A non-perturbative framework for N-point functions of locally non-Gaussian fields

This paper introduces a non-perturbative framework for calculating N-point correlation functions and polyspectra of locally non-Gaussian fields, offering a semi-perturbative alternative to local expansions and demonstrating its efficacy by deriving exact analytic results for fields with exponential tails in the strongly non-Gaussian limit.

The Gist

Imagine the early universe as a giant, invisible ocean. In the standard story of cosmology, this ocean is mostly calm, with tiny, random ripples (Gaussian fields) that eventually grow into the stars, galaxies, and black holes we see today.

However, sometimes the ocean gets turbulent. The ripples don't just stay small; they interact in weird, non-linear ways. This is called non-Gaussianity. When things get really wild, the usual math tools we use to predict what happens (like adding up small waves) break down. It's like trying to predict the weather by only looking at the average temperature; you miss the hurricanes.

This paper, by Hardi Veermäe, introduces a new, more robust toolkit to study these "turbulent" universes without relying on those broken tools.

Here is the breakdown using simple analogies:

1. The Problem: The "Local" Mess

The paper focuses on a specific type of chaos called locally non-Gaussian.

• The Analogy: Imagine you have a smooth, calm sheet of fabric (the Gaussian field). Now, imagine a machine that takes a tiny patch of that fabric and stretches or crumples it based only on how high or low that specific patch is right now. It doesn't care about the neighbors; it just reacts to the local height.
• The Issue: If the machine stretches the fabric too much (strong non-Gaussianity), the relationship between the original smooth sheet and the crumpled result becomes a nightmare to calculate using standard math. The standard method tries to expand the machine's rule into a long list of simple steps (a series expansion), but if the machine is too crazy, that list never ends or gives the wrong answer.

2. The Solution: The "Universal Translator"

The author's big idea is to stop trying to solve the whole messy ocean at once. Instead, he proposes a two-step process:

• Step A: The "One-Point" Dictionary.
  Instead of looking at the whole ocean, look at just one single drop of water. How does the machine transform a single drop? The paper shows that if you know how the machine transforms a single drop (the "1-point distribution"), you can build a Universal Translator (called the $G_n$ function).

  • Metaphor: Think of this like a dictionary. You don't need to know the whole language to translate a sentence; you just need a perfect dictionary that tells you how to translate every single word. Once you have the dictionary, you can translate any sentence, no matter how long or complex.

• Step B: The "Legos" of Correlation.
  The paper uses a mathematical trick (the Kibble–Slepian decomposition) to break the complex interactions between different points in the ocean into simple building blocks.

  • Metaphor: Imagine you are building a castle out of Legos. Standard math tries to glue the whole castle together at once. This new method says, "Let's just look at how two bricks connect, then three, then four." It builds the complex picture by stacking simple connections (correlations) on top of each other.

3. The "Semi-Perturbative" Framework

The author calls this a "semi-perturbative" framework.

• Old Way (Perturbative): "Let's assume the chaos is small and add tiny corrections." (Fails when chaos is huge).
• New Way (Semi-Perturbative): "Let's assume the connections between points are small, but the transformation of the points themselves can be absolutely wild."
  • This allows the math to work even if the machine crumples the fabric into a ball, as long as the fabric isn't stretching infinitely far apart instantly.

4. The Case Study: The "Exponential Tail"

To prove the method works, the author tests it on a specific scenario where the fabric gets stretched so much that it has "exponential tails."

• The Analogy: Imagine a bell curve (the normal distribution of heights). Usually, the tails (very tall or very short people) are rare. In this model, the machine makes extremely tall or short people much more common than usual.
• The Result: The author found that when the chaos gets too strong, the "noise" in the universe actually starts to smooth out the peaks.
  • Visual: Imagine a mountain range. If you shake the ground too hard (strong non-Gaussianity), the sharp peaks get flattened into a plateau. The paper shows that the "power spectrum" (a map of how much energy is in the waves) gets flattened and develops a specific shape (a $k^3$ tail) that is a signature of this extreme chaos.

5. Why Does This Matter?

This isn't just abstract math. It helps us understand:

• Primordial Black Holes: If the early universe had these wild fluctuations, they could collapse into black holes immediately. This math helps us predict how many might exist.
• Gravitational Waves: These fluctuations create ripples in space-time. If our math is wrong, we might miss the signal of these waves in future detectors.

Summary

Think of the universe as a complex recipe.

