🔭 astrophysics

Efficient evaluation of the dark-matter two-loop power spectrum in the EFT of LSS

This paper presents a fast and efficient method for evaluating the two-loop dark-matter power spectrum in the Effective Field Theory of Large-Scale Structure by decomposing cosmological differences into a fixed basis of functions, precomputing cosmology-independent integrals with local UV and IR regularization, and publicly releasing these results to enable rapid analysis of upcoming LSS survey data.

Original authors: Charalampos Anastasiou, Andrea Favorito, Matthew Lewandowski, Leonardo Senatore, Henry Zheng

Published 2026-02-26

📖 6 min read🧠 Deep dive

CC BY 4.0

Original authors: Charalampos Anastasiou, Andrea Favorito, Matthew Lewandowski, Leonardo Senatore, Henry Zheng

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

The Big Picture: Mapping the Cosmic Web

Imagine the universe as a giant, invisible spiderweb made of dark matter. Over billions of years, gravity has pulled this web into clumps (galaxies) and stretched it into vast empty spaces. Astronomers are currently building massive telescopes (like DESI and Euclid) to take a 3D picture of this web.

To understand what they see, they need a perfect map. But the web is messy. In the early universe, it was smooth and easy to predict. Today, it's chaotic and tangled. The paper presents a new, super-fast way to calculate the "messiness" of this web so scientists can compare their maps to the real universe and figure out what the universe is made of.

The Problem: The "Too Hard" Math

Scientists use a method called Perturbation Theory to predict how the web changes. Think of this like baking a cake:

Level 1 (Linear): You mix the flour and sugar. Easy. This predicts the smooth, early universe.
Level 2 (One-Loop): You add eggs and butter. The mixing gets a bit messy, but you can still do it.
Level 3 (Two-Loop): Now you are kneading dough with your hands while the oven is hot. It gets incredibly complicated.

The paper focuses on Level 3 (Two-Loop). This level is necessary to match the extreme precision of modern telescopes. However, calculating this is like trying to solve a million-piece puzzle while blindfolded. The math involves "loop integrals"—essentially summing up the effects of every possible interaction in the universe. Doing this for every single possible version of the universe (changing the amount of dark matter, dark energy, etc.) takes too long. It would take a supercomputer years to just check one set of parameters.

The Solution: The "Reference Cake" Strategy

The authors realized they didn't need to bake a new cake from scratch for every variation. Instead, they baked one perfect "Reference Cake" (based on the standard Planck cosmology model) and then calculated how much the recipe changes for other versions.

Here is how they made it fast and accurate:

1. The "Reference" and the "Difference"

Imagine you have a standard chocolate cake recipe. You want to know what a "strawberry" cake or a "lemon" cake would look like.

Old Way: Bake a strawberry cake from scratch. Bake a lemon cake from scratch. (Takes forever).
New Way: Bake the perfect chocolate cake once. Then, just calculate the difference between the chocolate and the strawberry (e.g., "add 20% more sugar, swap cocoa for strawberry extract").
In the paper: They calculate the complex "Two-Loop" math for the standard universe once. For any other universe, they only calculate the tiny difference between that universe and the standard one. Since the difference is small, the math is much easier.

2. Cleaning Up the "Noise" (UV Subtractions)

When you try to calculate the messiness of the web, you run into a problem: the math blows up when you look at tiny, tiny scales (like trying to count individual atoms in a cloud). This is called an Ultraviolet (UV) Singularity. It's like a microphone picking up a screeching feedback loop that drowns out the music.

The authors developed a way to "mute the feedback." They identified the parts of the math that cause the screeching noise and subtracted them out before doing the heavy calculation.

Analogy: Imagine listening to a song with a loud hiss. Instead of trying to record the whole song and then editing out the hiss later, they put a filter on the microphone that stops the hiss from being recorded in the first place. This leaves a clean signal that is easy to analyze.

3. Fixing the "Silence" (IR Cancellations)

On the flip side, looking at the largest scales (the biggest clumps) also causes math problems, called Infrared (IR) Singularities. It's like trying to hear a whisper in a library; the background noise cancels out the signal.

Analogy: They rearranged the math so that the "whispers" from different parts of the calculation cancel each other out locally (right where they happen), leaving a smooth, quiet result that is easy to measure.

4. The "Universal Library" (Pre-computation)

This is the magic trick. Because they cleaned up the noise (UV) and the silence (IR), and because they are only looking at the difference from a standard model, they realized they could break the calculation down into a set of universal building blocks.

Analogy: Imagine you are a chef who needs to make 10,000 different soups. Instead of chopping vegetables for every single soup, you pre-chop all the possible vegetables (carrots, onions, celery) and store them in labeled jars.
The Paper's Method: They pre-computed all the "jars" (the complex integrals) for the standard universe. These jars are cosmology-independent. They don't care if you are making a "dark matter soup" or a "dark energy soup."
The Result: When a scientist wants to test a new theory of the universe, they don't run the supercomputer for days. They just grab the pre-made jars, mix them together in the right proportions (based on the new theory's parameters), and get the answer in seconds.

