⚛️ general relativity

Evading the CMB $μ$ -distortion bound on Supermassive Primordial Black Hole seeds with Non-Gaussian tails

This paper demonstrates that the CMB $\mu$ -distortion bound, which typically precludes the formation of supermassive primordial black hole seeds via Gaussian perturbations, can be evaded by invoking non-Gaussian probability distribution tails—specifically power-law or sufficiently heavy log-normal shapes—that decouple the small-scale variance from the collapse probability.

Original authors: Sanket Dave, Sheng-Feng Yan, Amara Ilyas, Yi-Fu Cai

Published 2026-07-14✓ Author reviewed ⓘ

📖 6 min read🧠 Deep dive

CC BY 4.0

Original authors: Sanket Dave, Sheng-Feng Yan, Amara Ilyas, Yi-Fu Cai

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine the early universe as a giant, cosmic bakery. For decades, astronomers have been puzzled by a mystery: how did some black holes grow to be so massive—billions of times heavier than our Sun—so quickly? They were already fully grown when the universe was less than a billion years old. If they started as tiny "seeds" left over from dead stars, they simply wouldn't have had enough time to eat their way to that size, even if they ate at the maximum speed allowed by physics.

To solve this, scientists proposed that these black holes started as "supermassive seeds" right from the beginning, born from the raw energy of the Big Bang. But there was a huge problem: a cosmic speed limit.

The Cosmic Speed Limit (The Gaussian Barrier)
Think of the early universe's energy as a smooth, rolling hill. If the bumps on this hill are perfectly random and follow a standard "bell curve" (what scientists call Gaussian statistics), there's a strict rule: you can't make the hill too bumpy without breaking the rules of the Cosmic Microwave Background (CMB), the afterglow of the Big Bang.

Specifically, a telescope called COBE/FIRAS measured the "temperature" of this afterglow and found a tiny distortion, labeled $\mu < 9 \times 10^{-5}$ . This measurement acts like a ruler. If the hills (curvature perturbations) were bumpy enough to create supermassive black hole seeds (masses between $10^5$ and $10^7$ times the mass of our Sun, $M_\odot$ ), they would have created a distortion way too big to be hidden.

Under the old "bell curve" rules, the math says the probability of a bump big enough to make a supermassive seed is so tiny—like $10^{-1600}$ —that it's effectively zero. The authors call this the "Gaussian barrier." It's like trying to win the lottery by buying a ticket where the odds are written in a number so small you'd need more atoms in the universe than there are tickets to have a chance.

The Loophole: Twisting the Rules
But what if the hills aren't smooth and bell-shaped? What if the universe's energy distribution has a weird, "heavy tail"?

The authors of this paper suggest that if the distribution of energy bumps isn't a standard bell curve, the rules change. Imagine a bell curve as a smooth slide that drops off quickly. A "heavy tail" is like a slide that suddenly flattens out into a long, gentle ramp. Even if the average bumpiness of the hill stays the same (satisfying the COBE/FIRAS ruler), the extreme bumps at the very end of the ramp can be much more common than the bell curve predicts.

The team tested four different shapes for these "ramps" to see if they could sneak past the cosmic speed limit:

The Stretched-Exponential: A ramp that drops off slower than a normal slide but still eventually falls.
The Power-Law: A ramp that stays high for a long time, like a cliff that doesn't drop off sharply.
The Log-Normal: A ramp that gets very heavy very quickly, like a slide that turns into a wall of probability.
The Generalized Normal: A symmetric version of the above.

The Results: What Works and What Fails
The team ran the numbers, keeping the "average bumpiness" fixed at the strict limit allowed by the COBE/FIRAS data. Here is what they found:

The Standard "Non-Attractor" Models Fail: Many popular theories about the early universe predict a specific kind of bumpiness that creates an "ordinary exponential" tail (like a standard slide). The paper shows that even these models are too light. They still drop off too fast. If the universe followed these standard rules, the supermassive seeds would still be impossible to form. The "Gaussian barrier" holds firm against these specific ideas.
The "Heavy" Tails Win: The window for supermassive seeds only reopens if the tail is "heavy" enough.
- Power-Law Tails: These come from models where the energy potential has a "fractional" shape (a specific, weird curve). These tails are heavy enough to produce the necessary number of seeds while respecting the distortion limit.
- Log-Normal Tails: These are the heaviest of all, arising from "multiplicative" dynamics (where effects stack up like a chain reaction). These also work, producing plenty of seeds.

