WorksPhysics and Astronomy
Little Red Dots from Primordial Vacuum Bubbles
Abstract
Quantum tunneling transverse to the inflaton trajectory can produce – primordial black holes at the abundance of JWST's little red dots, without the enhanced curvature perturbations excluded by limits on CMB spectral distortions. Rare vacuum bubbles nucleate during a brief interval of inflation and later form black holes, with adjustable mass and abundance and a narrow mass spectrum of percent-level width. The leading constraint, from accretion, is satisfied for a population matching the bright little red dots even if all the emitted radiation escapes as hard X-rays.
Introduction. The James Webb Space Telescope (JWST) has revealed compact, high-redshift sources known as little red dots (LRDs), several of which exhibit broad Balmer emission lines consistent with accreting black holes [1, 2]. Standard virial analyses of broad-line LRD samples infer black-hole masses of approximately –, while modeling of the host emission places several sources above the local black-hole–stellar-mass relation [2]. For Abell 2744-QSO1 at , a virial estimate of about [3] is consistent with a subsequent dynamical measurement near [4]. Primordial black hole (PBH) formation within a factor of a few of the observed masses would substantially reduce the required accretion growth and its sensitivity to uncertain feedback physics. This is particularly relevant for systems such as QSO1, where growth histories must also remain compatible with the low stellar content and metallicity [5, 6]. Cosmological simulations with a massive primordial seed can reproduce several properties of QSO1 with only modest subsequent mass growth [5].
Forming PBHs directly at LRD masses from enhanced, approximately Gaussian curvature perturbations (i.e. the usual PBH production mechanism from inflation) is strongly constrained by CMB spectral distortions. Because collapse occurs only in the tail of the distribution, a substantial abundance requires a large curvature variance, and the many fluctuations that fail to collapse are erased by Silk damping, injecting energy that distorts the CMB spectrum [7]. The -distortion era, approximately , is primarily sensitive to modes with comoving wavenumber , a range that includes the scales associated with LRD-mass PBHs in the standard radiation-era mapping [8, 9]. The precise mass range constrained depends on the perturbation spectrum and collapse prescription, as well as the gradual boundaries of the distortion window [7, 8]. For approximately Gaussian perturbations, the FIRAS constraint [10, 11] sharply suppresses the primordial abundance above an initial seed-mass scale of roughly – in the scenarios considered in Ref. [6], well below the black-hole masses inferred for LRDs.
This has motivated lighter primordial seeds that form on smaller scales and grow by accretion or mergers [6, 12], and mechanisms that decouple the PBH abundance from the curvature power, such as rare regions with distinct inflationary histories [8], strongly non-Gaussian perturbations [9, 13, 14, 15], or false-vacuum domains surviving a catalyzed transition [16]. Here we point out that vacuum bubbles nucleated during inflation can produce supermassive PBHs with no enhancement of the curvature perturbations, avoiding the spectral-distortion bound on enhanced curvature power [17, 18, 19, 20]. The bubbles nucleate by tunneling in a field direction transverse to the inflaton trajectory (Figure 1), expand to super-horizon size, and form black holes when they re-enter the horizon after inflation. Each PBH comes from a single, rare tunneling event, whose probability is set by an instanton action and not by the amplitude of perturbative fluctuations. The inflaton perturbations keep their usual slow-roll amplitude near the transition point, and the transverse field is heavy, with no cosmologically significant fluctuations of its own apart from the bubbles. The black-hole mass is exponentially sensitive to the number of -folds remaining when the bubble nucleated. In earlier work, two of us showed that in a broad class of models nucleation is naturally localized near a particular value of the inflaton, yielding an approximately monochromatic PBH spectrum [19]. Here we calculate the nucleation rate and the width of the mass spectrum in a controlled approximation, match the abundance to the LRDs, and check the constraints. While we focus on vacuum bubbles, domain walls or other seeds produced by non-perturbative transitions during inflation could serve as well [17, 19].
Figure 1. Illustrative two-field potential with a rolling valley (the inflaton trajectory) and a nearby basin (“lake”) separated by a barrier [19]. The arrow along the valley indicates the inflaton slow-roll trajectory. The dashed line is a schematic for the instanton across the barrier, and is the value of the transverse field at the basin minimum. The tunneling probability across the barrier peaks at . In our scenario, regions of space where the fields tunnel to the “lake” eventually become PBHs, as explained in the main text.
Supermassive PBHs have been proposed as the origin of LRDs [21], but whether LRDs require primordial seeds is not settled: astrophysical heavy seeds may suffice [22, 23], supermassive Dark Stars may collapse to black holes [24, 25, 26], and the black-hole masses themselves are debated [27]. Regardless, the mechanism of this paper supplies supermassive PBHs without the constraints that exclude the more standard approach.
We set the speed of light and to unity and normalize the scale factor to today. Here is the reduced Planck mass. Number densities are comoving, and overdots denote derivatives with respect to cosmic time .
Population targets. Let be the observed comoving density of a selected LRD sample. We consider two LRD number-density targets. One of them is the fiducial number density: the rest-frame UV luminosity functions motivate an integrated-population range – [2, 28]. The other is the bright end of the distribution, a useful lower limit for the number density, [2]. The sample selections and bin sums are given in the Supplemental Material. For each target we consider a single PBH population with a narrow mass spectrum that accounts for the corresponding LRD sample; the observed masses follow from the accretion histories considered below. The comoving number density of this population, , is conserved if mergers are neglected, and we define as the fraction of these PBHs that appear in the selected LRD sample,
Thus gives the minimum PBH number density needed to match the observed LRD density; smaller requires more PBHs.
