Hunting Dark Matter with the Einstein Telescope A. Riotto Abstract Too light primordial black holes evaporate and are therefore strongly constrained by various bounds, e.g. Cosmic Microwave Background distortion. However, if they are formed strongly clustered, the corresponding haloes may collapse in heavier black holes which may form the entirety of the dark matter of the universe. The indirect signal of such scenario is the production of a flat stochastic background of gravitational waves which is detectable by the Einstein Telescope. 1 Introduction Primordial Black Holes (PBHs) [ 34 , 35 ] have recently attracted significant attention, as they may constitute the entirety of Dark Matter (DM) in the so-called asteoridal mass range defined as 10^{-16}M_{\odot}\lesssim M_{\rm PBH}\lesssim 10^{-11}M_{\odot}\, (1.1) and generate gravitational-wave (GW) signals potentially detectable by current and future experiments (see e.g. Ref. [ 33 ] for a recent review). Among the various formation mechanisms, the standard scenario involves the collapse of large-amplitude curvature perturbations generated during inflation. The characteristic scale of these perturbations determines the typical PBH mass and, at the same time, sets the frequency of the associated scalar-induced gravitational-wave (SIGW) background, according to (for a recent review see Ref. [ 44 ] ) f\simeq 2.7\,{\rm nHz}\,\left(\frac{M_{\rm PBH}}{M_{\odot}}\right)^{-1/2}. (1.2) This relation establishes a direct link between PBH mass and GW frequency, enabling the exploration of PBH formation across a wide mass range through different GW observatories. On the one hand, the space-based interferometer LISA [ 16 ] will probe the mHz frequency band, corresponding to asteroid-mass PBHs, which can account for the totality of DM (see Refs. [ 22 , 23 , 28 , 51 , 62 ] for works on the topic). On the other hand, the Einstein Telescope (ET) [ 80 , 71 , 1 ] will be sensitive to GW signals in the \mathcal{O}(10)\, Hz band, corresponding to the formation of much lighter PBHs, M_{\rm PBH}\sim 10^{-19}M_{\odot} , which would have already evaporated by today, leading to strong constraints on their abundance because of, e.g., the distortion in the CMB. As a consequence, it is often argued that ET cannot (despite indirectly) probe PBHs constituting the totality of DM in the universe. In this paper we show that in fact ET may give us a hint about the DM. Suppose PBHs are strongly clustered, beyond the Poisson expectation, at the time of formation [ 36 , 85 , 19 , 25 , 42 , 5 , 57 , 84 , 73 , 76 , 89 , 14 , 38 , 40 , 53 , 78 , 41 , 8 , 15 , 37 ] . If so, PBHs are formed in close proximity, and their subsequent gravitational interactions can lead to the collapse of the entire cluster into a heavier PBH, even during the radiation-dominated era, thereby evading the bounds associated with evaporated PBHs. This mechanism is commonly referred to as clusterogenesis [ 41 ] . Very light PBHs (which potentially may evaporate and so, for instance, distorting the CMB spectrum) if strongly clustered may not only form a heavier and harmless PBHs, but their mass can easily fall in the range where they can form the entirety of the DM. The intriguing bonus is that the stochastic SIGW spectrum generated at formation of the light PBHs at second-order falls in the range detectable by the ET so that the latter may help in the hunt of DM and understanding its nature. The paper is organized as follows. In section II we show why the correlation function of the compaction function and of the curvature perturbation are linked to each other, and using basic concept of probability conservation, we provide the expression for the correlation function in the case of broad spectrum. In section III we determine the condition for which clusterogenesis happens and the direct consequences on the gravitational waves signal produced, with a particular focus on the ET frequency band, are discussed in section IV. We conclude in section V. 2 The PBH clustering 2.1 Setting the stage Before launching ourselves into the details, some necessary technical details. The key starting object is the curvature perturbation \zeta(\bf x) on superhorizon scales and appearing in the metric in the comoving uniform-energy density gauge {\rm d}s^{2}=-{\rm d}t^{2}+a^{2}(t)e^{2\zeta(\bf x)}{\rm d}{\bf x}^{2}, (2.1) where a(t) is the scale factor in terms of the cosmic time. Cosmological perturbations may gravitationally