How Sure Are We?
The authors are very careful with their confidence. They didn't just guess; they used a mathematical tool called the " $\delta N$ formalism" to map how different inflationary dynamics (the physics of the early universe's expansion) translate into these probability shapes.

What is proven: They proved mathematically that if the universe follows standard "bell curve" rules or standard "exponential tail" rules, the supermassive seed window is closed.
What is suggested: They suggest that if the early universe had "fractional-potential" dynamics or "multiplicative" dynamics, the window opens up. However, they admit that for the log-normal case, they are treating it as a "phenomenological proxy"—a useful placeholder for a complex process they haven't fully derived from first principles yet.
The Verdict: The paper does not claim to have found the definitive answer. Instead, it says: "If you want supermassive black hole seeds to exist, the early universe must have had these specific, heavy-tailed shapes. The boring, standard shapes are ruled out."

The Bottom Line
The mystery of the early supermassive black holes isn't solved yet, but the authors have cleared the path. They've shown that the "boring" ways the universe could have behaved are dead ends. To find the seeds, we need to look for a universe that was a little bit more chaotic, with probability distributions that have "heavy tails" (specifically power-law or log-normal shapes). If the universe was built that way, the seeds could have formed without breaking the cosmic speed limit, and the giants we see today could have grown from them.

The paper concludes that while the "Gaussian barrier" is a formidable wall, it can be evaded—not by breaking the rules, but by realizing the rules of probability are more flexible than we thought, provided the universe's early dynamics were just right.

Technical Summary: Evading the CMB $\mu$ -distortion bound on Supermassive Primordial Black Hole seeds with Non-Gaussian tails

Problem Statement
The existence of luminous quasars at redshifts $z \gtrsim 6$ , hosting supermassive black holes (SMBHs) with masses $\sim 10^8 - 10^{10} M_\odot$ , presents a significant timing problem for standard galaxy formation models. Growing such objects from stellar-mass remnants ( $\sim 10^2 M_\odot$ ) via Eddington-limited accretion requires $\sim 0.9$ Gyr, which is comparable to the cosmic age at $z \sim 6-7$ . This necessitates "heavy seeds" in the mass range $10^5 - 10^7 M_\odot$ formed early in the universe, potentially as Primordial Black Holes (PBHs).

However, generating a population of PBHs in this mass range is constrained by Cosmic Microwave Background (CMB) spectral distortions. Photon diffusion during the $\mu$ -era ( $5 \times 10^4 \lesssim z \lesssim 2 \times 10^6$ ) damps small-scale acoustic waves, injecting energy that creates a chemical potential distortion $\mu$ . The COBE/FIRAS limit ( $\mu < 9 \times 10^{-5}$ ) constrains the variance of curvature perturbations ( $\sigma_\zeta^2$ ) on the relevant scales ( $k \sim 10^2 - 10^4 \text{ Mpc}^{-1}$ ) to $\sigma_\zeta^2 \lesssim 10^{-4}$ .

Under the assumption of Gaussian statistics, this variance constraint fixes the tail of the probability distribution function (PDF). Consequently, the probability of fluctuations exceeding the collapse threshold ( $\zeta_c \approx 0.67$ ) becomes exponentially suppressed ( $\beta \sim 10^{-1600}$ for $M \sim 10^6 M_\odot$ ), rendering the PBH abundance negligible for seeding. The authors term this obstruction the "Gaussian barrier."

Methodology
The paper proposes that the Gaussian barrier can be evaded if the variance (controlled by the $\mu$ -distortion) is decoupled from the deep tail probability (controlling PBH collapse). This requires non-Gaussian statistics where the variance does not uniquely determine the tail behavior.