Before recombination, Compton drag strongly suppresses infall onto PBHs. Recombination reduces the free electron abundance and weakens this drag, so we neglect the suppressed pre-recombination accretion and begin our accretion histories at [29]. Let be the PBH mass at and the fraction of dark matter in the form of PBHs at that time. Neglecting growth before , is equal to the PBH mass at formation. The comoving abundance and the fraction required to match the LRD density are
Here is the present-day dark matter density, equal to the co-moving dark matter density [30]. Subsequent accretion is described in terms of the mass-growth factor as , with , so an observed mass must be mapped back to before applying the abundance constraint.
Localized production. We describe a narrow PBH mass function by its characteristic formation mass , logarithmic width and comoving number density . In our scenario, all three are determined by how the nucleation rate (in units of ) of vacuum bubbles depends on the inflaton. It reaches its maximum at a preferred inflaton value , which the inflaton crosses e-folds before the end of inflation. Throughout, a subscript denotes a quantity evaluated at , or equivalently at . For a displacement , the rate is [19]
where is the radius of a bubble nucleated at and is a dimensionless localization coefficient.1
The value of determines the size of these bubbles at collapse, and hence the peak PBH mass Eq. (5). The combination fixes the duration of production and the width of the peak Eq. (4). The peak rate , integrated over that duration, fixes the abundance Eq. (7). Nucleation events are independent and produce a population of PBHs that is Poisson-distributed in space.
A simple realization is a two-scalar field model of the type considered in Ref. [19], in which a slow-roll valley (the inflaton trajectory) is separated by a transverse barrier from a lower-energy basin Figure 1. Most regions remain in the valley, while rare transitions populate the basin. With the techniques of Ref. [31], and the leading semiclassical contribution to the tunneling rate can now be calculated in a controlled approximation. We summarize that calculation below; the full derivation and an explicit two-field potential are given in the Supplemental Material.
At approximately constant and , we can relate the field displacement to the change in e-folding number as , where is the number of e-folds remaining in inflation. Inserting this into Eq. (3) and using the dimensionless scalar amplitude , we obtain a tunneling rate that is a Gaussian function of the number of e-folds
tionary dilution.
Here is the width of the nucleation-rate peak in -folds and counts -folds remaining until the end of inflation. For nearly constant , the probability of nucleation in an -fold interval is approximately . As seen in Eq. (4), if the amplitude of the curvature power spectrum during bubble nucleation is taken to be similar to the CMB-scale amplitude [30], the width is naturally small, but could be increased significantly for smaller or larger [19].
A bubble nucleated -folds before prompt reheating has radius at reheating, and collapses to a black hole when it re-enters the horizon during radiation domination [17, 18], so that
Here is the initial primordial black hole mass. All masses and widths quoted below are values at formation. The normalization of Eq. (5) carries an order-unity uncertainty from the collapse and depends on the reheating history [18, 19]. For the scale used below, the interval – corresponds to –. For the Gaussian profile in Eq. (4), the number-weighted standard deviation3 is , including inflationary dilution.
Abundance requirements. Assuming one PBH forms per bubble, the integrated nucleation probability required to match the observed LRD density is, for a monochromatic population,4
For the narrow spectrum considered here, we evaluate Eq. (6) at the peak initial mass, . The nucleation history predicts
Here is the nucleation probability per -fold, and the exponential accounts for the additional dilution of bubbles nucleated earlier than . We require . For the narrow Gaussian profile of Eq. (4), the dilution factor varies little across the production interval, leading to . Writing the physical nucleation rate per unit volume and time as , with the bubble radius and using , gives
where is the dimensionless Euclidean action difference and is the dimensionless fluctuation prefactor [32, 33]. The abundance therefore constrains the action together with the bubble radius and duration of production.
Rate from the rolling background. Let denote the value of the inflaton on the slow-roll trajectory at the time a bubble nucleates, so that far from the bubble the fields are , with the transverse field. Since the inflaton rolls, is not a stationary point of the potential and cannot serve directly as the boundary value of a Euclidean bounce. For any homogeneous background , where is Euclidean time, that obeys its equation of motion, however, the action of a localized fluctuation about it is [31, 34], with5
When the background changes little over the size of the bubble, it can be approximated as frozen at its value at nucleation, . Equation (9) then defines an ordinary tunneling problem with a stationary point at : the slope of the valley cancels identically, its curvature survives only as a negligible mass for the inflaton fluctuation, and the dependence on arises entirely from the coupling between the two fields, which localizes the transition. The action entering the peak rate in Eq. (8) is the Coleman action of at . Expanding the action to second order in gives
which leads directly to Eq. (3). Here is defined by the response of the inflaton to a change in its value far from the bubble: outside the bubble this response is a harmonic function in four Euclidean dimensions, , where is the Euclidean radial coordinate. Freezing the background is accurate when the inflaton barely moves while a bubble forms. A bubble forms in a time of order , a fraction of an e-fold, while production lasts e-folds, so the expansion parameter is for the benchmark below. At first order in the time dependence enters the Euclidean exponent only through a term odd in Euclidean time, which integrates to zero on the time-symmetric bounce. The leading correction is second order, . The derivation, error budget, and numerical procedure are given in the Supplemental Material.
As an example, for the explicit potential specified in the Supplemental Material [Eq. (S6)], with , the same reference curvature power , and , matching the scales and epoch to the bright sample with and gives
Here is the Euclidean radius at which the transverse field has fallen halfway from its value at the bubble center to its value far from the bubble, , i.e. . We have not calculated the prefactor, which would shift by a few. Table I illustrates three cases in which the number density of PBHs matches the bright LRD abundance with , for three masses . Adjusting the potential normalization, field scale, and production epoch for each target mass, while keeping the dimensionless potential shape fixed, we find that their mass widths are close to 4%. The , , solution is our bright-sample benchmark, marked by the star in Figure 2.
