenumber at which ∆A O(1). In the following, we role of baryons and DDM that of the standard DM after will therefore concentrate in∼ modes with wave numbers recombination. Sadly, the analogy stops here: firstly, the smaller than k1, axion density field already possesses large fluctuations of its own which are typically not correlated with those k < k1, (46) of the background field; secondly, DDM is decaying, so it stops being relevant for the overall expansion of the which are still linear at a1. If the DDM perturbations are adiabatic perturbations from a vanilla slow-roll infla- universe at aRH and it would be completely absent af- 1 ton, we could extrapolate the measured perturbations ter t Γ− . Axions dragged into DDM over-densities φ 2 9 ns 1  at CMB scales ∆ 10− (k/k ) − with ns would suddenly find themselves left alone, freed from the ∼ CMB ' gravitational pull of the DDM. If the DDM perturbations 0.968(8) [4]. These fluctuations are so small that they would not affect the typical size of axion miniclusters in were linear, axions just continue their free-streaming in- 6 wards of the former DDM perturbations, but if the DDM NSCs . Large adiabatic fluctuations are not entirely un- perturbations had already become non-linear the axions expected in exotic models of inflation, where they can would free-stream away from them. be originated from features in the inflaton potential [172] To understand qualitatively the impact on miniclus- or inflection points [173]. They can also originate from ters, recall that soon after the onset of oscillations, the particle production [174], waterfall transitions [175], and spectrum of axion DM fluctuations at small scales5 can

inflationary cosmology, we expect an uncorrelated additive con- tribution from the adiabatic temperature fluctuations imprinted Kination stage facilitates the treatment for axion CDM, since the in the plasma by the inflaton fluctuations. The temperature fluc- growth of perturbations in Kination cosmology is not understood tuations get imprinted as fluctuations in the axion DM density beyond the linear regime [150], and gravitational collapse has due to the later’s dependence on T1. never been proven. 6 They could affect substructure and the size of large clusters of 5 These are the fluctuations of the axion DM field at small scales miniclusters, though, but these do not have clear observational due to the misalignment and string emission contribution. In the consequences yet studied. 14 other exotic phenomena, see e.g. Refs. [176, 177]. In large as to produce too many black-holes). A more con- 2 5 practice, the amplitude at very small scales is obser- servative value ∆φ(ak) 10− yields a more modest en- 2 ∼ vationally constrained mostly from the requirement to hancement Mc0 10 Mc,0 and only for the most extreme avoid an overproduction of primordial black holes. values of the reheat∼ temperature. In this scenario, struc- Assume that DDM has an adiabatic spectrum of am- tures have time to become strongly non-linear above k0 2 1 plitude consistent with the PBH limits ∆φ 0.01. The so these it is likely that these typical miniclusters have a ∼ DDM density fluctuation δφ(k) entering the horizon at lot of substructures. ak (a1, aRH) experiences linear growth so that the It is tempting to think that the substructures of the power∈ spectrum at later times is large Mc0-mass miniclusters are something like the pre-  2  2  4 viously discussed Mc-mass objects (Mc given by H1). 2 2 a a k ∆φ(a) ∆φ(ak) 0.01 , (47) However, we think that this would not be the case. The ∼ ak ∼ a1 k1 axion CDM inhomogeneities are so small compared to the background that they play no dynamical role. The where we used the relation H = H (a /a)3/2 valid for 1 1 driving perturbations are those of the DDM, which corre- MD cosmology which leads to a /a = (k /k)2. For ax- k 1 1 spond to slightly (or not so-slightly) different DDM den- ion perturbations that are linear at a , the largest possi- 1 sities which grow and collapse into DDM halos that pre- ble growth happens for those that enter the the horizon 2 sumably merge hierarchically from small to large scales. earliest, i.e. k k1, for which ∆φ grows by a factor 2 ∼ The axion overdensities that would have later lead to Mc (aRH/a1) . This quantity is determined by our assump- miniclusters just follow the DDM gravitational potential. tion of having the totality of CDM in axions. In the MD Since these inhomogeneities have negligible self-gravity, scenario we have they would probably be disrupted inside DDM halos af- 1 28/9 ter a dynamical time scale H− once DDM becomes a 2 T std  ∼ RH = 1 108, (48) non-linear. a1 TRH ∼ The comoving scales that can be significantly affected by the NSC period are bounded from below by the mode where the numerical factor has been estimated for the that enters the horizon at the minimum reheating tem- choice TRH = 5 MeV and αtot O(1). ∼ perature, TRH 5 MeV, In the modified MD scenario described, the mass of ∼ the typical miniclusters would be given by the size of the 1 k > kmin TRH = HRH aRH T =5 MeV (20pc)− . (51) largest DDM fluctuation that became O(1), dragging { } | RH ∼ CDM axions with it. Here, we refer∼ to the associated momentum k10 for its similar role to k1 in setting the value of the minicluster mass, without any relation to 3. Observational parameters Eq. (1). The maximum effect takes place at a = aRH. 2 Solving for k10 in the expression ∆φ(aRH, k10 ) = 1, (47), Velocity dispersion- The clustering of axion CDM has we obtain observational consequences, since detectors would be 1   2 triggered by a larger energy density when the Earth a1 1/4 2
− k10 k1 ∆φ(ak) passes through one such substructures. In fact, the lo- ∼ aRH ∼ cal energy density of CDM is estimated to be ρ 3 ≈ 28/36 !1/4 0.4 GeV/cm , see Ref. [178], while the density obtained  T  0.01 RH for the minicluster is given in Eq. (31). The density en- 3k1 std 2 , (49) ∼ T1 ∆φ(ak) hancement during the encounter with a minicluster can then reach which can be up to a factor of 30 smaller than k1 for 2 ∼ std the largest values of ∆φ(ak) 0.01 and T1 2 GeV. ρc 7 3 ∼2 ∼ 1.7 10 (1 + Φ)Φ . (52) Note that k0 > HRHaRH if ∆φ(ak) is small enough so ρ ' × 1 that the black-hole constraint is still respected. In this 1 case, the typical minicluster has a size R0 (k0 /aRH)− The velocity dispersion of the axions in the minicluster 1 ∼ 1 2 at aRH, and thus a mass, δv is directly related to the coherence time of the axion field tcoh during such encounter. The latter quantity is  3  3 4π 1 k1 a key parameter in an axion search experiment. Both Mc0 = (1 + Φφ)ρCDM = Mc,0 , (50) these quantities can be computed from the gravitational 3 k0 k0 1 1 potential by the virial theorem where Mc,0 is the minicluster mass in the NSC if DDM fluctuations are negligible δφ 0. Summing up, mini- 2 Mc ∼ δv 2 . (53) clusters formed during MD and hosted in DDM pertur- ∼ mPlRc bations can be a factor up to 30 times larger and 303 heavier due to early structure∼ formation during the∼ NSC. In the standard cosmological model, once the deviation However, this requires large DDM fluctuations (not as from the size of the horizon r = R H = (1) is intro- 1 1 O 15 duced, this gives Given a galactic year τ = 2πr /v 230 My, the time between two encounters can be estimated ∼ as,  0.3336 2 19 2 50µeV δvstd 2 10− r Φ(1 + Φ) , (54)  0.1668 ∼ × mA std τ 5 50µeV T = 4 10 years . (61) 0.666 btw std   Nenc ∼ × mA std 1 8 1 50µeV δtcoh 2 = 0.9 10 s 2 (55), ∼ mA δv × r Φ(1 + Φ) mA During the encounter, the energy density in the mini- where in Eq. (55) we have assumed that the axion mass cluster is enhanced by the factor given in Eq. (52). Given the size of the minicluster in the standard scenario in is equal to mCDM. As we discussed in Sec. IV B 1, the mass and the radius of a minicluster computed in a NSC Eq. (36) and the velocity of the Solar System around the Galactic centre v 230 km/s, which coincides with typ- are modified by the same parameter 1/ (H1a1). Thus, ∼ the ratio of the velocity dispersions and the ratio of the ical virial velocities in the DM halo, the typical encounter coherence time when computed in the NSC and in the has a duration standard cosmology read std  0.1668 std 2Rc r mA 1−β 3+γ ∆tenc = 6 days , (62) 2  2/3  2( β 3+αγ 1) v ∼ Φ 50µeV δv Mc TRH − 2 std = std , (56) δvstd ' Mc T1 where we have assumed that the relative velocity between 1+β 3+γ  3 β 3+γα the Solar System and the minicluster is of the order of v . δt T − coh RH , (57) According to Eq. (55), the axion field is coherent during δtstd ' T std coh 1 the whole encounter, since δtcoh ∆tenc. The situation for standard miniclusters is a bit unfortunate. Although where the computations for δt involves using m coh CDM the enhancement of the signal in an axion search exper- from Eq. (25). Remarkably, the coherence time in the iment during an encounter would be extraordinary, the NSCs we have considered seems to be always longer than collision rates are so small that it seems quite hopeless in the standard scenario. Referring to Eq. (55), the ax- to expect a minicluster encounter. ion mass in both the Kination and KD scenarios is larger than in the standard case, however this trend is counter- Let us now turn our attention to the case of an early balanced by the large decrease in the minicluster mass NSC. In the Kination and KD cosmologies, the typical minicluster is generally lighter and smaller than in the which lowers the velocity dispersion. On the contrary, in std std the MD scenario the increase in the velocity dispersion standard cosmology, Mc < Mc and Rc < Rc . Ac- due to the more massive miniclusters in this scenario is cording to Eq. (58), we thus expect a larger number of modest, while the axion mass would be lower by orders of miniclusters in the halo, each with a smaller encounter magnitude with respect to the standard value. Our nu- probability with the Earth. On the contrary, miniclusters merical calculations for the velocity dispersion and the formed during a MD cosmology are slightly more massive coherence time are shown in Fig.6. and larger than what obtained in the standard scenario. We estimate the number of encounters of a minicluster Encounter rates with the Earth.- Assuming the DM 12 with the Earth Nenc and the average lapse time between mass of the Milky Way as MMW 10 M , the number two encounters in a NSC by using the results in Eqs. (59) of miniclusters in the halo in the∼ standard scenario is, and (60) and the expression in Eq. (61), to obtain  0.50 std MMW 23 mA   1  2 std N = 10 , (58) Nenc Tbtw − Rc Mc c std = Mc ∼ 50µeV std std std Nenc ' Tbtw ∝ Rc Mc 3 1 1−β 3+γ assuming 1/(r (1 + Φ)) 1 on average. Since we take Rstd  T  − β 3+αγ ' c RH most of the CDM in the form of miniclusters, the local = std , (63) Rc ∝ T number density of axion miniclusters is then 1 where in the last expression we used Eq. (37). Chances  0.50 std ρ 9 mA 3 to encounter a minicluster halo are then higher for mini- n 10 pc . (59) c std − clusters forming during a Kination NSC, with a smaller ' Mc ∼ 50µeV lapse time Tbtw between two subsequent encounters. The During a complete revolution around the galactic halo, drawback of the scenario lies in the smaller time for which the Solar System transverses a length l = 2π r , where the encounter lasts, which in the standard scenario is r = 8.3 kpc is the distance of the Solar System from given by Eq. (62) and for a NSC gives the galactic centre. On this path, the Earth encounters 1−β 3+γ   β 3+αγ 1 a number of miniclusters equal to, ∆tenc Rc TRH − std std std . (64)  0.1668 ∆tenc ' Rc ∝ T1 std std 2 std mA Nenc = (2π r )(π (Rc ) ) nc 600 . ∼ 50µeV In TableI we have summarised the results for the relevant (60) astrophysical quantities of a typical axion minicluster as 16 obtained in this Section for the NSCs we have consid- 100 101 102 103 104 ered. We have computed the quantities presented in this 20 Section by assuming that all of the axions clump into 10− miniclusters structures and that axions make up the to- 22 tality of the CDM budget. In the following, we have set 10− 2 for convenience TMeV = TRH/ MeV, and we have fixed RH 24 T δv 10 αtot = 10 to account for the relative contribution to the − axion energy density from the decay of topological de- Kination 26 fects. We have also set for simplicity r = 1, and we 10− Matter decay Kination decay