collapse to form a PBH depending upon the amplitude measured at the peak of the compaction function [ 83 ] , defined to be the mass excess compared to the background value in a given radius C({\bf x})=2\frac{M({\bf x},t)-M_{\rm b}({\bf x},t)}{R({\bf x},t)}, (2.2) where M({\bf x},t) is the Misner-Sharp mass, with M_{\rm b}({\bf x},t) its background value. The Misner-Sharp mass gives the mass within a sphere of areal radius R({\bf x},t)=a(t)re^{\zeta(\bf x)} with spherical coordinate radius r , centred around position {\bf x} and evaluated at time t . The compaction directly measures the overabundance of mass in a region and is therefore better suited than the curvature perturbation for determining when a region collapses. Furthermore, the compaction has the advantage to be time-independent on superhorizon scales. It can be written in terms of the density contrast as C({\bf x})=\frac{2}{R({\bf x},t)}\int{\rm d}^{3}{\bf x}\,\rho_{\rm b}\,\delta({\bf x},t), (2.3) where \rho_{\rm b} is the background energy density and on superhorizon scales the density contrast is related to the curvature perturbation in real space by the nonlinear relation [ 55 ] \delta({\bf x},t)=-\frac{4}{9}\frac{1}{a^{2}H^{2}}e^{-2\zeta(\bf x)}\left(\nabla^{2}\zeta({\bf x})+\frac{1}{2}\left({\mathbf{\nabla}}\zeta({\bf x})\right)^{2}\right). (2.4) Assuming spherical symmetry and defining \zeta^{\prime}={\rm d}\zeta/{\rm d}r , the compaction function becomes C(r)=\frac{8\pi}{R(r,t)}\rho_{\rm b}\int_{0}^{R}{\rm d}\widetilde{R}\,\widetilde{R}^{2}(r,t)\delta(r,t)=C_{1}(r)-\frac{3}{8}C_{1}^{2}(r),\quad\textrm{with}\quad C_{1}(r)=-\frac{4}{3}r\zeta^{\prime}(r). (2.5) Notice that the compaction function is half of the density perturbation averaged inside the radius r at horizon entry. To characterize the PBH two-point correlation function \xi(r) (or, simply, correlation function) at any comoving separation r=|{\bf r}| , we can use the overdensity of discrete PBH centers at position {{\bf r}}_{i} (eventually smoothed over a sphere with a radius equal to the Hubble radius at the moment the perturbations re-enter the horizon) \delta_{\rm PBH}({\bf r})=\frac{1}{\overline{n}_{\rm PBH}}\sum_{i}\delta_{D}({\bf r}-{\bf r}_{i})-1, (2.6) where \delta_{D}({\bf x}) is the three-dimensional Dirac distribution, n_{\rm PBH} is the average comoving number density of PBHs, and i runs over the initial positions of PBHs. The corresponding two-point correlation function must take the general form \displaystyle\big\langle\delta_{\rm PBH}({\bf x})\delta_{\rm PBH}(0)\big\rangle \displaystyle= \displaystyle\frac{1}{\overline{n}_{\rm PBH}}\delta_{D}({\bf x})-1+\frac{1}{\overline{n}_{\rm PBH}^{2}}\Big<\sum_{i\neq j}\delta_{D}({\bf x}-{\bf x}_{i})\delta_{D}({\bf x}_{j})\Big>=\frac{1}{\overline{n}_{\rm PBH}}\delta_{D}({\bf x})+\xi({\bf x}), where [ 41 ] \overline{n}_{\text{\rm PBH}}\simeq 30f_{\text{\rm PBH}}\left(\frac{M_{\rm PBH}}{M_{\odot}}\right)^{-1}\mathrm{kpc}^{-3}. (2.8) The first term in the final step of Eq. ( 2.1 ) is the unavoidable Poisson contribution coming from the fact that PBHs are discrete objects. In the scenario where PBHs are formed at the collapse of sizeable overdensities at re-entry within the comoving horizon {\cal H}^{-1} , we have \overline{n}_{\text{\rm PBH}}\sim\frac{1}{P_{1}{\cal H}^{3}}, (2.9) where P_{1} is the probability to form a PBH inside the comoving horizon. 2.2 A peak of the compaction function for a peak of the curvature perturbation Suppose now that there is peak in the curvature perturbation \zeta({\bf x}) with a given peak value \zeta(0) and profile \zeta(r) away from the center, which we arbitrarily can set at the origin of the coordinates. It is easy to convince oneself that to each peak of the curvature perturbation corresponds a peak of the compaction function. Performing a rotation of the coordinate axes to be aligned with the principal axes of length \lambda_{i} ( i=1,2,3 ) of the constant-curvature perturbation ellipsoid and Taylor expanding up to second-order one obtains [ 21 ] \zeta(r)\simeq\zeta(0)-\frac{1}{2}\sum_{i=1}^{3}\lambda_{i}x_{i}^{2},\quad r^{2}=\sum_{i=1}^{3}x_{i}^{2}. (2.10) Large amplitude peaks are approximately spherical with a common positive \lambda [ 20 ] . Therefore C_{1}(r)\simeq\frac{4}{3}\lambda r^{2}, (2.11) from which we deduce that, around the point {\bf x}=0 , the function C_{1}(r) increases. Since at large radii it must decrease and vanish, there must be a value, let