Figure 2. Initial PBH mass fraction versus initial mass at accretion onset, . The teal band and line show the abundances required by Eq. (2) for the integrated LRD population, –, and the bright subset, [2, 28], both for (every PBH appears as an LRD). For both shift upward by . Horizontal lines are our estimated CMB upper bounds on from Eq. (13): solid for Hubble-limited accretion with , , and from bottom to top, and dashed for Eddington-limited accretion with and . The right axis gives the largest allowed at each for Hubble-limited accretion. The green Bondi–Hoyle–Lyttleton (BHL) [45] and orange Park–Ricotti (PR) [46, 47, 55] curves are the published accretion bounds compiled in Ref. [6], shaded over their stated mass ranges. The star corresponds to the benchmark point shown in the middle row of Table I.
For the full-sample reference , the corresponding actions range from 56.1 to 49.3 as the mass increases, while . Each matching defines a
| 40.15 | 60.6 | 2.44 | 0.0401 | |
| 41.30 | 57.2 | 2.41 | 0.0404 | |
| 42.45 | 53.8 | 2.39 | 0.0408 |
Table I. Abundance-matched parameter sets for and , with , , and the dimensionless potential parameters of Eq. (19). The masses listed are PBH masses at formation. Each row denotes a separate choice of dimensional scales and production epoch. Actions and widths are obtained by matching the numerical bounce solutions to the target abundance set by JWST LRD data.
separate narrow population, and the small rates make bubble overlap negligible. The Supplemental Material gives the scale matching and its mass dependence.
Spectral distortion and other bounds
The PBH formation mechanism of this paper does not produce significant CMB spectral distortions. The enhanced Gaussian curvature fluctuations of the standard PBH formation mechanism source acoustic waves whose dissipation by Silk damping over the approximate redshift interval produces a dimensionless chemical-potential distortion [7, 35]. Defining as the sky-averaged distortion, a reanalysis of the COBE/FIRAS data gives at 95% confidence [11], constraining the curvature variance on the scales relevant for Gaussian PBH formation. The tunneling mechanism of this paper, on the other hand, sets the abundance through a rare transition and does not require enhanced curvature variance.
Bubble walls can nevertheless launch shocks and compensating underdensity waves. The idealized reflecting-wall estimate of Ref. [36] gives
The mass entering this formation-related estimate is the primordial PBH mass, before accretion begins. At , this estimate lies well below the FIRAS limit for both population targets. Equation (12) estimates the sky-averaged distortion, bounded by the FIRAS monopole limit; the Planck -map analysis instead constrains the statistical occurrence of rare, localized patches with unusually large . Under Poisson spatial statistics, the largest expected patch at the full-sample, upper-mass reference has , far below the approximate rare-patch limit inferred by Deng et al. [36] from the Planck -map distribution of Ref. [37]. The acoustic and occupancy assumptions are detailed in the Supplemental Material. For the narrow spectrum considered here, replacing in Eq. (12) by the number-weighted moment gives a correction of only within the Gaussian approximation.
Because the PBHs are discrete and Poisson distributed, they produce an irreducible white-noise PBH isocurvature component even without an enhanced Gaussian curvature spectrum [38, 39, 40]. Applying the CMB+BAO limits of Ref. [40] to the white-noise spectrum , we find an amplitude that is at least four orders of magnitude below current sensitivity for all masses and abundances considered here at . Since the amplitude scales as , our model remains compatible with these limits for . Therefore, our model safely evades bounds from -distortions and isocurvature.
Random nucleation also implies that the PBHs cluster like matter on large scales, with bias [41, 42]. Improved clustering measurements could distinguish this scenario from models in which LRDs preferentially occupy strongly biased massive halos.
Accretion. The most important constraint that does not depend on how the PBHs formed comes from accretion: radiation from gas falling onto the black holes after recombination ionizes the intergalactic gas and alters the CMB temperature and polarization spectra [43, 44]. Published bounds assume quasi-steady inflow, which fails above about where the inflow time approaches the expansion time [29, 45]. The accretion physics at higher masses is uncertain [46, 47, 48, 49, 50, 51]. We do not attempt a systematic treatment here. Instead we ask how aggressive the assumptions must be for the constraint to matter, using two histories that begin at and neglect feedback. In Hubble-limited accretion baryons are supplied at the rate , the mean baryon flux through the radius at which the gravitational time equals the Hubble time ,6 and we assume that all of it reaches the black hole. The luminosity may then exceed the Eddington luminosity ; for radiative efficiency , at . In Eddington-limited accretion the inflow is reduced so that . The CMB responds to the part of the luminosity that escapes the source as hard X-rays. We write that part as with , where is the fraction of the luminosity emitted as an escaping hard continuum. Taking gives the strongest constraint. Normalizing our estimate of the CMB response to the Planck-based limit of Ref. [45] for continuous Eddington emission gives (Supplemental Material)
These are estimates that depend on the assumed history and spectrum, not a new likelihood analysis.
Figure 2 compares these limits with the abundances required by Eq. (2). For and , the bright LRDs are allowed under either history even if the entire luminosity emerges as hard X-rays. The full population is allowed under Eddington-limited accretion if –, and under Hubble-limited accretion if –. These requirements tighten in proportion to , where a mass observed at redshift corresponds to , with growth factor – at . For comparison, observed active nuclei emit of order of their luminosity in hard X-rays [52] and LRDs are X-ray weak [53, 54], although the relevance of either to isolated black holes before reionization is uncertain.