• Old Math: Tries to taste the whole dish to figure out the ingredients. If the dish is too spicy (non-Gaussian), the taste is confusing.
• This Paper: Says, "Let's just taste the salt (the single point) and the water (the Gaussian field). If we know exactly how the salt changes the water, we can mathematically reconstruct the whole dish, even if it's incredibly spicy."

The author has built a new calculator that works even when the universe is behaving in ways that break all our previous calculators.

Technical Summary

1. Problem Statement

The statistical characterization of primordial curvature perturbations ($\zeta$) is central to understanding early universe cosmology, including the formation of Primordial Black Holes (PBHs) and Scalar-Induced Gravitational Waves (SIGWs). While observations suggest perturbations are nearly Gaussian, non-Gaussianity (NG) can be significant on small scales.

Current analytical methods rely heavily on perturbative expansions of the non-linear function $F$ that maps a Gaussian field $\zeta_G$ to a non-Gaussian field $\zeta$ (i.e., $\zeta(x) = F(\zeta_G(x))$). This approach faces two major limitations:

1. Breakdown of Perturbation Theory: In regimes of strong NG (e.g., near inflection points in inflation or curvaton decay), the series expansion of $F$ may diverge or converge too slowly to be useful.
2. Non-Analyticity: Many physically motivated models involve non-analytic functions (e.g., logarithmic or piecewise functions) where a Taylor series expansion is impossible or ill-defined.
3. Computational Cost: Fully non-perturbative lattice simulations are computationally expensive and difficult to interpret analytically.

The paper aims to develop a non-perturbative framework for calculating $N$-point correlation functions and polyspectra that does not rely on expanding the function $F$, thereby remaining valid even in strongly non-Gaussian regimes.

2. Methodology

The authors propose a semi-perturbative framework based on the Kibble–Slepian decomposition of the multivariate Gaussian probability density function (PDF).

A. Exact Formulation in Position Space

The paper starts with the exact definition of an $N$-point function for a locally non-Gaussian field $\zeta(x) = F(\zeta_G(x))$:
$$\langle \prod_{i=1}^n \zeta(x_i) \rangle = \frac{1}{\sqrt{\det(2\pi \xi_{ij})}} \int \exp\left(-\frac{1}{2}\zeta_i (\xi^{-1})_{ij} \zeta_j\right) \prod_{i=1}^n F(\zeta_i) d\zeta_i$$
where $\xi_{ij} = \langle \zeta_G(x_i)\zeta_G(x_j) \rangle$ is the Gaussian correlation matrix. This reduces the problem to a multi-dimensional Gaussian integral.

B. Kibble–Slepian Decomposition

Instead of expanding $F$, the authors expand the Gaussian correlation matrix using the Kibble–Slepian formula. This decomposes the $N$-dimensional Gaussian PDF into a sum of products of single-point distributions weighted by powers of the correlation function $\xi_{ij}$.
The resulting expansion for the $N$-point function is:
$$\langle \prod_{i} \zeta(x_i) \rangle = \sum_{\nu_{ij} \in \mathcal{N}} \left[ \prod_{i<j} \frac{\xi_{ij}^{\nu_{ij}}}{\nu_{ij}!} \right] \prod_{i} s_i! C_{s_i}$$
where:

• $\nu_{ij}$ are multiplicity matrices representing the number of "lines" connecting points $i$ and $j$.
• $s_i = \sum_j \nu_{ij}$ is the total multiplicity at vertex $i$.
• $C_s$ are the fundamental coefficients, defined as single-point expectation values:
  $$C_s = \frac{1}{s!} \langle \bar{H}_s(\zeta_G) F(\zeta_G) \rangle$$
  Here, $\bar{H}_s$ are rescaled Hermite polynomials. Crucially, $C_s$ depends only on the local mapping $F$ and the variance $\xi_0$, not on the shape of the power spectrum.

C. Diagrammatic Interpretation

The authors map this expansion to a Feynman-like diagrammatic formalism:

• Lines: Represent powers of the Gaussian correlation function $\xi_{ij}$ (or convolution powers of the power spectrum $P_G$ in momentum space).
• Vertices: Represent the resummed coefficients $C_s$.
• Resummation: Unlike standard perturbation theory which expands $F$ in powers of $\zeta_G$ (yielding $F_{NL,n}$), this framework resums all "tadpole" diagrams (self-contractions) into the coefficients $C_s$. This allows the framework to handle non-analytic $F$ because $C_s$ can be computed via numerical integration of the exact $F$, bypassing the need for a Taylor series.