have considered miniclusters that form with an overden- Too low 282 sity Φ = 1. For these reasons, the results only depend on 10−10 106 the reheat temperature TRH. The mass of the axion for 101 which we have 100% CDM, here mCDM, differs by various 105 orders of magnitude among the different cosmologies, as 0 first noted in Ref. [107]. Likewise, the mass and size for a 10 RH T 4 [yrs] [days] 10 minicluster in either the standard and MD scenarios can 1 enc 10− btw t have similar ranges. 103 T ∆ 2 10−

Too low 102 39 1010− Minicluster streams.- Tidal axionic streams [87, 179] form by the encounter 108 and the subsequent disruption of a minicluster with 7 nearby stars in the disc or by the gravitational field in 10 RH T the halo, similarly to other types of clumps that might [days] 106 have formed in the early universe [180]. These structures coh might be crucial for the direct detection of axions in that t 105 the energy density, while diminished with respect to what Too low is attained in a minicluster produced in the standard cos- 104 mological picture, can nevertheless be larger than the av- 100 101 102 103 104 erage DM density ρ . TRH[GeV] In the standard picture, axion miniclusters are too dense to be disrupted by the gravitational field of the FIG. 6. Top panel: the velocity dispersion of a minicluster as halo, and the main process for their disruption is by tidal a function of the reheat temperature for different cosmological stripping after an encounter with a star. Defining a crit- models before nucleosynthesis. We have assumed that 100% ical impact parameter bc as [181, 182] of the CDM is in axions with αtot = 10. Colour coding is the same as in Fig.5. Middle panel: typical duration in days s of a single encounter of a minicluster with the Earth. The MsRc bc 2 , (65) right vertical axis gives the mean time interval (in years) be- ≡ mPl vrelvc tween two Earth encounters with a minicluster with the same settings as in the Top panel. Bottom panel: coherence time any minicluster whose impact parameter with a nearby of the axion field in the minicluster. Note that it is always star is b < bc would be disrupted after a single encounter. sensibly larger than the time of an encounter. In Eq. (65), we have defined the velocity dispersion of 2 the minicluster vc = √δv from Eq. (53) and vrel is the velocity of the minicluster relative to the star of mass the relation between the mass and the radius of a mini- Ms. Given a column density of stars Σ = dMs/d for cluster as in Eq. (32). Notice that the disruption prob- a given area in the direction orthogonal⊥ to theA disc, A ability depends on the density of the minicluster but the total probability of encounter over the period ∆tobs not on Mc or Rc separately, so pdisr is the same for is then any cosmological model used. This because in larger r miniclusters, the escape velocity of bound axions is also 2πRc(4Σ )∆tobs 8πΣ ∆tobs 4π pdisr(Φ) = ⊥ = ⊥ , (66) larger by the same factor when the density of the ob- m2 v v m v 3ρ Pl rel c Pl rel c ject is constant. Using the expression in Eq. (32) for the density of a minicluster and the estimate for the col- where the additional factor of four in (4Σ ) has been 2 ⊥ umn density Σ (35 5) M pc− [183] gives pdisr = estimated in Ref. [87] from the averaging over all di- 1/⊥2 ≈ 3/2 ± rections, and the extra factor of two comes from sum- 0.1%(1 + Φ)− Φ− per galactic year. ming up all non-disruptive encounters at impact radii Owing to the disruption of the minicluster, an en- b > bc [181]. For obtaining the result, we have used hanced dark matter density ρstream(Φ, t) with respect to 17

Scenario STD MD Kination (no decay) Kination (w/ decay) α 1 3/8 1 3/4 β 1/2 2/3 1/3 1/3 −1 4/3 4 −1 4 −1 mCDM (µeV) 100 10 TMeV 10 TMeV 10 TMeV 5/12 1/8 T1 (MeV) 1500 70 TMeV 1500 600 TMeV −11 −9 −2/3 −21 3 −24 4 Mc (M ) 10 10 TMeV 10 TMeV 10 TMeV −2/9 −4 −5 4/3 Rc (AU) 0.5 TMeV 10 TMeV 3 × 10 TMeV Enhancement 107 107 107 107 7 10 −8/9 11 −1 13 −5/3 δtcoh (s) 3 × 10 10 TMeV 10 TMeV 10 TMeV −2/9 −3 −4 4/3 ∆tenc (days) 5.5 28 TMeV 4 × 10 TMeV 3 × 10 TMeV 2 2/9 6 −1 7 −4/3 Nenc 500 10 TMeV 10 TMeV 10 TMeV 5 6 −2/9 4/3 Tbtw (yr) 4 × 10 10 TMeV 300TMeV 20TMeV

TABLE I. The parameters α and β describing the various pre-BBN cosmologies: Standard radiation-dominated (STD), matter- dominated low-reheat temperature (MD) cosmology, Kination respectively without or with (KD) the decay of the φ field considered. For each cosmology, we provide the value of the relevant quantities describing the structure of the axion minicluster and the details of encounter with the Earth, setting αtot = 10, Φ = 1, and r = 1. We have defined TMeV = TRH/ MeV. mCDM is the value of the axion mass for which the axion is the CDM particle in the specific cosmology considered, in which the axion field begin the coherent oscillations at temperature T1. Rc and Mc are respectively the radius and the mass of the minicluster. The local CDM energy density is enhanced by the quantity under “Enhancement” by the presence of the minicluster. The encounter of the Earth with an axion minicluster would last ∆tenc days, with a period between two encounters given by Tbtw. Nenc is the number of miniclusters encountered by one galactic revolution.

the background valueρ ¯A is expected. Axions that have standard cosmological history with a black line, and we been stripped away from the bound orbits of the mini- have included the contributions from other NSC with cluster stream away, traveling a distance L = vct in a different colour codings: MD (red dashed line), Kina- time t and with a density of the stream that drops lin- tion (blue dotted line) and KD (green dot-dashed line). early with time as [87] Thicker lines correspond to larger reheat temperatures in TRH = (5, 50, 500) MeV. The result in the standard cos- Rc 7 GeV 1/2 3/2 τG mology is smaller than what obtained in Ref. [87] because ρstream(Φ, t)=ρc 10 3 (1 + Φ) Φ , (67) vct ≈ cm tG of the more massive miniclusters we used in our model, std 11 which again is independent on the cosmological model. Mc 10− M , consistently with the analysis in the previous∼ sections. For the same density, the miniclusters We have introduced the characteristic time τG = we obtain in our analysis are thus slightly larger than 35500 yr, while tG 12 Gyr is the age of the galac- tic disc. The enhancement∼ E of the axion energy den- what used in previous analyses, and it takes more time sity due to tidal stream then ranges from a minimum to transverse them affecting the denominator in Ref. (68). This reasoning also explains the trend observed in Fig.7 Em(Φ) = max (E, ρstream(Φ, t)/ρ¯A) to a maximum value for NSC scenarios, for which a more massive and larger EM = ρc/ρ¯A. Following closely Eq. (4.5) in Ref. [87], we obtain the rate of stream encounters minicluster is realised in the MD cosmology while smaller miniclusters are obtained in Kination and KD cosmolo- Z + Z EM ∞ dE0 pdisr(Φ) f(Φ) ρstream(Φ, t) gies. N(E) = dΦ 3 . 0 Em(Φ) (E0) 2Rc/vrel ρ¯A (68) Here, we have introduced the mass fraction in miniclus- V. DISCUSSION AND CONCLUSIONS ters, first estimated in Ref. [184], although we note that the refined calculations of Ref. [95] seem to point to a In this paper, we have discussed the properties of axion much smaller density of ultra-dense miniclusters. miniclusters emerging in different cosmological scenar- In Fig.7, we show the result obtained from comput- ios before Big Bang Nucleosynthesis (BBN) took place. ing the number of encounters N(E) in Eq. (68) that In particular, we have considered different scenarios in yield to an enhancement in density larger than E, plot- which the cosmology before BBN was governed by either ted as a function of E. We have fixed the observa- i) a matter component, ii) a fast-rolling field φ leading tion period ∆tobs = 20 years and we have assumed that to a Kination period, or iii) a decaying Kination field φ. strings contribute to the total axion energy density so Using assumptions commonly made in the literature, we that αtot = 10. We show the results obtained in the have obtained the mass and size of the minicluster, as 18

