⚛️ high-energy theory

Two-loop matter power spectrum from FFTLog and kite integrals

This paper presents a fast and accurate method for computing the two-loop matter power spectrum in the Effective Field Theory of Large-Scale Structure by decomposing the linear power spectrum via FFTLog into massless kite integrals, which are evaluated with high precision using a Mellin-Barnes representation to enable millisecond-level cosmological predictions.

Original authors: Paola Cappellazzi, Johannes M. Henn, Marko Simonović

Published 2026-11-02

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Paola Cappellazzi, Johannes M. Henn, Marko Simonović

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 or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

To understand the universe on its grandest scales, cosmologists do not simply look at where galaxies are; they measure how matter clumps together. This clumping is not random. In the early universe, tiny fluctuations in density grew under the pull of gravity, eventually forming the vast cosmic web of filaments and voids we see today. By measuring the statistical pattern of this clumping across different distances, scientists can reconstruct the history of the universe and test theories about dark matter and dark energy. However, as gravity pulls matter together over billions of years, the simple, smooth patterns of the early universe become distorted. Gravity couples different scales of motion, creating a complex, non-linear tangle that is difficult to predict with standard equations. To make accurate predictions for modern surveys, which measure the universe with extreme precision, scientists must calculate these gravitational interactions with increasing levels of detail, pushing their mathematical models to the very edge of what is computationally possible.

For decades, the standard tool for these calculations has been a method called perturbation theory, which treats the complex gravitational interactions as a series of corrections to a simple starting point. The first layer of correction, known as the one-loop calculation, has become a standard part of cosmological analysis. But as surveys like DESI and Euclid begin to map the universe with percent-level precision, the one-loop approximation is no longer sufficient. Scientists need the next level of detail: the two-loop correction. This step involves calculating how three different waves of matter density interact simultaneously. The mathematics required for this is notoriously difficult. It involves integrating over every possible direction and speed of invisible, internal motions within the cosmic web, a task that traditionally requires massive computing power and must be repeated for every new theory of the universe.

In a new study, researchers have developed a method to perform these difficult two-loop calculations with unprecedented speed and accuracy. Instead of trying to solve the complex integrals from scratch for every new scenario, the team realized they could break the problem into two distinct parts: the specific details of the universe being studied, and the universal geometry of the gravitational interactions themselves. They used a mathematical technique called FFTLog to decompose the complex, wavy pattern of matter density into a sum of simple, smooth curves. This allowed them to separate the messy, real-world data from the underlying physics. Once separated, the complex gravitational interactions reduced to a specific, repeating shape known as a kite integral. This shape, which looks like a geometric kite formed by five connecting lines, represents the fundamental way gravity couples different scales of motion.

The challenge was that these kite integrals had to be calculated for a wide range of complex numbers, including those with very large imaginary parts. In the world of mathematics, these imaginary parts cause the functions to oscillate wildly, like a signal that vibrates so fast it becomes impossible to track with standard tools. Traditional computer methods often fail here, either missing the peak of the vibration or getting lost in the noise. The researchers overcame this by using a specialized mathematical representation that allowed them to sample the function on a uniform grid. This approach acted like a high-speed camera, capturing the rapid oscillations with perfect clarity. They found that even when the vibrations were extremely fast, the function still decayed in a predictable way, allowing them to calculate the value with ten digits of precision.

The result of this work is a set of precomputed matrices that act as a universal library for these gravitational interactions. Once these matrices are calculated, which takes a few hours on a powerful computer, they can be used to evaluate the two-loop power spectrum for any new cosmological model in just a few milliseconds. This is a massive leap forward. Previously, calculating the same result for a single new model might have taken minutes or even hours of computing time. The team tested their method by comparing it against direct, brute-force numerical integration, which is the gold standard but is too slow for practical use in analyzing large datasets. They found that their fast method agreed with the slow, direct method to within one part in ten thousand, a level of accuracy that is more than sufficient for current and future surveys.

This breakthrough does more than just speed up calculations; it clarifies the relationship between the mathematical shortcuts and the physical reality. The researchers showed that the small differences between their fast method and the direct physical calculation are not errors, but rather predictable artifacts that correspond to known physical effects called counterterms. These counterterms account for the influence of very small-scale physics that the large-scale model cannot see directly. By explicitly calculating and removing these effects, the team demonstrated that their method captures the true physics of the universe with high fidelity. They also confirmed that their results match previous independent calculations, providing a robust cross-check.

The implications for cosmology are significant. With this tool, scientists can now explore a much wider range of theories about the universe without being bottlenecked by computing time. They can test how different types of dark matter or modifications to gravity would affect the clustering of galaxies, running thousands of scenarios in the time it used to take to run a single one. The method is also flexible enough to be extended to other complex measurements, such as how galaxies move relative to the expansion of the universe or how their shapes are distorted by gravity. By turning a decades-old computational bottleneck into a routine calculation, this work provides the necessary foundation for the next generation of cosmic discovery, ensuring that the theoretical predictions keep pace with the flood of data pouring in from the world's most powerful telescopes.