Discussion. Rare vacuum bubbles nucleated during slow-roll inflation can produce PBHs with the mass and abundance needed to explain LRDs. We computed the tunneling rate in a field direction transverse to that of the slow-roll motion of the inflaton using a controlled adiabatic approximation. An explicit two-field realization, shown in Figure 1, leads to a narrow PBH mass spectrum with logarithmic width . The peak PBH mass is set by the time during inflation where tunneling is most effective, while the abundance is set by the magnitude of the peak tunneling rate; thus they can be independently varied. Because this mechanism does not rely on enhanced curvature perturbations, it is not subject to the spectral-distortion bounds on such enhancements. Distortions sourced by the bubbles themselves are negligible at the required abundance, as are formation-related Poisson isocurvature perturbations. An alternative realization of the orthogonal tunneling mechanism within the chain inflation framework [56, 57], where the universe tunnels through a series of ever lower energy vacua, will be presented separately [58].
The most important remaining constraint, from accretion after recombination, is satisfied for the bright LRDs under conservative assumptions. For the full population at it requires that less than 0.2–1% of the accreted rest-mass energy escape as hard X-rays for Hubble-limited accretion, or less than 10–50% of the luminosity for Eddington-limited accretion.
In summary, we have proposed a mechanism that produces supermassive PBHs by tunneling transverse to the inflationary trajectory. These PBHs could explain the little red dots discovered by JWST without violating CMB spectral-distortion bounds, and are consistent with accretion bounds for the bright LRD population.
Acknowledgments. We thank Yacine Ali-Haïmoud, Matthew C. Johnson, and Anthony Taylor for useful discussions. KF and EIS are grateful for support from the Jeff & Gail Kodosky Endowed Chair in Physics at the University of Texas. KF and EIS acknowledge support by the U.S. Department of Energy, Office of Science, Office of High Energy Physics program under Award Number DE-SC0026683. The work of MBV is supported by FCT (Fundação para a Ciência e Tecnologia, Portugal) through the grant 2024.00457.BD with DOI identifier https://doi.org/10.54499/2024.00457.BD. Centro de Física do Porto is partially funded by Fundação para a Ciência e Tecnologia (FCT) under the grant UID04650-FCUP. MK and CN are supported by NSF grant PHY2412899.
References
Supplemental Material
Population selections
For a specified sample and redshift interval, the comoving LRD density follows from the rest-frame UV luminosity function , where is the absolute AB magnitude at 1450 Å [2],
Here is the bin-averaged luminosity function and is the bin width. Summing the one-magnitude bins in Table 4 of Ref. [2] gives approximately at and at , with bin centers extending from to and , respectively. Together with the differently selected COSMOS-Web density of [28], these sums motivate the integrated-population range used in the paper. The bright-subset reference is motivated by the bins, whose densities are close to this value in both redshift intervals [2]. These selections define alternative abundance targets without imposing a magnitude-to-mass relation or identifying the full observed mass range with one primordial peak.
Dilution and width conventions
At constant , define the raw nucleation probability and its dilution-weighted counterpart by integrating over the production window,
For prompt reheating followed by entropy-conserving cooling, the reheating temperature and present comoving bubble density then satisfy [19]
Here is the present photon temperature [59]. The quantities and count the effective relativistic degrees of freedom in energy and entropy, with subscripts rh and 0 denoting reheating and today. We use the Standard Model values and . Combining Eq. (16) with gives , which explains the cancellation of in Eq. (6). The same mass relation places the central benchmark at .
The Gaussian approximation in Eq. (4) gives
The first expression is the raw probability . The action estimate neglects the difference between and in the narrow-window limit. Number dilution introduces a factor , shifting the mean of the Gaussian without changing its variance. Since , its moments, with averages weighted by , are
The characteristic interval quoted for localization is therefore distinct from the standard deviation. The relation provides a check on the numerical calculation, which retains dilution explicitly.
Local numerical production benchmark
The results of the paper rely on Eq. (3), not on a particular Lagrangian. To calculate and explicitly we consider a canonical inflaton and transverse field , with potential . The field scale defines the dimensionless coordinates and , in which12
Here is the valley energy at , while the dimensionless coupling sets the normalization of . The bias lowers the transverse basin while preserving a barrier, and couples the two fields. Along the valley , and set the slope and curvature of . We choose as local effective-potential parameters for the numerical benchmark, and write for the dimensionless inflaton value at nucleation. At , the transverse valley curvature is , so the chosen bias preserves the barrier while allowing a lower basin.13
Bounce about the rolling background
Because the inflaton rolls along , the far-field value is not a stationary point of , and a Euclidean solution asymptoting to it would run off along the slope. The problem is instead well-posed for fluctuations about the rolling background. Let be the continuation to Euclidean time of the homogeneous rolling background, which solves its own equation of motion, and write with localized in space and Euclidean time. Integrating the cross term by parts, the equation of motion for cancels the term linear in , and the boundary terms at early and late Euclidean time cancel for a bounce that is even in . In flat space, where the background obeys ,
This is the construction of Ref. [31] (where the fluctuation is denoted ), there applied to a background oscillating about a metastable minimum; here the background rolls slowly along a flat direction, and the same identity holds. For the slow-roll background the equation of motion includes Hubble friction, and the cancellation of the linear term holds on the expanding background, with corrections to the flat-space form of order . The frozen-background approximation replaces by its value at nucleation, . In the dimensionless variables of Eq. (19), the fluctuation potential of Eq. (9) is with
which does not depend on the slope . It is stationary at for every , with Hessian eigenvalues and and no mixing, so the far-field point is metastable and the dependence of the action arises entirely from the coupling .
For physical Euclidean radius , introduce . The bounce equations and dimensionless action are [32]
Primes denote derivatives with respect to , and acts on . CosmoTransitions supplies a path-deformation seed [60], which is refined by collocation using analytic derivatives. At large the solution is matched to the linear tails in the two Hessian eigendirections, where is the modified Bessel function of the second kind.
The tail has and is therefore over the whole computational domain. Its contribution to the action converges and is included. We define by , giving . For the reference bounce at , and . This convention applies to the thick-wall profiles obtained here, whose central field values do not reach the lower-energy minimum.