us call it r_{m} , at which C_{1}(r) has a maximum. Now, since C^{\prime}(r_{m})=C^{\prime}_{1}(r_{m})\left(1-\frac{3}{4}C_{1}(r_{m})\right)=0, (2.12) the extremum of the compaction function C(r) coincides with the extremum of its linear counterpart C_{1}(r) . Furthermore, C^{\prime\prime}(r_{m})=C^{\prime\prime}_{1}(r_{m})\left(1-\frac{3}{4}C_{1}(r_{m})\right), (2.13) and the maximum of the compaction function C(r) coincides with the maximum of its linear counterpart C_{1}(r) as long as C_{1}(r_{m})<4/3 (that is, we consider type I PBHs), which we safely assume from now on. To locate the thresholds of the compaction function is therefore enough to locate those of C_{1} which, in turn, correspond to the thresholds of the fully non-Gaussian curvature perturbation. The correlation function of the PBHs is therefore the correlation function of the fully non-Gaussian curvature perturbation, which we now study. 2.3 The full non-Gaussian two-point correlation Consider now a generic fully non-Gaussian curvature perturbation, whose probability has a non-Gaussian tail at large values. It is important to remark at this stage that from now on we will be dealing with classical quantities in the sense that we focus on the perturbations whose wavelengths are larger than the Hubble radius at the end of inflation. Without loss of generality, we can express as a functional of a Gaussian counterpart \zeta_{\rm G} , \zeta({\bf x})=F\left[\zeta_{\rm g}({\bf x})\right], (2.14) with joint probability P[\zeta({\bf x}_{1}),\cdots,\zeta({\bf x}_{n})] . As a concrete example, we can mention the relation \zeta=-\mu\ln\left(1-\frac{\zeta_{{\rm g}}}{\mu}\right), (2.15) which arises naturally in standard models such as the Ultra-slow roll inflationary scenario [ 12 ] and the curvaton mechanism [ 82 , 79 ] . In the first case, \mu is typically related to the slope of the inflationary power spectrum of \zeta_{{\rm g}} beyond its peak. In models in agreement with CMB observations we have 1.5\mathrel{\hbox to0.0pt{\lower 4.0pt\hbox{\hskip 0.5pt$\sim$}\hss}\raise 1.0pt\hbox{$<$}}\mu\mathrel{\hbox to0.0pt{\lower 4.0pt\hbox{\hskip 0.5pt$\sim$}\hss}\raise 1.0pt\hbox{$<$}}5 [ 13 , 63 , 50 ] . For curvaton models, \mu is intricately determined by the time of the curvaton’s decay into radiation and the shape of the power spectrum [ 69 , 45 , 70 , 7 , 47 ] , with typical values of \mathcal{O}(0.1) . Expanding Eq. ( 2.15 ) to second-order, we find that f_{\text{\rm NL}}=5/(6\mu) . We remark however that our arguments are not restricted to the expression ( 2.15 ). Now, in threshold statistics, the correlation of the thresholds of the fully non-Gaussian curvature perturbation is \displaystyle 1+\xi(r) \displaystyle= \displaystyle\frac{P_{2}}{P^{2}_{1}}, \displaystyle P_{1} \displaystyle= \displaystyle\int_{\zeta>\zeta_{\rm c}}{\rm d}\zeta\,P[\zeta], \displaystyle P_{2} \displaystyle= \displaystyle\int_{\zeta(0)>\zeta_{\rm c}}{\rm d}\zeta(0)\int_{\zeta(r)>\zeta_{\rm c}}{\rm d}\zeta(r)\,P[\zeta(0),\zeta(r)], (2.16) where \zeta_{\rm c} is the threshold . The key point is the conservation of probability \displaystyle P[\zeta]{\rm d}\zeta \displaystyle= \displaystyle P[\zeta_{{\rm g}}]{\rm d}\zeta_{{\rm g}}, \displaystyle P[\zeta(0),\zeta(r)]{\rm d}\zeta(0){\rm d}\zeta(r) \displaystyle= \displaystyle P[\zeta_{{\rm g}}(0),\zeta_{{\rm g}}(r)]{\rm d}\zeta_{{\rm g}}(0){\rm d}\zeta_{{\rm g}}(r), (2.17) from which we deduce that the correlation function of the fully non-Gaussian curvature perturbation can be derived from the Gaussian counterpart, provided that one appropriately shifts the threshold \zeta_{{\rm cg}}=F^{-1}\left[\zeta_{\rm c}\right]. (2.18) For the example, from Eq. ( 2.15 ), \zeta_{{\rm cg}}=\mu\left(e^{-\zeta_{{\rm c}}/\mu}-1\right). (2.19) We conclude that we can calculate everything in terms of Gaussian probabilities, \displaystyle 1+\xi(r) \displaystyle= \displaystyle\frac{P_{2{\rm g}}}{P^{2}_{1{\rm g}}}, \displaystyle P_{1{\rm cg}} \displaystyle= \displaystyle\int_{\nu_{{\rm g}}}^{\infty}\frac{{\rm d}x}{\sqrt{2\pi}}e^{-x^{2}/2}={\rm Erfc}\left(\frac{\nu_{{\rm cg}}}{\sqrt{2}}\right), \displaystyle P_{2{\rm cg}} \displaystyle= \displaystyle\int_{\nu_{{\rm g}}}^{\infty}\frac{{\rm d}x_{1}}{\sqrt{2\pi}}e^{-x_{1}^{2}/2}\int_{\nu_{{\rm cg}}}^{\infty}\frac{{\rm d}x_{2}}{\sqrt{2\pi}}e^{-x_{2}^{2}/2}{\rm exp}\left[-\frac{x_{1}^{2}+x_{2}^{2}-2wx_{1}x_{2}}{2(1-w^{2})}\right], (2.20) where \displaystyle\nu_{{\rm