�� clusters in the MD cosmology can have a mass is up to �� two orders of magnitude larger than standard (radius up to 5 larger), while in the Kination and KD scenarios �� �� the∼ mass can be up to a factor 109 smaller than standard (with a radius up to 103 times smaller). The miniclusters ���� obtained when considering the Kination or the KD cos- mologies are lighter and more compact, thus making it more frequent for the Earth to come into the vicinity of �� std these objects. As we obtained in Fig.4, when TRH T1 the axion field starts to oscillate in the standard scenario≥ � and we recover the standard results. When we assume that all of the DM is in the form of ��-�� axions, the typical minicluster density is set by the DM ��� 3 density at matter-radiation equality, Mc/R ρeq, and ��� c ∼ ��-�� does not depend on the early cosmology within our sim- ��� plified picture. The minicluster mass and radius however �� -�� can be very different from standard cosmology as they �� �� �� are set by the size of the horizon when the axion field � �� �� �� begins to oscillate and becomes non-relativistic. The as- trophysical quantities of relevance for detection tend to depend on different combinations of Mc and Rc and can be very different from the standard scenario: the veloc- p FIG. 7. The number of encounters N(E) with an axion stream ity dispersion δv Mc/Rc, the time between encoun- ∝ 2 produced from the tidal stripping of a minicluster, leading to ters with the Earth T Mc/R and the duration of btw ∝ c an enhancement larger than E = ρ/ρ¯A. We have assumed an an encounter ∆t Rc are different in non-standard enc ∝ observation interval ∆tobs = 20 years and a contribution from cosmologies for different values of the reheating temper- topological defects so that αtot = 10. The black solid line ature. In Fig.6 (Top panel) we show the typical dura- represents the result obtained in the standard cosmological tion of a minicluster encounter with the Earth (left ver- scenario, while the other NSC scenarios are plotted with dif- tical axis) and the expected time interval between two ferent colours and dashing: MD (red dashed line), Kination (blue dotted line) and KD (green dot-dashed line). Thicker consecutive encounters (right vertical axis) as a func- lines correspond to larger reheat temperatures, corresponding tion of TRH, as well as the velocity dispersion squared to TRH = (5, 50, 500) MeV. (Bottom panel) for the cosmological model studied. For std TRH T1 , both modified cosmologies show detection advantages≤ and disadvantages compared to the standard well as the enhancement in axion density over the local result. If the axion starts oscillating in a Kination model, CDM background, in different cosmological setups. the encounter would only last up to a few minutes ow- We have sketched the results for the relative quantities ing to the small size of the minicluster itself; on the other describing miniclusters in more detail in Fig.5 as a func- hand, the frequency of encounter in the Kination cosmol- 3 tion of the temperature TRH at which the modified cos- ogy can be enhanced by an O(10 ) factor with respect to mology transitions to the standard radiation-dominated the standard case, with the encounters possibly being as scenario. In the Figure, we show the mass of the mini- frequent as one per a few years. On the contrary, for an cluster in the case whether the early cosmological sce- axion field that begins to oscillate in a matter-dominated scenario, the encounter would last up to 50 days, al- nario is standard (black lines), matter-dominated (red ≈ lines), Kination without (blue lines) and with the decay though one such encounter during a Galactic year would of the φ field (green lines). Solid and dashed lines assume be much more rare. For an axion minicluster forming in αtot = 10 or αtot = 1, respectively. In order to produce the standard cosmology, the velocity dispersion is small the figure, we fix the relic abundance to the present CDM enough so that the coherence time of the axion field is abundance, so that for each value of TRH the axion mass much longer than the duration of a minicluster encounter is given by Eq. (21). We have cut the plots at the value with the Earth. In any modified cosmology we study, the of TRH for which the axion mass exceeds the bound from coherence time modifies but not as much as to invalidate the astrophysical considerations or the minimum reheat- the previous statement. ing temperature. In Fig.5, the right vertical axis gives We discuss the dependence of the solution on αtot span- the size of the minicluster, obtained using the fact that ning through various orders of magnitude, since at pres- the minicluster density is constant, see Eq. (35). For ence the effective value of this quantity is uncertain. In std TRH T1 , the mass and the size are steadily smaller Fig.8 we report the density plot showing the mass of the than≤ the standard value for Kination cosmologies, while axion minicluster, in units of M , depending on both

it is higher than what obtained in the standard scenario TRH and αtot. Again, the largest variations in mass are for the matter-dominated model. In more details, mini- shown for the Kination models, for which the mass of the 19

22 8 minicluster ranges between 10− to 10− solar masses Sec. IV B 3 following closely the treatment in Ref. [87] over the allowed range. The range over which the mass and the paper therein in which the issue has been ad- of the minicluster varies is much more contained in the dressed for miniclusters in the standard cosmology. As 11 standard cosmology, for which Mc 10− M , and in for any dark matter micro-halo [181, 182], the disrup- ∼ the MD cosmology for which Mc varies by just two orders tion probability after one passage of an axion miniclus- of magnitude around the standard value. The white re- ter through the Galactic disc is given in Eq. (66), which gion marks the area where the axion mass is excluded by we have shown to be independent on the details of the astrophysical considerations. The dot-dashed line marks NSC at the lowest order of the approximation. This re- the region where T < T std, where the modified cos- sult holds because the disruption probability is approx- RH 1 p mology takes place to the left of the dot-dashed line, and imately pdisr 1/ρc const. In the simplest model std ∝ ≈ the region TRH > T1 where the axion field starts to we have discussed, the probability of disruption is then oscillate in the standard radiation-dominated cosmology, independent on the details of the cosmology and on the for which Mc is given by the value in the standard cos- details of the physics of the axion. The result ps 1,  mological scenario. Overall, the actual value of αtot does valid in the standard scenario, is then expected to hold not change much the general picture. also in modified cosmological histories. We then expect One might question whether an early matter stage, as a sizeable fraction of the dark matter axions to be bound the one the universe experiences in the MD cosmology, into miniclusters even in modified cosmologies, since tidal leads to a pre-BBN growth of the structures since adia- stripping does not seem to provide a mechanism of dis- batic perturbations in the modulus field entering horizon ruption of these sub-structures. In any case, we have in such a modified cosmology would grow linearly with computed the contribution from the tidal stripping ox ax- the scale factor. This has been considered for CDM seeds ion miniclusters to the local energy density in the form in MD models in Refs. [185, 186], and in Kination models of axion streams, by extending the results discussed in in Refs. [147, 150]. Perturbations in the axion dark mat- Ref. [87] to a NSC scenario. As we show in Fig.7, the ter fluid would be dragged into such primordial perturba- number of encounters N(A) with an axion stream for a tions and possibly lead to an early growth. However, such given enhancement A is expected to be sensibly larger model would be depending on the initial power spectrum than in the standard scenario for the early Kination and of the perturbations in the new field. We have partly ad- KD cosmologies. In particular, a number of encounters dressed this issue in Sec. IV B 2 where we have shown that of the order of N 102 to 103 are expected even for an perturbations in the energy density of the axion grow lin- enhancement A ∼10 of the local axion density. early with time only in a MD cosmology, while in Kina- For these reasons,∼ we believe it is worth readapting the tion the growth is suppressed. This behaviour is peculiar existing experimental strategies of detecting axion DM to to axion CDM and differs from the WIMP scenario stud- take into account this broad range of minicluster masses ied in Refs. [147, 150]. We have been able to estimate and radii shown in Fig.5. In the event of a discovery, the largest mass that could grow into a minicluster in the minicluster size distribution could be a window to the MD model due to the large fluctuations in the mas- the cosmology in the still unexplored era prior to big- sive scalar field governing the expansion rate in the NSC bang-nucleosynthesis. δφ, by demanding that fluctuations are not so large as to Note added: During the completion of the present produce an excessive number of primordial black holes. work, Ref. [188] appeared, with their results for an early We have found that the radius of the axion minicluster matter-dominated epoch overlapping with our work. could be enhanced by a factor up to 30, corresponding to an enhancement 303 104
in∼ mass with respect ∼ ≈ O to the case in which fluctuations are suppressed δφ 0. ACKNOWLEDGMENTS In this scenario, miniclusters would then attain a mass' 6 Mc 10− M which is of the same order of magnitude as the∼ mass of the first halos that form in WIMP models The authors would like to thank Pierre Sikivie, Sebas- from collisional damping and free-streaming that erase tian Baum, and Sunny Vagnozzi for the useful discussions density perturbations within that scale7. We leave fur- and comments that led to the present work. L.V. would ther details on the important and interesting issue of an like to thank the University of Zaragoza, where part of early growth of perturbations to a subsequent work this work was conducted, for hospitality. L.V. acknowl- A further aspect which is worth discussing is the edges support by the Vetenskapsr˚adet(Swedish Research eventual survival of axion minicluster from tidal strip- Council) through contract No. 638-2013-8993, the Os- ping. Here, we have dealt with these computations in kar Klein Centre for Cosmoparticle Physics, and support from the NWO Physics Vrij Programme “The Hidden Universe of Weakly Interacting Particles” with project number 680.92.18.03 (NWO Vrije Programma), which is 7 Depending on WIMP-lepton scattering cross-section, the value of (partly) financed by the Dutch Research Council (NWO). the free-streaming length can also vary by orders of magnitude. J.R. is supported by the Ramon y Cajal Fellowship 2012- The WIMP free-streaming length in a NSC has been estimated 10597, the grant FPA2015-65745-P (MINECO/FEDER), in Ref. [187]. the EU through the ITN “Elusives” H2020-MSCA-ITN- 20

FIG. 8. Density plot showing the mass of an axion minicluster, depending on the values of the reheat temperature and the parameter αtot, for different cosmological models before nucleosynthesis. Top left: Standard scenario. Top right: Low-reheat temperature scenario. Bottom left: Kination scenario. Bottom right: Kination scenario with a decaying φ field. The dot-dashed line marks the region where the axion field starts to oscillate in the standard scenario (right side) or in the modified scenario (left side). 21

2015/674896 and the Deutsche Forschungsgemeinschaft axions are produced from the decay of parent par- under grant SFB-1258 as a Mercator Fellow. ticles, since axions as dark radiation do not pile up to the present CDM budget.