Localization coefficient
The Gaussian dependence of the rate on the inflaton displacement, Eq. (3), follows from Eq. (21) by expanding the action in . At the potential is even in , so the bounce has . Since the action is stationary with respect to the fields, its first derivative is the explicit one,
which vanishes at : the rate is maximal there. Differentiating again and setting , the term containing is multiplied by and drops out, leaving
where is the profile of the reference bounce. The function is the linear response of the inflaton to a change of its value far from the bubble. Differentiating the equation of motion gives, in the range where the mass term can be neglected,
which defines the length : outside the bubble is a harmonic function in four dimensions, and is the coefficient of its tail. The integrand above is the Laplacian of , so the integral reduces to a surface term, . With , , and ,
which is Eq. (10) of the paper. The variation of the prefactor across the window is neglected. Physically, the coupling pulls the inflaton toward inside the bubble, and the cost of nucleating at displacement is the gradient energy of the inflaton profile that interpolates between the bubble and the background. When the coupling is strong, throughout the bubble, the inflaton is held at inside, approaches the radius of that region, and , where is the radius of that region; this equals when coincides with the radius used to normalize . When it is weak, and , proportional to . Solving Eq. (26) on the reference bounce gives , so that and . Equation (27) is a local expansion about ; the computed action exceeds its quadratic form by less than 1%, or 0.03 in , across the interval that contains 99% of the number integral.
Matching and the number spectrum
At approximately constant , the scalar amplitude fixes the rolling speed and relates the dimensional scales to ,
The reference transition is at , where the longitudinal slope is . The slope enters only through this mapping from to , not through the bounce of Eq. (21). The number spectrum follows by combining the nucleation rate with the dilution of bubbles formed earlier [18, 19],
The Jacobian is included in the normalization. As in the paper, we set rather than calculating the fluctuation determinant [33], and impose the abundance condition
Together with the amplitude relation, this fixes and from the computed dimensionless profiles. For the central benchmark, gives and . The Gaussian estimate reproduces to 1% with these values. Comparing the computed width with Eq. (4) gives an effective localization coefficient in the radius convention adopted here. This is about 2% larger than the value obtained from Eq. (27), the difference arising from the terms beyond quadratic order in and the variation of the prefactor across the window, and is close to the value for an inflaton held fixed within radius . The width is then obtained by integrating the spectrum rather than fitting a Gaussian localization coefficient.
Numerical checks and corrections to the frozen background
The rate is sampled across , covering its suppressed tails. Refining the radial grid, extending the radial domain, and widening the sampled interval in change the abundance and width by less than fractionally. An independent one-field shooting calculation reproduces the central action to the same accuracy. The fluctuation check at the central bounce finds one negative radial mode and a translation mode approaching zero under refinement, as expected for the bounce [33]. These checks support numerical convergence of the sampled branch, but do not exclude competing transitions.
Freezing the background neglects three effects, whose sizes we estimate. (i) Velocity. The Euclidean continuation of the slow-roll background is , where overdots on denote derivatives with respect to the rescaled time , so that for the benchmark, where . The velocity term is odd in , and the bounce is even, so its first-order contribution to integrates to zero. The leading correction is second order in , where is the Gaussian width of the rate in , the interval over which rises by unity, and for the benchmark; equivalently . At the inflaton induced by the velocity is , with and at large , and including the response of the bounce gives
For an inflaton held fixed within radius the ratio of integrals is 2. Solving for on the reference bounce gives 2.29, so at the peak of the rate. (ii) Acceleration. For the slow-roll background the acceleration is suppressed relative to the slope, with , and the displacement it produces across the bounce, of order , is negligible. (iii) Preparation of the state. We assume that the quantum state far from the bubble is the adiabatic vacuum of the slow-roll trajectory. In the terminology of [31], slow-roll prepares the state gradually, with no power at the transverse transition frequencies, and the transverse field is heavy, . Stochastic displacements of the inflaton are suppressed relative to its classical motion by per e-fold. These estimates are all well below the uncertainty in the fluctuation prefactor [18, 19, 33].
Applicability of the PBH mass relation
For the local benchmark, the lower basin has vacuum density , with , so its interior expansion rate satisfies . Provided that the interior remains in this basin, the bubble radius at reheating therefore obeys
The bubbles are consequently much larger than their interior de Sitter horizon, placing them in the large supercritical regime for which the exterior black-hole mass follows the scaling adopted in Eq. (5) [18, 19].
Acoustic normalization and localized fluctuations
The reflecting-wall estimate of Ref. [36] gives acoustic energy per PBH, where is the physical photon-diffusion length. The dominant contribution near the thermalization epoch gives
The initial quantities and also describe the population during acoustic dissipation because the adopted history neglects growth before . Using at s [36] gives a comoving diffusion length Mpc. For present density parameters and of dark matter and radiation, [30], and substitution of Eq. (2) yields Eq. (12). Its normalization retains the hydrodynamic uncertainty of the reflecting-wall approximation.
To estimate localized fluctuations, take a comoving Silk volume [36], with PBH count , Poisson mean , and single-object contribution . For comoving distance Gpc to last scattering [30], the sky contains approximately such regions. A characteristic largest occupancy , corresponding to , is set by
For the full-sample reference at and , this gives and , as quoted in the paper. The bright-subset estimate is smaller. This comparison uses the approximate rare-patch threshold discussed in Ref. [36] from Ref. [37], which is distinct from the FIRAS monopole limit [10].