cg}} \displaystyle= \displaystyle\frac{\zeta_{{\rm cg}}}{\sigma_{{\rm g}}},\quad\sigma_{{\rm g}}^{2}=\int\frac{{\rm d}k}{k}{\cal P}_{\zeta_{\rm g}}(k), (2.21) and w_{\rm g}(r)=\frac{\xi_{\rm g}(r)}{\sigma_{{\rm g}}^{2}}, (2.22) is the (normalized) two-point correlation of the Gaussian curvature perturbation. A valid approximation for \nu_{{\rm g}}\gg 1 has been found in Ref. [ 5 ] to be 1+\xi(r)\simeq(1+w_{\rm g})\frac{{\rm Erfc}\left(\sqrt{\frac{1-w_{\rm g}}{1+w_{\rm g}}}\frac{\nu_{{\rm cg}}}{\sqrt{2}}\right)}{{\rm Erfc}\left(\frac{\nu_{{\rm cg}}}{\sqrt{2}}\right)}. (2.23) An important remark is the following: the numerical value of the critical thresholds or of the shift from \zeta_{\rm c} to \zeta_{\rm cg} is not relevant. By the same argument of probability conservation, the PBH abundance for the fully non-Gaussian \zeta can be derived from the one-point probability of \zeta_{\rm g} for which only the value of \nu_{\rm g} is needed. This, in turn, is fixed phenomenologically by requiring that the PBH abundance has a given value. This fixes the parameter \nu , independently from the value of the threshold \zeta_{\rm cg} . Another interesting remark is the following. In the limit of large non-Gaussianities, \mu\ll 1 , the threshold of the non-linear curvature perturbation is in the non-Gaussian exponential tail of the probability when \zeta_{\rm c}\gg 1 , or \zeta_{\rm cg}\simeq\mu\ll 1 from Eq. ( 2.15 ). This means that, when the threshold for the curvature perturbation is in the non-Gaussian exponential tail, the correlation becomes independent from the formation threshold: when expressed in terms of the Gaussian counterpart through the conservation of the probability, the Gaussian threshold \zeta_{\rm cg} tends to zero. This simple reasoning confirms the result of Ref. [ 8 ] , that when the threshold of the non-linear \zeta is in the exponential tail region of the probability, the correlation looses its dependence on the threshold itself. What values of \nu_{\rm cg} should we use? The present abundance of DM in the form of PBHs per logarithmic mass interval \mathrm{d}\ln M is given by \displaystyle f_{\rm PBH}(M)\equiv\frac{1}{\rho_{{\rm DM}}}\frac{{\rm d}\rho_{\rm PBH}}{{\rm d}\ln M} \displaystyle\simeq \displaystyle\left(\frac{P_{1}}{6\cdot 10^{-9}}\right)\left(\frac{\gamma}{0.2}\right)^{1/2}\left(\frac{106.75}{g_{*}}\right)^{1/4}\left(\frac{M_{\odot}}{M}\right)^{1/2} (2.24) \displaystyle= \displaystyle\left(\frac{P_{1{\rm g}}}{6\cdot 10^{-9}}\right)\left(\frac{\gamma}{0.2}\right)^{1/2}\left(\frac{106.75}{g_{*}}\right)^{1/4}\left(\frac{M_{\odot}}{M}\right)^{1/2}, where \gamma\mathrel{\hbox to0.0pt{\lower 4.0pt\hbox{\hskip 0.5pt$\sim$}\hss}\raise 1.0pt\hbox{$<$}}1 is a parameter introduced to take into account the efficiency of the collapse and, for the masses of interest, the number of relativistic degrees of freedom g_{*} can be taken to be the SM value 106.75. In the second passage we have used the key information that P_{1}=P_{1\rm g} . By using again the conservation of the probability we can replace P_{1} with P_{1{\rm g}} . For PBH masses of the order of 10^{-19}M_{\odot} , potentially testable by ET, P_{1}=P_{1{\rm g}} should be of the order 10^{-19} to get f_{\rm PBH}(10^{-19}M_{\odot})\simeq 1 , corresponding to \nu_{\rm cg}\simeq 9 . Provided that the threshold is appropriately shifted, we have demonstrated that the correlation function of the full non-Gaussian curvature perturbation is identical to that of the Gaussian counterpart 1+\xi_{>\nu_{\rm c}}(r)=\frac{P_{2{\rm g}}}{P^{2}_{1{\rm g}}}=1+\xi^{\rm g}_{>\nu_{\rm cg}}(r), (2.25) where \xi^{\rm g}_{>\nu_{\rm cg}}(r) is the correlation of the thresholds of the Gaussian curvature perturbation. Given a profile of \zeta_{\rm g}(r) we can compute the corresponding value r_{m} at which the compaction function C_{1} has its maximum. For the profile, we take the averaged profile \langle\zeta_{\rm g}(r)\rangle_{\zeta_{\rm g}(0)}=\langle\zeta_{\rm g}(r)|\zeta_{\rm g}(0)\rangle=\frac{\langle\zeta_{\rm g}(r)\zeta_{\rm g}(0)\rangle}{\sigma_{\rm g}^{2}}\zeta_{\rm g}(0)=w_{\rm g}(r)\zeta_{\rm g}(0). (2.26) For the most optimistic case, in terms of clustering, of a broad spectrum {\cal P}_{\zeta_{\rm g}}(k)=A_{\rm b}\theta(k_{\rm max}-k)\theta(k-k_{\rm min})\,, (2.27) one has \displaystyle w_{\rm bg}(r) \displaystyle=\frac{1}{{\rm ln}(k_{\rm max}/k_{\rm min})}\left[{\rm Ci}(k_{\rm max}r)-{\rm sinc}(k_{\rm max}r)-(k_{\rm max}\rightarrow k_{\rm