Appendix A: Axion Production mechanisms Vacuum realignment • Vacuum realignment, one of the main mechanisms In this section, we review the axion physics and cos- to produce a cold axion population, occurs after mological production. For an excellent introduction to the breaking of the PQ symmetry that sets the the subject we refer to Ref. [189], while thorough reviews axion field at the bottom of a “Mexican hat” po- are found in Refs. [37, 38, 41, 42, 190–193] and in the tential [127, 151, 152]. Axions are massless from appendix to Ref. [194]. the breaking of the PQ symmetry down to temper- Populations of cosmological axions are produced atures of the order of the QCD phase transition, through five main mechanisms: thermalisation [195], the when instanton effects generate an effective axion decay of a parent particle [196–200], vacuum realign- potential [114, 235], ment [127, 151, 152], the decay of topological string de- 4  q  fects [201–207], and wall decay [117, 205, 208–210]. Of ΛA 2 V (θ) = 1 1 4cz sin (θ/2) , (A2) these mechanisms, only the latter three contribute to a cz − − sizeable cold dark matter population. We briefly revise 4 these production methods. where θ = Na/fA is an angular variable, ΛA = (75.5 MeV)4 is the topological susceptibility, and Thermal axions c = z/(1 + z)2 = 0.22 with the ratio of the up • z Thermal axions are produced in the early universe and down quark masses z = mu/md = 0.48. The mainly through the process π + π π + a [211]. square of the axion mass at zero temperature is Similarly to neutrinos, thermal axions→ would con- then [13, 14] tribute the hot dark matter component. For this 1 d2V Λ4 reason, an upper bound mA < 1 eV can be placed 2 A mA 2 2 = 2 . (A3) from the requirement that thermal∼ axions do not ≡ fA dθ θ=0 fA overclose the universe [212–215]. We discuss the vacuum realignment mechanism in Decay of a parent particle a generic cosmological scenario in Sec.B. • A decaying massive particle or a modulus cou- Decay of topological strings pled to the axion field would lead to an incre- • ment of the hot dark matter or dark radiation com- Topological strings are produced because the an- ponents, through the decay of the modulus into gular variable θ takes different values at each spa- two axions. An effective model for the massive tial point after the breaking of the PQ symmetry, modulus would be a low-energy manifestation of through the Kibble mechanism [236]. After pro- a larger theory involving both supersymmetry and duction, the energy density in strings scales with extra dimensions [216–222] like a string theory ax- the string unit length, and the string continuously ion [53, 197, 223–231]. For this reason, dark ra- emit low-frequency modes axions which eventually diation from a string model [197, 198] is able to contribute to the present cold dark matter energy constrain string and M-theory compactification sce- density. The actual emission spectrum is crucial in narios through the change in the effective number determining the present abundance of cold axions, of relativistic degrees of freedom Neff [232], with which is computed in Refs. [201, 202] by using an constraints coming from both the CMB polarisa- energy spectrum with a sharp peak at the horizon tion and big bang nucleosynthesis. In some mod- scale, and in Refs. [203–205] by using a spectrum els, the parent particle is a modulus field which, proportional to the inverse of the axion momentum if it dominates the universe, must decay prior Big 1/q. Results are often expressed in terms of the str mis Bang Nucleosynthesis (BBN) at a reheat tempera- ratio αstr = ρA (t0)/ρA (t0) of the present energy str ture [102–106] density of cold axions from axionic strings ρA (t0) and that from axions produced via the misalign- mis TRH > 5 MeV, (A1) ment mechanism ρA (t0). Refs. [201, 202] report ∼ αstr 200, while Refs. [203–205] report αstr 0.1, in order to avoid the so-called “moduli decay prob- thus∼ the estimation of the CDM axion mass differs∼ lem” [233, 234]. The limit on TRH results from gen- by order of magnitudes in the two models. The eral considerations on the successes of BBN, and it controversy between these different models is solved is then a general lower bound below which the uni- with lattice QCD numerical simulations [206, 207], verse has to be dominantly filled with radiation. which show that the energy spectrum peaks at the Here we do not treat further the possibility that horizon scale and is exponentially suppressed at 22

higher momenta. This method yields an interme- In this last expression, we have written the Laplacian op- diate value α 10. However the recent nu- erator in terms of a rescaled co-moving spatial coordinate str ∼ merical simulations in Refs. [118, 120] find an or- x = H1 a1 x¯. Defining χ = 2 1/(2β), and setting der of magnitude discrepancy with the results in − ψ Refs. [206, 207], showing that a consensus on the θ = χ , (B3) detail on the axion string radiation into a spectrum a of axions has not been reached yet. All of the re- Eq. (B2) is rewritten as sults discussed are valid in a radiation-dominated  2   ψ χ mA 3 ψ cosmology, however the value of αstr also depends 4(1 ) 2 2β ψ00+χ(1 χ) 2 a − ψ+9 a sin χ =0, on the properties of the cosmological model before − a − ∇ m1 a BBN [107–109]. (B4) where a prime indicates a derivation with respect to a. Decay of domain walls The expression above is the generalization of the equation • When the primordial plasma undergoes the QCD of motion for the axion field in any cosmological model, phase transition, the effective axion potential in and reduces to the usual expression in the radiation- Eq. (A2) takes place, showing N minima separated dominated limit β = 1/2, by domain walls attached to strings. Similarly to  2   2 mA 3 ψ what discussed for axions from strings, there has ψ00 ψ + 9 a sin = 0. (B5) been some controversy regarding the spectrum of − ∇ m1 a axion radiated from domain walls. Ref. [208] claims Eq. (B5) coincides with the results in Ref. [59], where the that the energy spectrum peaks around the axion conformal time η is used as the independent variable in mass, while in Refs. [205] a larger axion population place of the scale factor a. We remark that this choice is is obtained by using an emission spectrum propor- possible in the radiation-dominated cosmology because tional to the axion wave number. The evolution of η a, whereas in a generic cosmological model this rela- ∼ 1/β 1 the string-wall network with N = 1 has been ex- tion reads η a − and the use of η as the independent plored in Refs. [117, 209], where numerical simula- variable leads∼ to a more complicated form of Eq. (B4). tions have been performed to settle the controversy Thus, in a modified cosmology the choice of the scale fac- and a spectrum peaking at a wave number of the tor as the independent variable leads to a simpler form order of the axion mass is obtained. The contri- of the equation of motion. Taking the Fourier transform bution of cold axions from wall decay is found as of the axion field as wall mis αwall = ρA (t0)/ρA (t0) = (32 16) [209, 210]. Z ± iq x ψ(x) = e− ψ(q), (B6)

Appendix B: Vacuum realignment mechanism we find  2   ψ 2 4 2 mA 3 ψ β − 2β The axion field originates from the breaking of the PQ ψ00+χ(1 χ) 2 +a q ψ+9 a sin χ = 0. symmetry at a temperature of the order of f /N. Ax- − a m1 a A (B7) ions, which are the quanta of the axion field, are massless Eq. (B7) expresses the equation of motion for the axion from the moment of production down to the temperature field in the variable a and it is conveniently written to be of QCD transition, when the mass term in Eq. (13) turns solved numerically. in. In this picture, the equation of motion for the angular variable of the axion field at any time is ¯ 2 1. Approximate solutions of the equation of motion θ¨ + 3H θ˙ ∇ θ + m2 sin θ = 0, (B1) − a2 A Analytic solutions to Eq. (B7) can be obtained in the where ¯ is the Laplacian operator with respect to the co- limiting regime θ 1, where Eq. (B7) reads ∇  moving coordinatesx ¯. We re-scale time t and scale factor 2 ψ00 + κ (a) ψ = 0, (B8) a so that these quantities are dimensionless, t t/t1 and a a/a , and we use the definition in Eq. (→13) for the → 1 with the wave number axion mass at the time at which the coherent field oscilla-  2 χ(1 χ) mA 1−β  1−2β 2 tions begin, m1 = mA(T1) = 3H1 = 3β/t1, as a function 2 β β κ (a) = −2 + 9 a + q a . (B9) of the exponent β that expresses the time dependence of a m1 β the scale factor a as a = t in Eq. (10). Using this re- An approximate solution of Eq. (B8), valid in the adia- lation, we obtain t1 = 3β/m1, so that Eq. (B1) in these batic regime in which higher derivatives are neglected, is rescaled quantities reads given by setting 2  2  Z a  ¨ a˙ ˙ 2 2 mA θ + 3 θ β ∇2 θ + 9β sin θ = 0 . (B2) ψ = ψ0(a) exp i κ(a0) da0 , (B10) a − a m1 23 where the amplitude ψ is given by Solution for the zero mode at the onset of oscilla- 0 • 2 tions ψ0(a) κ(a) = const. (B11) | | An approximate solution of Eq. (B8) for the zero- Finally, an approximate solution to Eq. (B8) is [38, 59] momentum mode q = 0, valid after the onset of axion oscillations when t t , is obtained by set-  Z a  1 const. ting ∼ ψ = p exp i κ(a0) da0 . (B12) κ(a) mA(a) 1−β κ(a) 3 a β , (B19) Each of the three terms appearing in Eq. (B9) is the ≈ m1 leading term in a particular regime of the evolution of the axion field. We analyse these approximate behaviour so that the adiabatic solution for ψ in Eq. (B12) in in depths in the following. this slowly oscillating regime gives the axion num- ber density Solution at early times, outside the horizon • 1/β 2   3 At early times t a < t1 prior to the onset of mis 1 2 ψ(a) mis a − ∼ nA (a) = mA(a) fA χ = nA , (B20) axion oscillations, the mass∼ term in Eq. (B8) can be 2 a a1 neglected since mA(a) m1. Defining the physical  mis wavelength λ = a/q, we distinguish two different where nA is the number density of axions from regimes in this approximation, corresponding to the the misalignment mechanism at temperature T1, evolution of the modes outside the horizon (λ > t) or inside the horizon (λ < t). In the first case λ ∼> t, mis 1 2 2 n = mA(T1) f θ . (B21) Eq. (B8) at early times∼ reduces to ∼ A 2 A h i i

2 a ψ00 + χ(1 χ) ψ = 0, (B13) Eq. (B20) shows that, regardless of the dominating − cosmological model, the axion number density of χ with solution (θ = ψ/a ) the zero modes after the onset of axion oscillations 3 1−3β scales with a− . The energy density at temperature β 1 3β θ(q, t) = θ1(q)+θ2(q) a = θ1(q)+θ2(q) t − . (B14) T1 is obtained as

One of the two solutions to Eq. (B13) is thus a con- 4 2   γ mis mis ΛA θi T1 − stant value θ (q), while the second solution drops ρ (T1) = mA n = h i , (B22) 1 A A 2 T to zero for cosmological models with β > 1/3. Re- Λ gardless of the cosmological model considered, the where we have used Eq. (13) to express m1 in terms axion field for modes larger than the horizon is of T . “frozen by causality”. For example, in a radiation- 1 dominated model with β = 1/2, Eq. (B14) coin- cides with the result in Ref. [38], Appendix C: A note on primordial black hole 1/2 formation θ = θ1(q) + θ2(q) t− . (B15)