D. Momentum Space Translation

The position-space expansion is converted to momentum space (polyspectra) by replacing $\xi_{ij}^n$ with the $n$-th convolution power of the Gaussian power spectrum, $P_G^{*n}$.

3. Key Contributions

1. Exact Non-Perturbative Formulation: The derivation of a closed-form expression for $N$-point functions (Eq. 2.14 and 3.9) that separates the dependence on the local non-linearity ($C_s$) from the Gaussian background statistics ($\xi_{ij}$).
2. Resummation of Vertices: The introduction of coefficients $C_s$ which effectively resum the infinite series of $F_{NL,n}$ terms. This allows the framework to remain valid even when the expansion of $F$ diverges.
3. Diagrammatic Rules for Locally NG Fields: A complete set of Feynman rules (lines, vertices, momentum conservation) for calculating polyspectra (bispectra, trispectra, etc.) in this semi-perturbative regime.
4. Analytic Results for Strong NG: The application of the framework to a specific model with exponential tails (motivated by USR inflation and curvaton models), deriving exact analytic expressions for the correlation function and power spectrum in the limit of strong non-Gaussianity ($\bar{\beta} \to \infty$).

4. Results

A. The Exponential Tail Model

The authors test the framework on a model where $\zeta = -\frac{1}{2\beta} \ln|1 - 2\beta \zeta_G|$.

• Perturbative Failure: The standard expansion in $F_{NL,n}$ (powers of $\beta$) has a zero radius of convergence and fails rapidly as the non-Gaussianity parameter $\bar{\beta} = \beta\sqrt{\xi_0}$ increases.
• Success of $C_s$ Expansion: The expansion in terms of $C_s$ remains convergent and accurate even for $\bar{\beta} \gtrsim 1$, capturing non-linear effects that standard perturbation theory misses.

B. Strong Non-Gaussianity Limit ($\bar{\beta} \to \infty$)

In the limit of very strong NG, the authors derive exact analytic results:

• The correlation function becomes:
  $$\xi(x) \approx \frac{\xi_0}{8\bar{\beta}^2} \arcsin^2\left(\frac{\xi_G(x)}{\xi_0}\right)$$
• The normalized correlation function becomes independent of $\bar{\beta}$, indicating a saturation of non-Gaussian effects.
• The coefficients $C_n$ scale as $1/\bar{\beta}$, causing the overall amplitude of the power spectrum to be suppressed as NG increases.

C. Impact on Power Spectra

Numerical simulations using the framework reveal distinct features in the non-Gaussian power spectrum $P(k)$:

1. Amplitude Suppression: As NG increases, the peak amplitude of the power spectrum decreases (contrary to naive perturbative expectations where higher orders add positive terms).
2. Flattening: The sharp peaks in the Gaussian spectrum are flattened, forming a plateau.
3. IR Tails: The non-Gaussian power spectrum develops a characteristic $k^3$ infrared (IR) tail at low momenta ($k \ll k_{IR}$). This is a generic feature of locally non-Gaussian fields derived from the causality constraints of the convolution integrals.

5. Significance

• Validity Testing: The framework provides a rigorous tool to test the validity of perturbative approximations in early universe cosmology. It demonstrates that perturbative estimates can fail significantly earlier than the onset of "strong" non-Gaussianity (e.g., when $\xi_0 f_{NL}^2 \sim 10^{-2}$).
• PBH and SIGW Formation: Since PBH formation and SIGW generation are highly sensitive to the tails of the probability distribution and the amplitude of the power spectrum, this non-perturbative approach offers more reliable predictions for these phenomena in regimes where standard perturbation theory breaks down.
• Computational Efficiency: By pre-calculating the coefficients $C_s$ (which depend only on the local model $F$), the framework allows for the rapid generation of polyspectra for any Gaussian power spectrum, avoiding the need for repeated expensive numerical integrations for every new cosmological model.
• Handling Non-Analyticity: It opens the door to studying physically motivated models with non-analytic potentials or phase transitions that were previously inaccessible to analytical treatment.

In summary, Veermäe presents a robust mathematical bridge between local non-Gaussianity and observable polyspectra, offering a semi-perturbative tool that remains accurate in the strong coupling regime where traditional methods fail.