min})\right], (2.28) \displaystyle C_{1{\rm b}}(r) \displaystyle\simeq-\frac{4}{3}\zeta_{\rm g}(0)\left[{\rm sinc}(k_{\rm max}r)-{\rm sinc}(k_{\rm min}r)\right], (2.29) with k_{\rm max}r_{m}\simeq 3.5 . As shown in Ref. [ 76 ] , in the case of a broad spectrum the PBH mass function is peaked at the mass formed when the comoving wavelength k^{-1}_{\rm max} re-enters the comoving Hubble radius. The corresponding full non-Gaussian two-point correlations are plotted in Fig. 1 , where from now on, we identify the length of the power spectrum as k_{\rm max}\equiv 10^{\Delta}k_{\rm min} . Figure 1 : The two point correlation length \xi(r) from Eq. ( 2.23 ) for a broad USR power spectrum from Eq. ( 2.27 ) , fixing the benchmark value with \nu_{\rm cg}=9 and tuning the length \Delta . 3 The PBH clusterogenesis 3.1 The number of PBHs in the cluster At what minimum value of the radius one should evaluate the correlation function? In its original form, the thresholded regions are a continuous field and thus do not include any exclusion. This spatial exclusion arises because distinct PBHs cannot form arbitrarily close to each other. As a result, the conditional probability to find a PBH at a comoving distance r from another PBH, which is proportional to [1+\xi(r)] must vanish for some r\mathrel{\hbox to0.0pt{\lower 4.0pt\hbox{\hskip 0.5pt$\sim$}\hss}\raise 1.0pt\hbox{$<$}}r_{\rm exc} and the correlation function of thresholded regions should go to -1 for comoving distances smaller than r_{\rm exc} . Now, the physical meaning of r_{m} is the following. Imagine to have a peak in the Gaussian curvature \zeta . Given our assumptions, this has a spherical profile whose maximum value is at {\bf x}=0 . The corresponding compaction function C_{1}(r) has a maximum at a distance r_{m} from the peak of the curvature perturbation. It is the spherical region of radius r_{m} which will eventually collapse into a PBH upon horizon re-entry when r_{m}={\cal H}^{-1} , where {\cal H} is the comoving horizon scale. From such considerations, we conclude that the correlation function must be evaluated at least at the distance r_{\rm exc}\gtrsim 2r_{m}, (3.1) to avoid the superposition of two peak profiles overlapping. In order to determine if PBHs are born clustered at formation, we should calculate the average number of PBHs in the comoving volume V \langle N\rangle=\overline{n}_{\rm PBH}\int_{V}{\rm d}^{3}r\,\left[1+\xi(r)\right]=\overline{n}_{\rm PBH}V+\overline{n}_{\rm PBH}\int_{V}{\rm d}^{3}r\,\xi(r). (3.2) The mean count \langle N\rangle significantly deviates from Poisson if the second contribution from the clustering is larger than the discreteness noise \overline{n}_{\rm PBH}V . To quantify the importance of the former, we will compute the characteristic (comoving) clustering length r_{\rm cl} at which \xi(r_{\rm cl})\simeq 1 . However, the existence of a clustering length does not imply automatically that correlation is relevant. Indeed, from Eq. ( 3.2 ), the Poisson contribution increases like the volume centered around a given bubble and therefore scale like r^{3} . On the other hand, the piece from the correlation function scales slower than r^{3} . Therefore, even if the two contributions become of the same order at r_{\rm cl} , one has to also estimate the average number of bubbles in a volume of size the correlation length, that is \overline{n}_{\rm PBH}r_{\rm cl}^{3} . Since \overline{n}_{\rm PBH}\sim P_{1}{\cal H}^{3}=(P_{1}/r_{m}^{3}) , all in all, the average number of PBHs in the cluster is \langle N\rangle\equiv\frac{4\pi^{2}P_{1}}{r_{m}^{3}}\int_{r_{\rm exc}}^{r_{\rm cl}}{\rm d}r\,r^{2}\,\xi(r)=4\pi^{2}P_{1}\int_{2}^{r_{\rm cl}/r_{m}}{\rm d}x\,x^{2}\,\xi(x)\mathrel{\hbox to0.0pt{\lower 4.0pt\hbox{\hskip 0.5pt$\sim$}\hss}\raise 1.0pt\hbox{$>$}}1, (3.3) with x\equiv r/r_{m} . The corresponding mass of the cluster is simply given by M_{\rm cl}\equiv\langle N\rangle M_{\rm PBH} . 