Formalism: The authors utilize the non-perturbative $\delta N$ formalism. They argue that perturbative expansions in $f_{NL}$ and higher cumulants are inadequate because PBH collapse occurs at $\sim 50-100$ standard deviations from the mean, far outside the perturbative regime. Instead, they relate the asymptotic PDF tails directly to the global shape of the $\delta N$ map.
Standardization: To isolate the effect of tail shape, the authors standardize all candidate distributions to unit variance ( $\sigma_\zeta = 1$ in standardized units) and zero mean. The physical variance is then capped by the FIRAS limit: $\sigma_{\zeta, \text{max}}^2(M) = \mu_{\text{lim}} / (2.2 W_\mu(M))$ .
Tail Families: Four Gaussian-cored families of PDFs are analyzed, representing different asymptotic behaviors of the $\delta N$ $δ N$ map:
- Family A (Generalized Normal): Symmetric deformation of the Gaussian.
- Family B (Stretched Exponential): Represents single-field non-attractor dynamics (e.g., ultra-slow-roll, upward steps).
- Family C (Power Law): Represents algebraic tails arising from fractional-potential dynamics (e.g., radiatively corrected inflection points, non-canonical kinetic sectors).
- Family D (Log-Normal): Represents multiplicative dynamics (e.g., multifield scenarios or volume-weighted expansion), treated as a phenomenological proxy.
Calculation: For each family, the authors compute the maximum PBH abundance fraction $\beta_{\text{max}}(M)$ allowed by the distortion bound. This involves integrating the standardized PDF from the standardized collapse threshold $u_c = \zeta_c / \sigma_{\zeta, \text{max}}$ to infinity.

Key Results

The Gaussian Barrier: For Gaussian distributions, even saturating the FIRAS bound yields $\beta \ll 10^{-15}$ , insufficient for seeding.
Exponential Tails (Family B): Standard single-field non-attractor dynamics produce exponential tails ( $P \propto e^{-\lambda \zeta}$ , corresponding to $p=1$ in stretched exponentials). While heavier than Gaussian, these tails are still too light; they yield $\beta \sim 10^{-30}$ at $M=10^6 M_\odot$ , failing to reopen the seed window.
Sub-Exponential Tails (Families C & D):
- Power-law tails (Family C), arising from fractional potentials ( $V \propto |\phi - \bar{\phi}|^m$ with $2 < m < 2.5$ ), possess an asymptotic log-slope $D_\infty = 0$ . These tails lack exponential suppression at large thresholds. For indices $q \sim 4$ (where $P \propto \zeta^{-(q+1)}$ ), the abundance reaches $\beta \sim 10^{-15}$ across the seed window, satisfying seeding requirements.
- Log-normal tails (Family D), representing multiplicative dynamics, are the heaviest considered. For shape parameters $s \gtrsim 0.5$ , they easily exceed the reference abundance level ( $\beta \sim 10^{-15}$ ).
Viable Parameter Space: The viable regions for seeding are found in the parameter space of heavy tails ( $D_\infty = 0$ ). Specifically, the stretched-exponential family requires $p \lesssim 0.6$ , and the power-law family requires $q \lesssim 4$ .

Significance and Claims
The paper claims that the "Gaussian barrier" is not a fundamental prohibition on PBH seeds but a consequence of assuming Gaussian statistics. By decoupling the variance (constrained by CMB distortions) from the tail shape (controlled by non-perturbative inflationary dynamics), the seed window for $10^5 - 10^7 M_\odot$ PBHs can be reopened.

Dynamical Implications: The results imply that generic single-field non-attractor transitions (which produce exponential tails) are insufficient for SMBH seeding. Instead, viable models must involve specific dynamics: either fractional-potential dynamics (leading to power-law tails) or multiplicative/multifield dynamics (leading to log-normal tails).
Methodological Shift: The authors emphasize that constraints on PBH formation and inflation should be derived from the full non-perturbative one-point PDF rather than relying solely on the power spectrum or perturbative non-Gaussianity parameters ( $f_{NL}$ ), which fail to capture the behavior of rare, large-amplitude fluctuations.
Modesty: The authors note that the log-normal case is treated as a phenomenological proxy for multiplicative dynamics rather than a derived prediction from a specific single-field model. They also acknowledge that the narrow-feature approximation for the distortion bound and the Press-Schechter collapse criterion are simplifications, though they argue these do not alter the relative ordering of the tail families.

In conclusion, the paper demonstrates that a primordial origin for high-redshift SMBHs remains viable provided the underlying inflationary dynamics generates sufficiently heavy, sub-exponential tails in the curvature perturbation PDF, thereby evading the CMB $\mu$ -distortion bound that rules out Gaussian seeds.

Technical Summary: Evading the CMB μ\muμ-distortion bound on Supermassive Primordial Black Hole seeds with Non-Gaussian tails

More like this

Technical Summary: Evading the CMB $\mu$ -distortion bound on Supermassive Primordial Black Hole seeds with Non-Gaussian tails