Solution at early times, inside the horizon Primordial black holes formed through various mech- • anisms, of which one consists in the growing of large in- Eq. (B8) for modes that evolve inside the horizon homogeneities around the QCD phase transition. The < λ t reduces to question is, should axion inhomogeneities also form black ∼ 2 holes instead of condensing into miniclusters? To an-  2 1/β ψ00 + q a − ψ = 0, (B16) swer this question, we compute the Schwarzschild radius 2 rs = M/mPl for the primordial plasma and for the axion whose solution in a closed form, obtained through energy density at the onset of oscillations. Eq. (B12) and θ = ψ/aχ, reads When overdensities in the primordial plasma grow  a  larger than one, a condition for the formation of primor- const. Z 2β−1 θ = exp iq (a0) β da0 . (B17) dial black holes is met. At time t, the mass enclosed a within a Hubble radius is M = ρ/H3, and the ratio be- tween the Schwarzschild radius r and the horizon length The dependence of the amplitude θ 1/a in s 1/H is Eq. (B17) is crucial, since it shows that| | ∼ the axion number density scales with r ρ 1 s = = , (C1) 2 1/H H2 m2 8π θ 3 Pl nA(q, t) | | a− , (B18) ∼ λ ∼ where in the last equality we have used the Friedmann 2 2
for any cosmological model considered. equation H = 8π/3mPl ρ. Thus, the Schwarzschild 24 radius is about one order of magnitude smaller than all cosmological models considered and for all physical the horizon length, so a significant fraction of inhomo- values of TRH and αtot. Primordial black holes cannot geneities can condense into black holes. form from axion cold dark matter using this mechanism, For axion miniclusters of radius R1, the ratio is mainly because the axion field is a subdominant com- ponent of the total energy density at the QCD phase 2 rs H1 Mc 4π (1 + Φ) r ρ1 transition. Recently, the production of primordial black = 2 = 2 2 , (C2) R1 r mPl 3 H1 mPl holes from topological defects arising in the QCD axion theory has been considered in Ref. [237]. The constraint where in the last expression we have used Eq. (33) in on scenarios in which primordial black holes accrete dark mis terms of the energy density of axions ρ1 αtotρA (T1). matter has been considered in Ref. [238]. ≡8 13 The ratio in Eq. (C2) is of the order of 10− to 10− for

[1] V. C. Rubin, W. K. Ford, Jr., N. Thonnard, M. S. arXiv:1401.0709 [physics.class-ph]. Roberts, and J. A. Graham, Astronomical Journal 81, [25] H. Ter¸cas,J. D. Rodrigues, and J. T. Mendon¸ca, Phys. 687 (1976). Rev. Lett. 120, 181803 (2018), arXiv:1801.06254 [hep- [2] V. C. Rubin, N. Thonnard, W. K. Ford, Jr., and M. S. ph]. Roberts, Astronomical Journal 81, 719 (1976). [26] L. Visinelli and H. Ter¸cas, (2018), arXiv:1807.06828 [3] V. Trimble, Annual Review of Astron- [hep-ph]. omy and Astrophysics 25, 425 (1987), [27] I. Stern, Proceedings, 38th International Conference https://doi.org/10.1146/annurev.aa.25.090187.002233. on High Energy Physics (ICHEP 2016): Chicago, IL, [4] P. A. R. Ade et al. (Planck), Astron. Astrophys. 594, USA, August 3-10, 2016, PoS ICHEP2016, 198 (2016), A13 (2016), arXiv:1502.01589 [astro-ph.CO]. arXiv:1612.08296 [physics.ins-det]. [5] N. Aghanim et al. (Planck), (2018), arXiv:1807.06209 [28] M. Raggi and V. Kozhuharov, Adv. High Energy Phys. [astro-ph.CO]. 2014, 959802 (2014), arXiv:1403.3041 [physics.ins-det]. [6] J. Lesgourgues, M. Viel, M. G. Haehnelt, and [29] B. Majorovits and J. Redondo (MADMAX Working R. Massey, JCAP 0711, 008 (2007), arXiv:0705.0533 Group), in Proceedings, 12th Patras Workshop on Ax- [astro-ph]. ions, WIMPs and WISPs (PATRAS 2016): Jeju Is- [7] V. Springel et al., Nature 435, 629 (2005), arXiv:astro- land, South Korea, June 20-24, 2016 (2017) pp. 94–97, ph/0504097 [astro-ph]. arXiv:1611.04549 [astro-ph.IM]. [8] M. Davis, G. Efstathiou, C. S. Frenk, and S. D. M. [30] Y. Kahn, B. R. Safdi, and J. Thaler, Phys. Rev. Lett. White, Astrophys. J. 292, 371 (1985), [,105(1985)]. 117, 141801 (2016), arXiv:1602.01086 [hep-ph]. [9] G. Efstathiou, M. Davis, C. S. Frenk, and S. D. M. [31] D. Alesini, D. Babusci, D. Di Gioacchino, C. Gatti, White, Astrophys. J. Suppl. 57, 241 (1985). G. Lamanna, and C. Ligi, (2017), arXiv:1707.06010 [10] V. Springel, C. S. Frenk, and S. D. M. White, Nature [physics.ins-det]. 440, 1137 (2006), arXiv:astro-ph/0604561 [astro-ph]. [32] D. Babusci et al., (2019), arXiv:1911.02427 [physics.ins- [11] D. Clowe, A. Gonzalez, and M. Markevitch, Astrophys. det]. J. 604, 596 (2004), arXiv:astro-ph/0312273 [astro-ph]. [33] K. J. Bae, H. Baer, E. J. Chun, and C. S. Shin, Phys. [12] M. Markevitch, A. H. Gonzalez, D. Clowe, A. Vikhlinin, Rev. D91, 075011 (2015), arXiv:1410.3857 [hep-ph]. L. David, W. Forman, C. Jones, S. Murray, and [34] K. J. Bae, H. Baer, A. Lessa, and H. Serce, Front.in W. Tucker, Astrophys. J. 606, 819 (2004), arXiv:astro- Phys. 3, 49 (2015), arXiv:1502.07198 [hep-ph]. ph/0309303 [astro-ph]. [35] S. Baum, L. Visinelli, K. Freese, and P. Stengel, Phys. [13] S. Weinberg, Phys. Rev. Lett. 40, 223 (1978). Rev. D95, 043007 (2017), arXiv:1611.09665 [astro- [14] F. Wilczek, Phys. Rev. Lett. 40, 279 (1978). ph.CO]. [15] R. D. Peccei and H. R. Quinn, Phys. Rev. Lett. 38, 1440 [36] G. G. Raffelt, in Dark matter in cosmology, clocks and (1977). test of fundamental laws. Proceedings, 30th Rencontres [16] R. D. Peccei and H. R. Quinn, Phys. Rev. D16, 1791 de Moriond, 15th Moriond Workshop, Villars sur Ollon, (1977). Switzerland, January 22-29, 1995 (1995) pp. 159–168, [17] A. A. Belavin, A. M. Polyakov, A. S. Schwartz, arXiv:hep-ph/9502358 [hep-ph]. and Yu. S. Tyupkin, Phys. Lett. B59, 85 (1975), [37] G. G. Raffelt, Quantum theories and renormalization [,350(1975)]. group in gravity and cosmology: Proceedings, 2nd Inter- [18] G. ’t Hooft, Phys. Rev. Lett. 37, 8 (1976), [,226(1976)]. national Conference, IRGAC 2006, Barcelona, Spain, [19] R. Jackiw and C. Rebbi, Phys. Rev. Lett. 37, 172 July 11-15, 2006, J. Phys. A40, 6607 (2007), arXiv:hep- (1976). ph/0611118 [hep-ph]. [20] C. G. Callan, Jr., R. F. Dashen, and D. J. Gross, Phys. [38] P. Sikivie, Axions: Theory, cosmology, and experi- Lett. B63, 334 (1976), [,357(1976)]. mental searches. Proceedings, 1st Joint ILIAS-CERN- [21] F. Wilczek, Phys. Rev. Lett. 58, 1799 (1987). CAST axion training, Geneva, Switzerland, November [22] S. V. Krasnikov, Phys. Rev. Lett. 76, 2633 (1996). 30-December 2, 2005, Lect. Notes Phys. 741, 19 (2008), [23] R. Li, J. Wang, X. Qi, and S.-C. Zhang, Nature Phys. [,19(2006)], arXiv:astro-ph/0610440 [astro-ph]. 6, 284 (2010), arXiv:0908.1537 [cond-mat.other]. [39] J. E. Kim and G. Carosi, Rev. Mod. Phys. 82, 557 [24] L. Visinelli, Mod. Phys. Lett. A28, 1350162 (2013), (2010), arXiv:0807.3125 [hep-ph]. 25