Accretion histories and CMB response
Accretion histories
We motivate the gas supply by the limiting picture of cold, approximately spherical infall on a cosmological timescale [29]. Equating the gravitational dynamical time to the Hubble time selects the radius , at which the characteristic gravitational speed is . Evaluating the spherical flux at this speed and at the mean baryon density gives
where is the instantaneous mass and we have used . The order-unity normalization is an assumption, and the neglect of pressure, angular momentum, feedback, and halo structure means that Eq. (35) should not be read as a strict bound on gas delivery. We assume that all supplied gas reaches the black hole. Accretion begins at , near recombination, when the falling free-electron abundance weakens Compton drag on the infalling gas [43]; earlier accretion and its delayed deposition are omitted.
For radiative efficiency the luminosity is and the mass growth as . We define the Eddington ratio , with and Myr, and the growth factor . We consider two histories. In Hubble-limited accretion all the gas supplied by Eq. (35) is accreted, so that, using ,
The Eddington ratio is independent of the mass and is largest at onset, . For it falls from at to 5.52 at and reaches unity near . In Eddington-limited accretion the luminosity is not allowed to exceed , which replaces the inflow by
so that . In both histories the mass follows from
For the Eddington limit is never reached and the two histories coincide. In both, the inflow is proportional to , so the growth factor is independent of the initial mass: accretion shifts the mass spectrum in without changing , as long as all objects share the same history. For , for Hubble-limited accretion, and and 2.29 for Hubble- and Eddington-limited accretion, respectively.
Escaping radiation
The deposition calculation follows the part of the luminosity that leaves the neighborhood of the black hole as a hard continuum with a cutoff near 200 keV. We write it as , with
where is the fraction of the luminosity that both is emitted in this form and escapes. Radiation that is absorbed near the source and re-emitted below the hydrogen ionization threshold is excluded from , while escaping ultraviolet or soft X-ray radiation would add to the ionization and strengthen the limits. Taking assigns the entire luminosity to hard escaping photons and gives the tightest limit for this spectrum. The effective efficiency controls the CMB signal, whereas alone controls the mass growth and the Eddington limit.
Deposition and calibration
Conservation of comoving number gives the power injected in escaping hard photons per proper volume,
where is the proper dark-matter density. To account for the delay between emission and absorption, define the dimensionless deposited power
For the adopted hard continuum, the approximate Compton-loss treatment of Ref. [43] gives
where is the total hydrogen number density, is the effective Compton energy-loss rate, and is the Thomson cross section. The coefficient 4 follows by dividing the deposited-power evolution equation, whose expansion term is , by . We impose , consistent with omitting radiation emitted before onset.
We calibrate to Ref. [45], which uses Planck 2018, baryon acoustic oscillation, and supernova data, with ExoCLASS to compute how the deposited energy modifies the ionization history and the CMB anisotropy spectra. For PBHs radiating continuously at the Eddington luminosity it obtains the mass-independent bound . For this reference history we set in Eq. (42) and integrate from , so that the solution loses memory of its initial condition before the CMB-sensitive interval.
We project the model and reference histories onto the published Planck-forecast decay-basis modes[9] of Ref. [61], whose density scaling matches Eq. (41). Interpolation onto the native 90-bin grid, covering approximately , gives
where and are the mode vectors and eigenvalues, and is the number of retained principal components. The published vectors are discretely orthonormal, so no additional integration weight is introduced. Since the response is linear in the abundance,
The calculation uses a flat cosmology with , , , , and hydrogen mass fraction 0.755, consistent with the Planck reference cosmology [30]. For Hubble-limited accretion with and , ten response modes give and hence . The first mode contains only 58% of the model’s squared response, but increasing from five to ten modes changes the limit only from to . Eddington-limited accretion with and gives .
The dependence on the efficiencies follows from Eq. (42). At fixed growth history is proportional to , so both limits scale as . For Hubble-limited accretion , so apart from the dependence of on , which changes the limit at the percent level for . These scalings give Eq. (13) of the paper. Limiting the inflow can only reduce the deposited power, so Hubble-limited accretion is the more constrained history at a given . Photon trapping would lower and is covered by the same scaling.
Population requirements
Figure 2 of the paper compares the limits of Eq. (13) with the abundances required by Eq. (2). The required abundance increases linearly with , whereas the limits are independent of for the adopted histories. The published BHL and PR curves in that figure are shown with their fixed-mass conventions identified with the initial parameters, and . The comparison is made at accretion onset, so a black-hole mass observed at redshift is first mapped to
For example, an observed mass of at corresponds to for Hubble-limited and for Eddington-limited accretion with . Requiring in Eq. (2) gives
At the bright subset requires , so that for Hubble-limited accretion and for Eddington-limited accretion with . Equivalently, the bright subset is allowed for
At the same initial mass the integrated population requires , which is allowed for
with the tighter value in each range corresponding to the larger number density. The X-ray fractions of observed active nuclei [52] and the X-ray weakness of the LRDs [53, 54] motivate considering , although their relevance to isolated black holes before reionization is uncertain.
Limitations
These estimates rescale a published Planck-based constraint and are not a new likelihood analysis. They assume spatially averaged deposition, and the mean emitted power does not establish that rare sources produce a sufficiently homogeneous ionization pattern. The forecast modes adopt an ionization and heating prescription that omits the additional Lyman- excitation contribution [61]. The reference constraint uses spectral transfer functions in ExoCLASS [45], whereas both histories here are propagated with the approximate Compton-loss equation, so the calibration retains uncertainties from the response shape and the deposition treatment. A quantitative likelihood bound requires a consistent spectral deposition and ionization calculation with the relevant CMB data.
References
- [1]J. Matthee et al., Astrophys. J. 963, 129 (2024), arXiv:2306.05448 [astro-ph.GA].arxiv.org/abs/2306.05448
- [2]D. D. Kocevski et al., Astrophys. J. 986, 126 (2025), arXiv:2404.03576 [astro-ph.GA].arxiv.org/abs/2404.03576
- [3]L. J. Furtak et al., Nature 628, 57 (2024), arXiv:2308.05735 [astro-ph.GA].