Technical Summary: Two-loop matter power spectrum from FFTLog and kite integrals

Problem Statement
The Effective Field Theory of Large-Scale Structure (EFTofLSS) provides a framework for modeling the clustering of matter on scales where gravity couples different modes, extending standard perturbation theory with counterterms for short-scale physics. While one-loop analyses are standard and supported by fast computational codes, two-loop corrections are necessary to extend the theory to smaller scales and quantify theoretical errors. A primary computational bottleneck arises because each loop introduces a three-dimensional integral over unobserved wavenumbers, where the integrand depends on the linear power spectrum $P_{\text{lin}}(k)$ . For realistic cosmologies, $P_{\text{lin}}(k)$ is known only numerically, requiring these integrals to be repeated at every point in parameter space. Direct numerical integration is computationally expensive (seconds to minutes per wavenumber), while existing fast methods often rely on bases fitted to specific cosmologies or struggle with the complexity of two-loop topologies.

Methodology
The authors propose a method combining the FFTLog decomposition with the evaluation of specific massless Feynman integrals known as "kite integrals."

FFTLog Decomposition: The linear power spectrum is decomposed into a sum of complex power laws, $P_{\text{lin}}(k) \approx \sum c_m k^{\nu + i\eta_m}$ . This transforms the loop integrals into a sum of cosmology-independent integrals where each power law acts as a massless propagator raised to a complex power.
Reduction to Kite Integrals: At the two-loop level, the corrections reduce to massless Feynman integrals of "kite" topology (a two-loop self-energy integral with five propagators). The complexity of the cosmology is entirely contained in the coefficients $c_m$ , while the integrals themselves depend only on the complex powers of the propagators.
Mellin–Barnes (MB) Representation: The central numerical challenge is evaluating these kite integrals for complex powers with large imaginary parts (up to $\sim 20$ $\sim 20$ ), where integrands become highly oscillatory. The authors utilize a two-fold Mellin–Barnes representation. Instead of using adaptive integrators, which may fail to capture the dominant oscillatory regions, they evaluate the MB integrals by summing the integrand over a uniform lattice of points on straight contours.
- This approach relies on the property that for analytic integrands decaying exponentially, the lattice sum converges exponentially fast.
- Large imaginary parts of the powers shift the region of large integrand values but do not slow the decay, allowing the lattice to capture the region automatically without prior knowledge of its location.
- Analytic continuation is used to handle powers where straight contours do not exist, involving residue calculations.
Precomputation and Regularization: The authors precompute tensors containing the kite integral values for a grid of frequencies capable of resolving Baryon Acoustic Oscillations (BAO). They address the discrepancy between the FFTLog calculation (which uses analytic regularization, setting power-law divergences to zero) and physical loop integrals. They explicitly compute the difference terms, which take the form of EFTofLSS counterterms (e.g., $k^2 P_{\text{lin}}$ , $k^4 P_{\text{lin}}$ ), derived from the moments of the spectrum outside the FFTLog interval and the large-momentum limits of the integrands.

Key Contributions

Algorithmic Implementation: A robust numerical method for evaluating two-loop kite integrals with complex powers using a uniform lattice summation of Mellin–Barnes representations, achieving $\sim 10$ -digit precision even for large imaginary parts.
Precomputed Matrices: The generation of cosmology-independent tensors that allow the two-loop matter power spectrum to be evaluated for a new cosmology in a few milliseconds (after a fast Fourier transform of the input spectrum).
Analytic Regularization Framework: A rigorous treatment of the difference between analytically regularized FFTLog integrals and physical loop integrals. The authors derive and compute the specific counterterm-like terms required to match the FFTLog results to direct numerical integration without introducing free parameters.
Validation: The code and precomputed matrices are made publicly available. The method is validated against direct numerical integration and existing hypergeometric representations.

Results

Accuracy: When comparing the FFTLog-based two-loop spectrum (after accounting for counterterm contributions and FFTLog decomposition errors) with direct numerical integration, the authors find agreement at the $10^{-4}$ level relative to the power spectrum up to $k = 0.2 \, h \, \text{Mpc}^{-1}$ .
Discrepancy Analysis: The remaining differences at larger $k$ are attributed to the reconstruction error of the input spectrum by the power-law decomposition (specifically ringing artifacts and BAO resolution limits), rather than errors in the kite integral evaluation.
Bias Independence: The method yields consistent results across different choices of the FFTLog bias parameter ( $\nu$ ), provided the appropriate analytic regularization terms are included.
Computational Cost: The precomputation of the tensors for a grid of $N=168$ points required approximately 400 core-hours. Once computed, evaluating the spectrum for a new cosmology takes roughly 6 ms for 100 wavenumbers.

Significance and Claims
The paper claims to provide a fast and accurate framework for two-loop calculations in EFTofLSS that extends the capabilities of existing FFTLog-based codes (which are typically limited to one loop) without sacrificing precision. The authors emphasize that their approach separates the cosmological input from the loop integration, allowing the same precomputed matrices to be applied to any linear spectrum resolvable by the grid.

The authors modestly note that while their method is highly accurate for the matter power spectrum, it relies on the accuracy of the FFTLog decomposition of the input spectrum. They state that their results can be directly incorporated into existing codes to extend them to two loops. Furthermore, they suggest the framework is extensible to biased tracers, redshift-space distortions, and potentially higher-order calculations (such as the two-loop bispectrum or three-loop spectrum), as the separation of the loop integration from the cosmological input provides a reusable basis for perturbative calculations. The paper does not claim to solve the problem of infrared resummation or time-dependent effects directly, noting these can be handled by modifying the input spectrum or coefficients within the same tensor framework.

Technical Summary: Two-loop matter power spectrum from FFTLog and kite integrals

More like this