[40] O. Wantz and E. P. S. Shellard, Phys. Rev. D82, 123508 [hep-ph]. (2010), arXiv:0910.1066 [astro-ph.CO]. [73] J. Eby, M. Leembruggen, P. Suranyi, and L. C. R. [41] M. Kawasaki and K. Nakayama, Ann. Rev. Nucl. Part. Wijewardhana, JHEP 12, 066 (2016), arXiv:1608.06911 Sci. 63, 69 (2013), arXiv:1301.1123 [hep-ph]. [astro-ph.CO]. [42] D. J. E. Marsh, Phys. Rept. 643, 1 (2016), [74] D. G. Levkov, A. G. Panin, and I. I. Tkachev, Phys. arXiv:1510.07633 [astro-ph.CO]. Rev. Lett. 118, 011301 (2017), arXiv:1609.03611 [astro- [43] J. E. Kim (2017) arXiv:1703.03114 [hep-ph]. ph.CO]. [44] A. D. Linde, Phys. Lett. B201, 437 (1988). [75] T. Helfer, D. J. E. Marsh, K. Clough, M. Fairbairn, [45] A. D. Linde, Phys. Lett. B259, 38 (1991). E. A. Lim, and R. Becerril, JCAP 1703, 055 (2017), [46] M. S. Turner and F. Wilczek, Phys. Rev. Lett. 66, 5 arXiv:1609.04724 [astro-ph.CO]. (1991). [76] E. Braaten, A. Mohapatra, and H. Zhang, Phys. Rev. [47] F. Wilczek, (2004), arXiv:hep-ph/0408167 [hep-ph]. D96, 031901 (2017), arXiv:1609.05182 [hep-ph]. [48] M. Tegmark, A. Aguirre, M. Rees, and F. Wilczek, [77] E. Braaten, A. Mohapatra, and H. Zhang, Phys. Rev. Phys. Rev. D73, 023505 (2006), arXiv:astro- D94, 076004 (2016), arXiv:1604.00669 [hep-ph]. ph/0511774 [astro-ph]. [78] Y. Bai, V. Barger, and J. Berger, JHEP 12, 127 (2016), [49] M. P. Hertzberg, M. Tegmark, and F. Wilczek, Phys. arXiv:1612.00438 [hep-ph]. Rev. D78, 083507 (2008), arXiv:0807.1726 [astro-ph]. [79] J. Eby, M. Leembruggen, J. Leeney, P. Suranyi, [50] B. Freivogel, JCAP 1003, 021 (2010), arXiv:0810.0703 and L. C. R. Wijewardhana, JHEP 04, 099 (2017), [hep-th]. arXiv:1701.01476 [astro-ph.CO]. [51] K. J. Mack, JCAP 1107, 021 (2011), arXiv:0911.0421 [80] V. Desjacques, A. Kehagias, and A. Riotto, Phys. Rev. [astro-ph.CO]. D97, 023529 (2018), arXiv:1709.07946 [astro-ph.CO]. [52] L. Visinelli and P. Gondolo, Phys. Rev. D80, 035024 [81] L. Visinelli, S. Baum, J. Redondo, K. Freese, (2009), arXiv:0903.4377 [astro-ph.CO]. and F. Wilczek, Phys. Lett. B777, 64 (2018), [53] B. S. Acharya, K. Bobkov, and P. Kumar, JHEP 11, arXiv:1710.08910 [astro-ph.CO]. 105 (2010), arXiv:1004.5138 [hep-th]. [82] P.-H. Chavanis, Phys. Rev. D98, 023009 (2018), [54] G. Ballesteros, J. Redondo, A. Ringwald, and arXiv:1710.06268 [gr-qc]. C. Tamarit, JCAP 1708, 001 (2017), arXiv:1610.01639 [83] S. Krippendorf, F. Muia, and F. Quevedo, JHEP 08, [hep-ph]. 070 (2018), arXiv:1806.04690 [hep-th]. [55] S. Hoof, F. Kahlhoefer, P. Scott, C. Weniger, and [84] B. Eggemeier and J. C. Niemeyer, Phys. Rev. D100, M. White, JHEP 03, 191 (2019), arXiv:1810.07192 [hep- 063528 (2019), arXiv:1906.01348 [astro-ph.CO]. ph]. [85] G. B. Gelmini and P. Gondolo, JCAP 0810, 002 (2008), [56] T. Tenkanen and L. Visinelli, JCAP 1908, 033 (2019), arXiv:0803.2349 [astro-ph]. arXiv:1906.11837 [astro-ph.CO]. [86] K. M. Zurek, C. J. Hogan, and T. R. Quinn, Phys. Rev. [57] C. J. Hogan and M. J. Rees, Phys. Lett. B205, 228 D75, 043511 (2007), arXiv:astro-ph/0607341 [astro-ph]. (1988). [87] P. Tinyakov, I. Tkachev, and K. Zioutas, JCAP 1601, [58] E. W. Kolb and I. I. Tkachev, Phys. Rev. Lett. 71, 3051 035 (2016), arXiv:1512.02884 [astro-ph.CO]. (1993), arXiv:hep-ph/9303313 [hep-ph]. [88] E. Hardy, JHEP 02, 046 (2017), arXiv:1609.00208 [hep- [59] E. W. Kolb and I. I. Tkachev, Phys. Rev. D49, 5040 ph]. (1994), arXiv:astro-ph/9311037 [astro-ph]. [89] V. I. Dokuchaev, Y. N. Eroshenko, and I. I. Tkachev, [60] E. W. Kolb and I. I. Tkachev, Phys. Rev. D50, 769 Journal of Experimental and Theoretical Physics 125, (1994), arXiv:astro-ph/9403011 [astro-ph]. 434 (2017). [61] A. S. Sakharov, D. D. Sokoloff, and M. Yu. [90] M. Fairbairn, D. J. E. Marsh, and J. Quevillon, Phys. Khlopov, Phys. Atom. Nucl. 59, 1005 (1996), [Yad. Rev. Lett. 119, 021101 (2017), arXiv:1701.04787 [astro- Fiz.59N6,1050(1996)]. ph.CO]. [62] D. J. Kaup, Phys. Rev. 172, 1331 (1968). [91] M. Fairbairn, D. J. E. Marsh, J. Quevillon, [63] R. Ruffini and S. Bonazzola, Phys. Rev. 187, 1767 and S. Rozier, Phys. Rev. D97, 083502 (2018), (1969). arXiv:1707.03310 [astro-ph.CO]. [64] A. Das, Journal of Mathematical Physics 4, 45 (1963), [92] A. Katz, J. Kopp, S. Sibiryakov, and W. Xue, JCAP http://dx.doi.org/10.1063/1.1703887. 1812, 005 (2018), arXiv:1807.11495 [astro-ph.CO]. [65] D. A. Feinblum and W. A. McKinley, Phys. Rev. 168, [93] K. Inomata, M. Kawasaki, K. Mukaida, and 1445 (1968). T. T. Yanagida, Phys. Rev. D97, 043514 (2018), [66] A. F. D. F. Teixeira, I. Wolk, and M. M. Som, Phys. arXiv:1711.06129 [astro-ph.CO]. Rev. D12, 319 (1975). [94] J. Enander, A. Pargner, and T. Schwetz, JCAP 1712, [67] M. Colpi, S. L. Shapiro, and I. Wasserman, Phys. Rev. 038 (2017), arXiv:1708.04466 [astro-ph.CO]. Lett. 57, 2485 (1986). [95] A. Vaquero, J. Redondo, and J. Stadler, JCAP 19, 012 [68] E. Seidel and W. M. Suen, Phys. Rev. Lett. 66, 1659 (2019), arXiv:1809.09241 [astro-ph.CO]. (1991). [96] B. Eggemeier, J. Redondo, K. Dolag, J. C. Niemeyer, [69] I. I. Tkachev, Phys. Lett. B261, 289 (1991). and A. Vaquero, (2019), arXiv:1911.09417 [astro- [70] P.-H. Chavanis, Phys. Rev. D 84, 043531 (2011), ph.CO]. arXiv:1103.2050. [97] C. A. J. O’Hare and A. M. Green, Phys. Rev. D95, [71] E. Braaten, A. Mohapatra, and H. Zhang, Phys. Rev. 063017 (2017), arXiv:1701.03118 [astro-ph.CO]. Lett. 117, 121801 (2016), arXiv:1512.00108 [hep-ph]. [98] M. Vogelsberger and S. D. M. White, Mon. Not. Roy. [72] J. Eby, P. Suranyi, and L. C. R. Wijewardhana, Mod. Astron. Soc. 413, 1419 (2011), arXiv:1002.3162 [astro- Phys. Lett. A31, 1650090 (2016), arXiv:1512.01709 ph.CO]. 26