- [4]I. Juodžbalis et al., Nature 653, 1017 (2026), arXiv:2508.21748 [astro-ph.GA].
- [5]S. Zhang, B. Liu, V. Bromm, and F. Kühnel, Astrophys. J. Lett. 1000, L19 (2026), arXiv:2512.14066 [astro-ph.GA].
- [6]V. De Luca, L. Del Grosso, G. Franciolini, K. Kritos, E. Berti, D. J. D’Orazio, and J. Silk, Phys. Rev. Lett. 136, 231402 (2026), arXiv:2512.19666 [astro-ph.CO].
- [7]T. Nakama, B. Carr, and J. Silk, Phys. Rev. D 97, 043525 (2018), arXiv:1710.06945 [astro-ph.CO].
- [8]T. Nakama, T. Suyama, and J. Yokoyama, Phys. Rev. D 94, 103522 (2016), arXiv:1609.02245 [gr-qc].
- [9]S. Balaji, B. Cyr, S. R. Geller, D. I. Kaiser, D. L. Lorenzoni, E. McDonough, and W. Qin, Little Red Dots from Primordial Black Holes: Non-Gaussianity and Accretion Evade μ-Distortion Constraints (2026), arXiv:2611.12861 [astro-ph.CO].
- [10]D. J. Fixsen, E. S. Cheng, J. M. Gales, J. C. Mather, R. A. Shafer, and E. L. Wright, Astrophys. J. 473, 576 (1996), arXiv:astro-ph/9605054.
- [11]F. Bianchini and G. Fabbian, Phys. Rev. D 106, 063527 (2022), arXiv:2206.02762 [astro-ph.CO].
- [12]B. Zhang, W.-X. Feng, and H. An, Little Red Dots from Small-Scale Primordial Black Hole Clustering (2025), arXiv:2507.07171 [astro-ph.CO].
- [13]D. Hooper, A. Ireland, G. Krnjaic, and A. Stebbins, JCAP 2024 (04), 021, arXiv:2308.00756 [astro-ph.CO].
- [14]S. Allegrini, A. J. Iovino, G. Perna, and H. Veermäe, JCAP 2026 (10), 085, arXiv:2606.28296 [astro-ph.CO].
- [15]S. Dave, S.-F. Yan, A. Ilyas, and Y.-F. Cai, Phys. Rev. D 114, 083546 (2026), arXiv:2607.03138 [astro-ph.CO].
- [16]J. Guo, J. Liu, M. Tanaka, X.-P. Wang, and H. Xiao, Supermassive Primordial Black Holes from a Catalyzed Dark Phase Transition for Little Red Dots (2026), arXiv:2604.01304 [hep-ph].
- [17]J. Garriga, A. Vilenkin, and J. Zhang, JCAP 2016 (02), 064, arXiv:1512.01819 [hep-th].
- [18]H. Deng and A. Vilenkin, JCAP 2017 (12), 044, arXiv:1710.02865 [gr-qc].
- [19]M. Kleban and C. E. Norton, Phys. Rev. D 111, 023538 (2025), arXiv:2310.09898 [hep-th].
- [20]H.-L. Huang and Y.-S. Piao, Towards supermassive primordial black holes from inflationary bubbles (2023), arXiv:2312.11982 [astro-ph.CO].
- [21]H.-L. Huang, J.-Q. Jiang, J. He, Y.-T. Wang, and Y.-S. Piao, Sub-Eddington accreting supermassive primordial black holes explain little red dots (2024), arXiv:2410.20663 [astro-ph.GA].
- [22]S. Chon, S. Hirano, T. Ishiyama, S.-J. Chang, and V. Springel, Overmassive black holes and little red dots naturally form in simulations (2026), arXiv:2604.04955.
- [23]P. Natarajan, F. Pacucci, A. Ricarte, A. Bogdan, A. D. Goulding, and N. Cappelluti, Astrophys. J. Lett. 960, L1 (2024), arXiv:2308.02654 [astro-ph.HE].
- [24]K. Freese, T. Rindler-Daller, D. Spolyar, and M. Valluri, Rept. Prog. Phys. 79, 066902 (2016), arXiv:1501.02394 [astro-ph.CO].
- [25]C. Ilie, J. Paulin, A. Petric, and K. Freese, Universe 12, 1 (2026), arXiv:2511.08477 [astro-ph.GA].
- [26]K. Freese, G. M. Fuller, S. Ghodla, C. Ilie, K. S. Kehrer, T. Rindler-Daller, and E. I. Sfakianakis, Early Formation of Supermassive Black Holes via Dark Star Gravitational Instability (2025), arXiv:2511.08578 [astro-ph.CO].
- [27]V. Rusakov, D. Watson, G. P. Nikolapoulos, G. Brammer, R. Gottumukkala, T. Harvey, K. E. Heintz, R. Damgaard, S. A. Sim, A. Sneppen, A. P. Vijayan, N. Adams, D. Austin, C. J. Conselice, C. M. Goolsby, S. Toft, and J. Witstok, Nature (London) 649, 574 (2026), arXiv:2503.16595 [astro-ph.GA].DOI
- [28]H. B. Aksin et al., Astrophys. J. 991, 37 (2025), arXiv:2406.10341 [astro-ph.GA].
- [29]M. Ricotti, Astrophys. J. 662, 53 (2007), arXiv:0706.0864 [astro-ph].DOI
- [30]N. Aghanim et al. (Planck), Astron. Astrophys. 641, A6 (2020), [Erratum: Astron. Astrophys. 652, C4 (2021)], arXiv:1807.06209 [astro-ph.CO].