3.2 Clusterogenesis Once the cluster has formed, it may form a virialized halo even during the radiation phase. Indeed, it is usually believed that overdensities may not grow during the radiation phase because of the presence of the radiation pressure. However, this is true only at the linear level. If the overdensities are large enough, the self gravity of these large non-linear fluctuations may become sizeable before the equality time, and produce, after they decouple from the Hubble flow, very dense clusters [ 65 ] . Adopting for simplicity, a spherical collapse model, the evolution equation for the parameter R , describing the deviation of the motion of each collapsing shell from the uniform Hubble flow of the background Friedmann universe reads [ 65 ] x(1+x)\frac{{\rm d}^{2}R}{{\rm d}x^{2}}+\left(1+\frac{3}{2}x\right)\frac{{\rm d}R}{{\rm d}x}+\frac{1}{2}\left(\frac{1+\Phi}{R^{2}}-R\right)=0, (3.4) where \Phi=\frac{\delta n_{\rm PBH}}{\overline{n}_{\rm PBH}}\simeq\langle N\rangle, (3.5) we have assumed that all the DM is in PBHs and x=a/a_{\rm eq} , in terms of the one at the epoch of matter-radiation equality a_{\rm eq} . For small values of x , one may obtain analytically R\simeq 1-\frac{\Phi x}{2}-\frac{\Phi^{2}x^{2}}{8}+\mathcal{O}(x^{3}). (3.6) The decoupling from the Hubble flow takes place at a_{\rm cl}=a_{\rm eq}/\Phi , well in the radiation phase. The physical radius of the cluster is [ 39 ] R\simeq 4\cdot 10^{-5}\langle N\rangle^{-1}\left(\frac{C}{200}\right)^{-1/3}\left(\frac{r_{\rm cl}}{\rm kpc}\right){\rm kpc}, (3.7) where C=\mathcal{O}(1-10^{2}) parametrizes the amplitude of the final overdensity upon virialization [ 64 , 65 ] . PBH clusters may collapse into BHs of mass M^{\rm final}_{\rm PBH}\simeq M_{\rm cl}\equiv\langle N\rangle M_{\rm PBH} if the final halo is sufficiently more compact than a BH, thereby giving rise to a population of heavier PBHs. This condition is described by the Hoop conjecture [ 74 ] R<2GM^{\rm final}_{\rm PBH}, (3.8) or [ 41 ] \langle N\rangle>6\cdot 10^{4}\left(\frac{C}{200}\right)^{-1/6}\left(\frac{r_{\rm cl}}{\textrm{kpc}}\right)^{-1}. (3.9) In our case, r_{\rm cl}\simeq k^{-1}_{\rm min}=10^{\Delta}k^{-1}_{\rm max} , and the relation between the mass of the lightest PBHs and the scale k_{\rm max} is given by [ 81 ] M_{\rm PBH}\simeq 2.5\cdot 10^{6}\left(\frac{k_{\rm max}}{\textrm{kpc}^{-1}}\right)^{-2}M_{\odot}\,. (3.10) with k_{\rm max}\simeq 5\cdot 10^{13}\textrm{kpc}^{-1} in order to get M_{\rm PBH}\sim 10^{-19}\ M_{\odot} 1 1 1 Notice that in this paper we are interested in asteroid-like PBH masses, much lighter than the ones discussed in Refs. [ 39 , 37 ] . . Setting for simplicity C=200 , the Hoop conjecture can then be recast in terms of the width of the power spectrum \Delta and the PBH mass as \langle N\rangle\gtrsim 50\cdot 10^{-\Delta}\sqrt{\frac{M_{\odot}}{M_{\rm PBH}}}\,. (3.11) As shown in Fig. 2 , satisfying the Hoop conjecture requires a rather extended power spectrum, with k_{\rm max}/k_{\rm min}\sim 10^{6.5} . Another important requirement is that PBHs, which typically evaporate on a timescale [ 33 ] \tau_{\rm eva}\approx 1.0\times 10^{65}\left(\frac{M_{\mathrm{PBH}}}{M_{\odot}}\right)^{3}\,\text{yr}\,, (3.12) do not evaporate before the characteristic free-fall time that sets the formation time of the heavier PBHs 2 2 2 An additional constraint arises from the dissipation time of the cluster due to repeated encounters among its constituents, which may lead to its disruption before collapse into heavier PBHs [ 26 ] . This effect is unrelated to Hawking evaporation. We neglect it here since, as shown in Ref. [ 41 ] , it is less stringent than the Hoop conjecture. [ 26 ] : \tau_{\rm cl}=\sqrt{\frac{3\pi}{32G\rho_{\rm cl}}}\simeq 1.2\cdot 10^{4}\langle N\rangle^{-2}\left(\frac{C}{200}\right)^{-1/2}\,\text{yr}\,. (3.13) Figure 2 : The average number of PBHs of masses \sim 10^{-19}\ M_{\odot} inside a volume of size r_{\rm cl}\sim k^{-1}_{\rm min} for different power spectrum lenghts \Delta . The red line set the minimum number of PBHs in the cluster such to satisfy the Hoop conjecture, i.e. Eq. ( 3.11 ). We set for simplicity C=200 . For M_{\rm PBH}\sim 10^{-19}M_{\odot} and k_{\rm max}/k_{\rm min}\sim 10^{6.5} , which implies \langle N\rangle\gtrsim 10^{5} , one can easily verify that, for any value C\in[1,100] , the condition \tau_{\rm cl}\ll\tau_{\rm eva} is satisfied 3 3 3 The same conclusion holds for spinning BHs, whose evaporation time is approximately a factor of two shorter than in the non-spinning case [ 9 ] . . In order for PBHs to account for the totality of DM, the final mass must lie in the range [ 33 ] 10^{-16}M_{\odot}\lesssim M^{\rm final}_{\rm PBH}\lesssim 10^{-11}M_{\odot}\,. (3.14) Starting from M_{\rm PBH}\sim 10^{-19}M_{\odot} , this requirement translates into a range for the width of the power spectrum, 6.5\lesssim\Delta\lesssim 8\,. (3.15) Such broad spectrum can be obtained in the context of curvaton models [ 47 ] and within USR models with a prolonged constant roll phase [ 66 , 49 ] . In the next subsection, we investigate the implications for detection prospects in current and future experiments. 