[99] I. G. Irastorza and J. A. Garcia, JCAP 1210, 022 51 (1983). (2012), arXiv:1207.6129 [physics.ins-det]. [129] M. S. Turner, Phys. Rev. D 28, 1243 (1983). [100] I. G. Irastorza and J. Redondo, Prog. Part. Nucl. Phys. [130] R. J. Scherrer and M. S. Turner, Phys. Rev. D 31, 681 102, 89 (2018), arXiv:1801.08127 [hep-ph]. (1985). [101] S. Knirck, A. J. Millar, C. A. J. O’Hare, J. Re- [131] G. Lazarides, C. Panagiotakopoulos, and Q. Shafi, dondo, and F. D. Steffen, JCAP 1811, 051 (2018), Phys. Lett. B192, 323 (1987). arXiv:1806.05927 [astro-ph.CO]. [132] G. Lazarides, R. K. Schaefer, D. Seckel, and Q. Shafi, [102] M. Kawasaki, K. Kohri, and N. Sugiyama, Phys. Rev. Nucl. Phys. B346, 193 (1990). Lett. 82, 4168 (1999), arXiv:astro-ph/9811437. [133] V. Balasubramanian, P. Berglund, J. P. Conlon, [103] M. Kawasaki, K. Kohri, and N. Sugiyama, Phys. Rev. and F. Quevedo, JHEP 03, 007 (2005), arXiv:hep- D62, 023506 (2000), arXiv:astro-ph/0002127. th/0502058 [hep-th]. [104] S. Hannestad, Phys. Rev. D70, 043506 (2004), [134] J. P. Conlon, F. Quevedo, and K. Suruliz, JHEP 08, arXiv:astro-ph/0403291. 007 (2005), arXiv:hep-th/0505076 [hep-th]. [105] K. Ichikawa, M. Kawasaki, and F. Takahashi, Phys. [135] J. D. Barrow, Nucl. Phys. B208, 501 (1982). Rev. D72, 043522 (2005), arXiv:astro-ph/0505395. [136] L. H. Ford, Phys. Rev. D35, 2955 (1987). [106] F. D. Bernardis, L. Pagano, and A. Melchiorri, As- [137] B. Spokoiny, Phys. Lett. B315, 40 (1993), arXiv:gr- tropart. Phys. 30, 192 (2008). qc/9306008 [gr-qc]. [107] L. Visinelli and P. Gondolo, Phys. Rev. D81, 063508 [138] M. Joyce, Phys. Rev. D55, 1875 (1997), arXiv:hep- (2010), arXiv:0912.0015 [astro-ph.CO]. ph/9606223 [hep-ph]. [108] L. Visinelli, Phys. Rev. D96, 023013 (2017), [139] P. Salati, Phys. Lett. B571, 121 (2003), arXiv:astro- arXiv:1703.08798 [astro-ph.CO]. ph/0207396 [astro-ph]. [109] N. Ramberg and L. Visinelli, Phys. Rev. D99, 123513 [140] S. Profumo and P. Ullio, JCAP 0311, 006 (2003), (2019), arXiv:1904.05707 [astro-ph.CO]. arXiv:hep-ph/0309220 [hep-ph]. [110] S. Borsanyi et al., Nature 539, 69 (2016), [141] C. Pallis, JCAP 0510, 015 (2005), arXiv:hep- arXiv:1606.07494 [hep-lat]. ph/0503080 [hep-ph]. [111] D. Grin, T. L. Smith, and M. Kamionkowski, Phys. [142] C. Pallis, Nucl. Phys. B751, 129 (2006), arXiv:hep- Rev. D77, 085020 (2008), arXiv:0711.1352 [astro-ph]. ph/0510234 [hep-ph]. [112] G. F. Giudice, E. W. Kolb, A. Riotto, D. V. Semikoz, [143] M. E. Gomez, S. Lola, C. Pallis, and J. Rodriguez- and I. I. Tkachev, Phys. Rev. D64, 043512 (2001), Quintero, Proceedings, 4th International Workshop on arXiv:hep-ph/0012317 [hep-ph]. the Dark Side of the Universe (DSU 2008): Cairo, [113] G. F. Giudice, A. Riotto, and I. I. Tkachev, JHEP 06, Egypt, June 1-5, 2008, AIP Conf. Proc. 1115, 157 020 (2001), arXiv:hep-ph/0103248 [hep-ph]. (2009), arXiv:0809.1982 [hep-ph]. [114] G. Grilli di Cortona, E. Hardy, J. Pardo Vega, and [144] S. Lola, C. Pallis, and E. Tzelati, JCAP 0911, 017 G. Villadoro, JHEP 01, 034 (2016), arXiv:1511.02867 (2009), arXiv:0907.2941 [hep-ph]. [hep-ph]. [145] M. Lewicki, T. Rindler-Daller, and J. D. Wells, JHEP [115] K. J. Bae, J.-H. Huh, and J. E. Kim, JCAP 0809, 005 06, 055 (2016), arXiv:1601.01681 [hep-ph]. (2008), arXiv:0806.0497 [hep-ph]. [146] M. Artymowski, M. Lewicki, and J. D. Wells, JHEP [116] P. Gondolo and L. Visinelli, Phys. Rev. Lett. 113, 03, 066 (2017), arXiv:1609.07143 [hep-ph]. 011802 (2014), arXiv:1403.4594 [hep-ph]. [147] K. Redmond and A. L. Erickcek, Phys. Rev. D96, [117] T. Hiramatsu, M. Kawasaki, and K. Saikawa, JCAP 043511 (2017), arXiv:1704.01056 [hep-ph]. 1108, 030 (2011), arXiv:1012.4558 [astro-ph.CO]. [148] F. D’Eramo, N. Fernandez, and S. Profumo, JCAP [118] V. B. Klaer and G. D. Moore, JCAP 1711, 049 (2017), 1705, 012 (2017), arXiv:1703.04793 [hep-ph]. arXiv:1708.07521 [hep-ph]. [149] L. Visinelli, Symmetry 10, 546 (2018), arXiv:1710.11006 [119] L. Fleury and G. D. Moore, JCAP 1601, 004 (2016), [astro-ph.CO]. arXiv:1509.00026 [hep-ph]. [150] K. Redmond, A. Trezza, and A. L. Erickcek, Phys. Rev. [120] V. B. Klaer and G. D. Moore, JCAP 1710, 043 (2017), D98, 063504 (2018), arXiv:1807.01327 [astro-ph.CO]. arXiv:1707.05566 [hep-ph]. [151] L. F. Abbott and P. Sikivie, Phys. Lett. B120, 133 [121] M. Gorghetto, E. Hardy, and G. Villadoro, JHEP 07, (1983). 151 (2018), arXiv:1806.04677 [hep-ph]. [152] J. Preskill, M. B. Wise, and F. Wilczek, Phys. Lett. [122] E. W. Kolb, D. J. H. Chung, and A. Riotto, Trends in B120, 127 (1983). theoretical physics II. Proceedings, 2nd La Plata Meet- [153] D. J. Gross, R. D. Pisarski, and L. G. Yaffe, Rev. Mod. ing, Buenos Aires, Argentina, November 29-December Phys. 53, 43 (1981). 4, 1998, AIP Conf. Proc. 484, 91 (1999), [,592(1999)], [154] G. G. Raffelt, Physics Letters B 166, 402 (1986). arXiv:hep-ph/9810361 [hep-ph]. [155] G. G. Raffelt, “Astrophysical axion bounds,” in Axions: [123] G. F. Giudice, E. W. Kolb, and A. Riotto, Phys. Rev. Theory, Cosmology, and Experimental Searches, edited D64, 023508 (2001), arXiv:hep-ph/0005123 [hep-ph]. by M. Kuster, G. Raffelt, and B. Beltr´an(Springer [124] L. Visinelli, JCAP 1501, 005 (2015), arXiv:1410.1187 Berlin Heidelberg, Berlin, Heidelberg, 2008) pp. 51–71. [astro-ph.CO]. [156] N. Viaux, M. Catelan, P. B. Stetson, G. Raffelt, [125] L. Visinelli, JCAP 1607, 054 (2016), arXiv:1605.06449 J. Redondo, A. A. R. Valcarce, and A. Weiss, Phys. [astro-ph.CO]. Rev. Lett. 111, 231301 (2013), arXiv:1311.1669 [astro- [126] K. Freese, E. I. Sfakianakis, P. Stengel, and L. Visinelli, ph.SR]. JCAP 1805, 067 (2018), arXiv:1712.03791 [hep-ph]. [157] M. Giannotti, I. G. Irastorza, J. Redondo, A. Ring- [127] M. Dine and W. Fischler, Phys. Lett. B120, 137 (1983). wald, and K. Saikawa, JCAP 1710, 010 (2017), [128] P. J. Steinhardt and M. S. Turner, Phys. Lett. B129, arXiv:1708.02111 [hep-ph]. 27