- [31]O. Janssen, M. Kleban, M. Mirbabayi, and C. Norton, Tunneling from an oscillating state in quantum field theory (2026), arXiv:2611.14022 [hep-th].arxiv.org/abs/2611.14022
- [32]S. R. Coleman, Phys. Rev. D 15, 2929 (1977), [Erratum: Phys. Rev. D 16, 1248 (1977)].
- [33]C. G. Callan, Jr. and S. R. Coleman, Phys. Rev. D 16, 1762 (1977).
- [34]M. Yamada, Tunneling from a Slowly Rolling Multifield Background Using Tadpole Subtraction (2026), arXiv:2608.20451 [hep-ph].arxiv.org/abs/2608.20451
- [35]J. Chluba and R. A. Sunyaev, Mon. Not. Roy. Astron. Soc. 419, 1294 (2012), arXiv:1109.6552 [astro-ph.CO].DOI
- [36]H. Deng, A. Vilenkin, and M. Yamada, JCAP 2018 (07), 059, arXiv:1804.10059 [gr-qc].DOI
- [37]R. Khatri and R. Sunyaev, JCAP 2015 (09), 026, arXiv:1507.05615 [astro-ph.CO].DOI
- [38]N. Afshordi, P. McDonald, and D. N. Spergel, Astrophys. J. Lett. 594, L71 (2003).
- [39]D. Inman and Y. Ali-Haïmoud, Phys. Rev. D 100, 083528 (2019).
- [40]C. Gerlach, Y. Gouttenoire, A. J. Iovino, and N. Leister, Phys. Rev. Lett. 137, 181401 (2026).
- [41]G. Hüts, Clustering of JWST LRDs. Implications for PBHs and other early-forming seeds (2026), arXiv:2610.23806 [astro-ph.CO].
- [42]Z. Wang, F. Jiang, H. Zhang, X. Shen, Z. Jia, L. C. Ho, K. Inayoshi, and L. Jiang, Mon. Not. Roy. Astron. Soc. 552, stag1712 (2026), arXiv:2603.15736 [astro-ph.CO].arxiv.org/abs/2603.15736
- [43]Y. Ali-Haïmoud and M. Kamionkowski, Phys. Rev. D 95, 043534 (2017), arXiv:1612.05644 [astro-ph.CO].
- [44]V. Poulin, P. D. Serpico, F. Calore, S. Clesse, and K. Kohri, Phys. Rev. D 96, 083524 (2017), arXiv:1707.04206 [astro-ph.CO].DOI
- [45]P. D. Serpico, V. Poulin, D. Inman, and K. Kohri, Phys. Rev. Res. 2, 023204 (2020), arXiv:2002.10771 [astro-ph.CO].DOI
- [46]K. Park and M. Ricotti, Astrophys. J. 739, 2 (2011), arXiv:1006.1302 [astro-ph.CO].DOI
- [47]K. Park and M. Ricotti, Astrophys. J. 767, 163 (2013), arXiv:1211.0542 [astro-ph.CO].DOI
- [48]K. Sugimura and M. Ricotti, Mon. Not. Roy. Astron. Soc. 495, 2966 (2020), arXiv:2003.05625 [astro-ph.GA].DOI
- [49]G. Facchinetti, M. Lucca, and S. Clesse, Phys. Rev. D 107, 043537 (2023), arXiv:2212.07969 [astro-ph.CO].DOI
- [50]D. Agius, R. Essig, D. Gaggero, F. Scarcella, G. Suczewski, and M. Valli, JCAP 2024 (07), 003, arXiv:2403.18895 [hep-ph].arxiv.org/abs/2403.18895
- [51]L. Piga, M. Lucca, N. Bellomo, V. Bosch-Ramon, S. Matarrese, A. Raccanelli, and L. Verde, JCAP 2022 (12), 016, arXiv:2210.14934 [astro-ph.CO].DOI
- [52]F. Duras, A. Bongiorno, F. Ricci, E. Piconcelli, F. Shankar, E. Lusso, S. Bianchi, F. Fiore, R. Maiolino, A. Marconi, F. Onori, E. Sani, R. Schneider, C. Vignali, and A. La Franca, Astron. Astrophys. 636, A73 (2020), arXiv:2001.09984 [astro-ph.GA].DOI
- [53]M. Yue, A.-C. Eilers, T. T. Ananna, C. Panagiotou, E. Kara, and T. Mijay, Astrophys. J. Lett. 974, L26 (2024), arXiv:2404.13290 [astro-ph.GA].arxiv.org/abs/2404.13290
- [54]R. Maiolino, G. Risaliti, M. Signorini, B. Trefoloni, I. Juodžbalis, J. Scholtz, H. Übler, F. D’Eugenio, et al., Mon. Not. Roy. Astron. Soc. 538, 1921 (2025), arXiv:2405.00504 [astro-ph.GA].arxiv.org/abs/2405.00504
- [55]K. Park and M. Ricotti, Astrophys. J. 747, 9 (2012), arXiv:1110.0454 [astro-ph.CO].
- [56]K. Freese and D. Spolyar, JCAP 07 (07), 007, arXiv:hep-ph/0412145.
- [57]J. M. Cline, G. D. Moore, and Y. Wang, JCAP 08 (08), 032, arXiv:1106.2188 [hep-th].
- [58]M. Barroso Varela, K. Freese, M. Kleban, C. Norton, and E. F. Sfakianakis, Forthcoming.
- [59]D. J. Fixsen, Astrophys. J. 707, 916 (2009), arXiv:0911.1955 [astro-ph.CO].DOI
- [60]C. L. Wainwright, Comput. Phys. Commun. 183, 2006 (2012), arXiv:1109.4189 [hep-ph].DOI
- [61]D. P. Finkbeiner, S. Galli, T. Lin, and T. R. Slatyer, Phys. Rev. D 85, 043522 (2012), arXiv:1109.6322 [astro-ph.CO].DOI