4 Implication for detection with Einstein Telescope As already mentioned, PBHs production is related to a signal of induced GWs. Such GWs are produced as a second order effect by scalar perturbations which re-enter the horizon and collapse to form a PBH [ 87 , 72 , 3 , 75 , 6 , 24 ] (see also [ 61 ] for a recent physical explanation of the origin of the signal). The spectral density of induced GWs associated to PBHs production can be computed as [ 46 ] \displaystyle h^{2}\Omega_{{\rm GW}}(f)= \displaystyle\frac{c_{g}h^{2}\Omega_{r}}{36}\int_{0}^{\frac{1}{\sqrt{3}}}{\rm d}t\int_{\frac{1}{\sqrt{3}}}^{\infty}{\rm d}s\left[\frac{\left(t^{2}-1/3\right)\left(s^{2}-1/3\right)}{t^{2}-s^{2}}\right]^{2} \displaystyle\left[\mathcal{I}^{2}_{c}(t,s)+\mathcal{I}_{s}^{2}(t,s)\right]\mathcal{P}_{\zeta_{\rm g}}\left[\frac{k\sqrt{3}}{2}(s+t)\right]\mathcal{P}_{\zeta_{\textrm{g}}}\left[\frac{k\sqrt{3}}{2}(s-t)\right], (4.1) with c_{g}\equiv\frac{g_{*}(M_{H})}{g_{*}^{0}}\left(\frac{g_{*S}^{0}}{g_{*S}(M_{H})}\right)^{4/3}\approx 0.4 (4.2) ( g_{*S} and g_{*} being the effective degrees of freedom of thermal radiation and the superscript 0 stands for present day values), \Omega_{r} current radiation density and \mathcal{I}_{c}(t,s) , \mathcal{I}_{s}(t,s) are transfer functions computed analytically assuming a perfect radiation cosmological background (see, for example, Ref. [ 46 ] ). Notice therefore Eq. ( 4 ) is strictly valid only during radiation domination. In principle, we should also consider the evolution of the relativistic degrees of freedom as the Electroweak phase transition takes place. We neglect this effects and just assume the Universe to be radiation dominated at all the scales relevant for GWs production. We also neglect possible primordial non-Gaussian corrections in the computation of the scalar-induced GW signal, which we leave for future work (we refer to refs. [ 32 , 30 , 88 , 31 , 54 , 11 , 90 , 43 , 52 , 68 , 91 , 67 , 4 , 77 , 58 , 60 , 59 ] for a discussion about these effects). Figure 3 : Signal of second order GWs associated with the broad power spectrum obtained within the two benchmark cases relevant in this work. We plot the power-law-integrated sensitivity curves (thick dashed lines) for three configurations of {\rm ET} : the triangular configuration ( {\rm ET}_{\rm TR} ), the 2L-aligned configuration ( {\rm ET}_{\rm LLa} ) and the 2L-misaligned ( {\rm ET}_{\rm LLm} ) (in the sense discussed in the main text). We assume 1 yr of observation and Signal-to-noise ratio 10 . We also plot the power-law-integrated sensitivity curves (dashed thin lines) of {\rm LLR} and {\rm eLO} [ 48 , 27 ] , {\rm LISA} [ 10 ] and {\rm AION-km} [ 17 , 18 , 2 ] . In Fig. 3 we report the gaussian SIGW signal for the relevant cases discussed in the previous section, i.e. a broad power spectrum (Eq. 2.27 ) with k_{\rm max}=5\cdot 10^{13}\textrm{kpc}^{-1} and k_{\rm min}\equiv 10^{-\Delta}k_{\rm max} with \Delta\sim(6.5,8) . Since we want to be general about the specific non-gaussian parameter values in the relation of Eq. ( 2.15 ), the precise values of the amplitude of the power spectrum A_{\rm b} in order to get f_{\rm PBH}\sim 1 is unknown but reasonable it lies in the range \sim(10^{-3}-10^{-2}) . For simplicity, as benchmark value, in the plot, we use A_{\rm b}=5\cdot 10^{-3} . Assuming a Signal-to-Noise ratio equal to 10 and a time of observations of 1 year, we assess the detectability of the SIGW signals at ET. We consider ET in its standard “xylophone” layout, in which each detector is made of a high-frequency (HF) interferometer and a low-frequency (LF) interferometer, and we consider the different geometries currently under study [ 29 ] , namely either a single-site observatory with a triangular configuration (ET TR ) made of three nested detectors, each made of a LF and a HF interferometer, and 10 km-long arms; or a network of two L-shaped instruments (again, each made of a LF and a HF interferometer) with 15 km-long arms each. 4 4 4 We follow the terminology of the 2020 ET Design Report Update, https://apps.et-gw.eu/tds/?r=18715 : the high-frequency (HF) and low-frequency (LF) interferometers of the xylophone configuration are indeed referred to as “interferometers”. The combination of a HF interferometer and a LF