[158] J. K. Vogel, E. Armengaud, F. T. Avignone, M. Betz, [177] J. Chluba, J. Hamann, and S. P. Patil, Int. J. Mod. P. Brax, P. Brun, G. Cantatore, J. M. Carmona, Phys. D24, 1530023 (2015), arXiv:1505.01834 [astro- G. P. Carosi, F. Caspers, S. Caspi, S. A. Cetin, ph.CO]. D. Chelouche, F. E. Christensen, A. Dael, T. Dafni, [178] J. I. Read, J. Phys. G41, 063101 (2014), M. Davenport, A. V. Derbin, K. Desch, A. Diago, arXiv:1404.1938 [astro-ph.GA]. B. D¨obrich, I. Dratchnev, A. Dudarev, C. Eleftheriadis, [179] V. I. Dokuchaev, Y. N. Eroshenko, and I. I. Tkachev, G. Fanourakis, E. Ferrer-Ribas, J. Gal´an,J. A. Garc´ıa, Soviet Journal of Experimental and Theoretical Physics J. G. Garza, T. Geralis, B. Gimeno, I. Giomataris, 125, 434 (2017), arXiv:1710.09586 [astro-ph.GA]. S. Gninenko, H. G´omez,D. Gonz´alez-D´ıaz,E. Guendel- [180] V. S. Berezinsky, V. I. Dokuchaev, and Y. N. man, C. J. Hailey, T. Hiramatsu, D. H. H. Hoffmann, Eroshenko, Phys. Usp. 57, 1 (2014), [Usp. Fiz. D. Horns, F. J. Iguaz, I. G. Irastorza, J. Isern, K. Imai, Nauk184,3(2014)], arXiv:1405.2204 [astro-ph.HE]. A. C. Jakobsen, J. Jaeckel, K. Jakovcic, J. Kaminski, [181] T. Goerdt, O. Y. Gnedin, B. Moore, J. Diemand, and M. Kawasaki, M. Karuza, M. Krcmar, K. Kousouris, J. Stadel, Mon. Not. Roy. Astron. Soc. 375, 191 (2007), C. Krieger, B. Lakic, O. Limousin, A. Lindner, A. Lio- arXiv:astro-ph/0608495 [astro-ph]. lios, G. Luz´on,S. Matsuki, V. N. Muratova, C. Nones, [182] A. Schneider, L. Krauss, and B. Moore, Phys. Rev. I. Ortega, T. Papaevangelou, M. J. Pivovaroff, G. Raf- D82, 063525 (2010), arXiv:1004.5432 [astro-ph.GA]. felt, J. Redondo, A. Ringwald, S. Russenschuck, J. Ruz, [183] K. Kuijken and G. Gilmore, Mon. Not. Roy. Astron. K. Saikawa, I. Savvidis, T. Sekiguchi, Y. K. Semertzidis, Soc. 239, 605 (1989). I. Shilon, P. Sikivie, H. Silva, H. ten Kate, A. Tomas, [184] E. W. Kolb and I. I. Tkachev, Astrophys. J. 460, L25 S. Troitsky, T. Vafeiadis, K. van Bibber, P. Vedrine, (1996), arXiv:astro-ph/9510043 [astro-ph]. J. A. Villar, L. Walckiers, A. Weltman, W. Wester, [185] A. L. Erickcek and K. Sigurdson, Phys. Rev. D84, S. C. Yildiz, and K. Zioutas, Physics Procedia 61, 193 083503 (2011), arXiv:1106.0536 [astro-ph.CO]. (2015). [186] A. L. Erickcek, K. Sinha, and S. Watson, Phys. Rev. [159] M. Yamaguchi, J. Yokoyama, and M. Kawasaki, Phys. D94, 063502 (2016), arXiv:1510.04291 [hep-ph]. Rev. D61, 061301 (2000), arXiv:hep-ph/9910352. [187] L. Visinelli and P. Gondolo, Phys. Rev. D91, 083526 [160] G. Nicholson and C. R. Contaldi, JCAP 0907, 011 (2015), arXiv:1501.02233 [astro-ph.CO]. (2009), arXiv:0903.1106 [astro-ph.CO]. [188] A. E. Nelson and H. Xiao, Phys. Rev. D98, 063516 [161] G. Nicholson, C. R. Contaldi, and P. Paykari, JCAP (2018), arXiv:1807.07176 [astro-ph.CO]. 1001, 016 (2010), arXiv:0909.5092 [astro-ph.CO]. [189] E. Kolb and M. Turner, The Early Universe, Frontiers [162] S. Bird, H. V. Peiris, M. Viel, and L. Verde, Mon. Not. in physics (Avalon Publishing, 1994). Roy. Astron. Soc. 413, 1717 (2011), arXiv:1010.1519 [190] R. D. Peccei, Axions: Theory, cosmology, and experi- [astro-ph.CO]. mental searches. Proceedings, 1st Joint ILIAS-CERN- [163] J. Chluba, R. Khatri, and R. A. Sunyaev, Mon. Not. CAST axion training, Geneva, Switzerland, November Roy. Astron. Soc. 425, 1129 (2012), arXiv:1202.0057 30-December 2, 2005, Lect. Notes Phys. 741, 3 (2008), [astro-ph.CO]. [,3(2006)], arXiv:hep-ph/0607268 [hep-ph]. [164] J. Chluba, A. L. Erickcek, and I. Ben-Dayan, Astro- [191] J. E. Kim and G. Carosi, Rev. Mod. Phys. 82, 557 phys. J. 758, 76 (2012), arXiv:1203.2681 [astro-ph.CO]. (2010). [165] J. Chluba and D. Grin, Mon. Not. Roy. Astron. Soc. [192] L. D. Duffy and K. van Bibber, New J. Phys. 11, 105008 434, 1619 (2013), arXiv:1304.4596 [astro-ph.CO]. (2009), arXiv:0904.3346 [hep-ph]. [166] J. Chluba et al., (2019), arXiv:1903.04218 [astro- [193] A. Ringwald, Phys. Dark Univ. 1, 116 (2012), ph.CO]. arXiv:1210.5081 [hep-ph]. [167] D. Jeong, J. Pradler, J. Chluba, and M. Kamionkowski, [194] A. G. Dias, A. C. B. Machado, C. C. Nishi, A. Ring- Phys. Rev. Lett. 113, 061301 (2014), arXiv:1403.3697 wald, and P. Vaudrevange, JHEP 06, 037 (2014), [astro-ph.CO]. arXiv:1403.5760 [hep-ph]. [168] A. S. Josan, A. M. Green, and K. A. Malik, Phys. Rev. [195] M. S. Turner, Phys. Rev. Lett. 59, 2489 (1987). D79, 103520 (2009), arXiv:0903.3184 [astro-ph.CO]. [196] T. Higaki and F. Takahashi, JHEP 11, 125 (2012), [169] T. Bringmann, P. Scott, and Y. Akrami, Phys. Rev. arXiv:1208.3563 [hep-ph]. D85, 125027 (2012), arXiv:1110.2484 [astro-ph.CO]. [197] M. Cicoli, M. Goodsell, and A. Ringwald, JHEP 10, [170] G. Aslanyan, L. C. Price, J. Adams, T. Bringmann, 146 (2012), arXiv:1206.0819 [hep-th]. H. A. Clark, R. Easther, G. F. Lewis, and P. Scott, [198] M. Cicoli, J. P. Conlon, and F. Quevedo, Phys. Rev. Phys. Rev. Lett. 117, 141102 (2016), arXiv:1512.04597 D87, 043520 (2013), arXiv:1208.3562 [hep-ph]. [astro-ph.CO]. [199] J. Hasenkamp and J. Kersten, JCAP 1308, 024 (2013), [171] F. Vernizzi, Phys. Rev. D69, 083526 (2004), arXiv:1212.4160 [hep-ph]. arXiv:astro-ph/0311167 [astro-ph]. [200] J. P. Conlon and M. C. D. Marsh, JHEP 10, 214 (2013), [172] A. A. Starobinsky, JETP Lett. 55, 489 (1992), [Pisma arXiv:1304.1804 [hep-ph]. Zh. Eksp. Teor. Fiz.55,477(1992)]. [201] R. L. Davis, Phys. Rev. D 32, 3172 (1985). [173] A. G. Polnarev and I. Musco, Class. Quant. Grav. 24, [202] R. A. Battye and E. P. S. Shellard, Nucl. Phys. B423, 1405 (2007), arXiv:gr-qc/0605122 [gr-qc]. 260 (1994), arXiv:astro-ph/9311017 [astro-ph]. [174] N. Barnaby and Z. Huang, Phys. Rev. D80, 126018 [203] D. Harari and P. Sikivie, Phys. Lett. B195, 361 (1987). (2009), arXiv:0909.0751 [astro-ph.CO]. [204] C. Hagmann and P. Sikivie, Nucl. Phys. B363, 247 [175] A. D. Linde, Phys. Rev. D49, 748 (1994), arXiv:astro- (1991). ph/9307002 [astro-ph]. [205] S. Chang, C. Hagmann, and P. Sikivie, Phys. Rev. D59, [176] N. Barnaby and M. Peloso, Phys. Rev. Lett. 106, 023505 (1999), arXiv:hep-ph/9807374 [hep-ph]. 181301 (2011), arXiv:1011.1500 [hep-ph]. [206] M. Yamaguchi, M. Kawasaki, and J. Yokoyama, Phys. 28

Rev. Lett. 82, 4578 (1999), arXiv:hep-ph/9811311 [hep- arXiv:hep-th/0605206 [hep-th]. ph]. [224] A. Arvanitaki, S. Dimopoulos, S. Dubovsky, N. Kaloper, [207] T. Hiramatsu, M. Kawasaki, T. Sekiguchi, M. Yam- and J. March-Russell, Phys. Rev. D81, 123530 (2010), aguchi, and J. Yokoyama, Phys. Rev. D83, 123531 arXiv:0905.4720 [hep-th]. (2011), arXiv:1012.5502 [hep-ph]. [225] M. Dine, G. Festuccia, J. Kehayias, and W. Wu, JHEP [208] M. Nagasawa and M. Kawasaki, Phys. Rev. D50, 4821 01, 012 (2011), arXiv:1010.4803 [hep-th]. (1994), arXiv:astro-ph/9402066 [astro-ph]. [226] A. Ringwald, Proceedings, 18th International Sympo- [209] T. Hiramatsu, M. Kawasaki, K. Saikawa, and sium on Particles, Strings and Cosmology (PASCOS T. Sekiguchi, Phys. Rev. D85, 105020 (2012), [Er- 2012): Merida, Yucatan, Mexico, June 3-8, 2012,J. ratum: Phys. Rev.D86,089902(2012)], arXiv:1202.5851 Phys. Conf. Ser. 485, 012013 (2014), arXiv:1209.2299 [hep-ph]. [hep-ph]. [210] M. Kawasaki, K. Saikawa, and T. Sekiguchi, Phys. Rev. [227] J. E. Kim, Phys. Rev. Lett. 111, 031801 (2013), D91, 065014 (2015), arXiv:1412.0789 [hep-ph]. arXiv:1303.1822 [hep-ph]. [211] S. Chang and K. Choi, Phys. Lett. B316, 51 (1993), [228] T. C. Bachlechner, C. Long, and L. McAllister, JHEP arXiv:hep-ph/9306216 [hep-ph]. 12, 042 (2015), arXiv:1412.1093 [hep-th]. [212] S. Hannestad and G. Raffelt, JCAP 0404, 008 (2004), [229] J. Halverson, C. Long, and P. Nath, Phys. Rev. D96, arXiv:hep-ph/0312154 [hep-ph]. 056025 (2017), arXiv:1703.07779 [hep-ph]. [213] S. Hannestad, A. Mirizzi, G. G. Raffelt, and Y. Y. Y. [230] M. J. Stott, D. J. E. Marsh, C. Pongkitivanichkul, L. C. Wong, JCAP 1008, 001 (2010), arXiv:1004.0695 [astro- Price, and B. S. Acharya, Phys. Rev. D96, 083510 ph.CO]. (2017), arXiv:1706.03236 [astro-ph.CO]. [214] M. Archidiacono, S. Hannestad, A. Mirizzi, G. Raf- [231] L. Visinelli and S. Vagnozzi, Phys. Rev. D99, 063517 felt, and Y. Y. Y. Wong, JCAP 1310, 020 (2013), (2019), arXiv:1809.06382 [hep-ph]. arXiv:1307.0615 [astro-ph.CO]. [232] G. Mangano, G. Miele, S. Pastor, and M. Peloso, Phys. [215] E. Di Valentino, E. Giusarma, M. Lattanzi, O. Mena, Lett. B534, 8 (2002), arXiv:astro-ph/0111408 [astro- A. Melchiorri, and J. Silk, Phys. Lett. B752, 182 ph]. (2016), arXiv:1507.08665 [astro-ph.CO]. [233] G. D. Coughlan, W. Fischler, E. W. Kolb, S. Raby, and [216] N. Arkani-Hamed, S. Dimopoulos, and G. R. Dvali, G. G. Ross, Phys. Lett. 131B, 59 (1983). Phys. Rev. D59, 086004 (1999), arXiv:hep-ph/9807344 [234] B. S. Acharya, P. Kumar, K. Bobkov, G. Kane, J. Shao, [hep-ph]. and S. Watson, JHEP 06, 064 (2008), arXiv:0804.0863 [217] L. Randall and R. Sundrum, Phys. Rev. Lett. 83, 4690 [hep-ph]. (1999), arXiv:hep-th/9906064 [hep-th]. [235] P. Di Vecchia and G. Veneziano, Nucl. Phys. B171, 253 [218] P. Binetruy, C. Deffayet, and D. Langlois, Nucl. Phys. (1980). B565, 269 (2000), arXiv:hep-th/9905012 [hep-th]. [236] T. W. B. Kibble, Journal of Physics A Mathematical [219] D. J. H. Chung and K. Freese, Phys. Rev. D62, 063513 General 9, 1387 (1976). (2000), arXiv:hep-ph/9910235 [hep-ph]. [237] F. Ferrer, E. Masso, G. Panico, O. Pujolas, and [220] R. R. Caldwell and D. Langlois, Phys. Lett. B511, 129 F. Rompineve, Phys. Rev. Lett. 122, 101301 (2019), (2001), arXiv:gr-qc/0103070 [gr-qc]. arXiv:1807.01707 [hep-ph]. [221] L. Visinelli, Gen. Rel. Grav. 47, 62 (2015), [238] S. M. Boucenna, F. Kuhnel, T. Ohlsson, and arXiv:1410.1523 [gr-qc]. L. Visinelli, JCAP 1807, 003 (2018), arXiv:1712.06383 [222] L. Visinelli, N. Bolis, and S. Vagnozzi, Phys. Rev. D97, [hep-ph]. 064039 (2018), arXiv:1711.06628 [gr-qc]. [223] P. Svrcek and E. Witten, JHEP 06, 051 (2006),