interferometer (whether in a L-shaped geometry, or with arms at 60^{\circ} as in the triangle configuration) is called a “detector”. The whole ensemble of detectors is called an “observatory”. So, ET in the triangle configuration is made of three detectors for a total of six interferometers, while in the 2L configuration it is made of two detectors, for a total of four interferometers. Currently, there are three candidate locations, which could in principle host either a triangle, or an L-shaped detector in a 2L configuration: the Sos Enattos site in Sardinia; the Euroregion Meuse-Rhine (EMR) across Belgium, Germany and the Netherlands; and the Lusatia region in Saxony (Germany). For definiteness, we place the triangle configuration in Sardinia, and the 2L in Sardinia and in the EMR region. 5 5 5 Note, however, that, for a 2L configuration, placing the second detector in the candidate Saxony region would not affect the results, as the distance between the EMR and Saxony sites relative to the Sardinian location is quite similar. As the sensitivity to a stochastic background strongly depends on the relative orientation between detector pairs, we further consider scenarios in which the two L-shaped ET detectors are either aligned (ET LLa ) or misaligned (ET LLm ). As in Ref. [ 29 ] , to take into account the curvature of the Earth, the relative orientation between the two detectors is defined with reference to the great circle that connects the two detectors. Denoting by \beta the angle describing the relative orientation of the two detectors, defined with reference to this great circle, \beta=0^{\circ} corresponds to the case where the arms of the two interferometers make the same angle with respect to the great circle, while \beta=45^{\circ} corresponds to the case where one of the two interferometers is rotated by 45^{\circ} from the \beta=0^{\circ} orientation. For \beta=0^{\circ} (with we refer as the “2L-aligned” configuration, ET LLa ) the sensitivity to stochastic backgrounds is maximized, while for \beta=45^{\circ} it is minimized, to the extent that the 2L configuration becomes completely blind to stochastic backgrounds in the limit fd\rightarrow 0 , where f is the GW frequency and d the distance between the two detectors. This is the opposite of what happens for compact binary coalescences (CBCs), where the accuracy of parameter reconstruction is maximized for \beta=45^{\circ} . However, as discussed in Section 2 of Ref. [ 29 ] , already a small misalignment angle, of order of a few degrees, allows us to recover an interesting sensitivity to stochastic backgrounds, without basically affecting CBC reconstruction. In the following we use the same small misalignment angle as in [ 29 ] , of about 2.5^{\circ} , and we refer to this configuration as “2L-misaligned”, ET LLm . Results for larger values of this relative angle will be intermediate between the results shown for the aligned and misaligned configuration. We therefore consider a total of three configurations (triangle, 2L-parallel and 2L-misaligned), for which the power-law-integrated sensitivity curves, as defined in [ 86 ] (see also App. A of [ 29 ] for conventions and definitions) are computed using the public code GWFAST [ 56 ] . The results are shown in Fig. 3 . Both corresponding SIGW spectra, with the lightest final masses respectively of M^{\rm final}_{\rm PBH}\sim 10^{-11} ( \Delta=8) ) and M^{\rm final}_{\rm PBH}\sim 10^{-14} ( \Delta=6.5) ), lies in the band range of ET and at the same time may be tested simultaneously by future observatories such as LISA [ 10 ] , AION-km [ 17 , 18 , 2 ] and, in the case of the broadest spectrum, also by the eLO experiment [ 48 ] . 5 Conclusions Giving an identity card to the DM is one of the main tasks in modern astroparticle physics. One attractive candidate is represented by PBHs since this would not require physics beyond the Standard Model in terms of particle content. In this paper we have made the observation that the DM might be due entirely by asteroid-like PBHs in the mass range ( 1.1 ), originated by the collapse of much lighter PBHs; in fact so light that they would have evaporated otherwise. During the formation of such light PBHs, an ineludible SIGW stochastic background is generated with frequencies falling not only in the LISA range, but also in the ET range, thus opening the possibility of revealing the presence of DM, albeit indirectly, by means of GW observations.