Why This Matters

Speed: What used to take days or weeks of supercomputer time now takes seconds.
Precision: They found that to get the math right, they needed 8 specific "counterterms" (adjustment knobs). Previous methods only used 4. This means their map is much more accurate and less likely to be fooled by the messy physics of the small scales.
Future Proof: This method allows astronomers to instantly test thousands of different universe models against real data from telescopes like DESI and Euclid.

Summary

The authors built a universal, pre-computed library of cosmic "ingredients." By cleaning up the mathematical noise and focusing on the small differences between universe models, they turned a task that was previously impossible to do quickly into a simple "mix-and-match" operation. This allows cosmologists to finally decode the secrets of the universe's structure with the speed and precision required by the next generation of telescopes.

1. Problem Statement

The Effective Field Theory of Large-Scale Structure (EFTofLSS) is the standard framework for modeling cosmological density fluctuations in the weakly non-linear regime. While one-loop calculations are routine and used in data analysis (e.g., by DESI), moving to two-loop order is essential for extracting cosmological parameters from next-generation surveys (like Euclid and DESI) with the required precision.

However, computing the two-loop power spectrum faces two major bottlenecks:

Computational Cost: Evaluating the two-loop integrals for every point in the vast cosmological parameter space (during MCMC sampling) is prohibitively slow if done via direct numerical integration for each model.
Numerical Instability: The loop integrands contain severe Ultraviolet (UV) divergences (sensitive to unknown short-distance physics) and Infrared (IR) singularities (cancellations between diagrams that are not manifest locally). These singularities make stochastic numerical integration unstable and inaccurate.

2. Methodology

The authors propose a novel framework that decouples the cosmological model dependence from the numerical integration, enabling fast and precise evaluation. The methodology consists of three main pillars:

A. UV and IR Regularization

To make the integrands suitable for numerical integration, the authors perform systematic subtractions:

UV Regularization: They identify the leading UV singularities in the loop integrands (diagrams $P_{15}, P_{42}, P_{33}$ ) by expanding the integrands in the limit of large loop momenta. They subtract these singular parts analytically. The subtracted terms are shown to be degenerate with the EFT counterterms, allowing them to be absorbed into the renormalization of the theory. This leaves a "UV-regular" remainder that is finite and smooth.
IR Cancellations: While the sum of all diagrams is IR finite, individual diagrams have singularities at soft momenta ( $\vec{q} \to 0$ ). The authors use partitions of unity to decompose the integrands into "channels." By shifting integration variables, they map all IR singularities to the same location (the origin), where they cancel locally within the integrand. This ensures the integrand is smooth everywhere, allowing for stable Monte Carlo integration.

B. Perturbative Expansion Around a Reference Cosmology

To avoid re-computing integrals for every cosmological model, the authors treat the linear power spectrum $P_{\text{lin}}(k)$ of a generic model $j$ as a small deviation from a reference model (the Planck best-fit cosmology, $j=0$ ):
$\Delta P^{[j]}_{\text{lin}}(k) = P^{[j]}_{\text{lin}}(k) - N^{[j]} P^{[0]}_{\text{lin}}(k)$
where $N^{[j]}$ is a rescaling factor.
They expand the loop corrections in powers of this small difference $\Delta P_{\text{lin}}$ . Since $\Delta P_{\text{lin}}$ is small for realistic cosmologies, the series converges rapidly. The two-loop correction is expanded up to cubic order in $\Delta P_{\text{lin}}$ , but the cubic term is shown to be negligible.

C. Cosmology-Independent Basis and Pre-computation

The core innovation is the factorization of the calculation:

Basis Functions: The difference $\Delta P_{\text{lin}}$ is fitted to a set of cosmology-independent basis functions (resembling massive scalar propagators).
Tensor Decomposition: The loop integrals are decomposed into a sum of cosmology-independent tensor integrals (denoted as $T$ ) multiplied by cosmology-dependent coefficients (the fit parameters of the basis functions).
Pre-computation: The authors numerically integrate these universal tensors $T$ once (using the reference cosmology). These pre-computed integrals are stored.
Fast Evaluation: For any new cosmological model, the prediction is obtained by simply contracting the pre-computed tensors with the new model's fit coefficients. This reduces the evaluation time from hours to milliseconds.

3. Key Contributions

Full Two-Loop Renormalization: The paper provides the first complete analytical and numerical renormalization of the two-loop dark-matter power spectrum in the EFTofLSS.
- They identify eight necessary EFT counterterms (including effects from generated vorticity and non-locality in time) required to fully cancel UV divergences. Previous works often used fewer counterterms based on numerical fits, but this work proves analytically that eight are required for full consistency.
Efficient Numerical Algorithm: They developed a method to evaluate two-loop integrals with high precision (sub-percent level) using stochastic integration, overcoming the UV/IR singularities that previously hindered such calculations.
Public Release of Integrals: The authors publicly release the pre-computed cosmology-independent integrals (tensors), enabling the community to perform two-loop analyses without re-doing the heavy numerical lifting.
Validation: They validated the method by comparing the "fit-based" prediction against direct numerical integration of a test cosmology (with significant deviations from the reference). The theoretical error was found to be well below the expected experimental uncertainty of surveys like DESI.

4. Results

Precision: The method achieves a theoretical error of $< 0.6\%$ (at redshift $z=1.23$ ) compared to the experimental uncertainty of DESI-like surveys.
Counterterm Necessity: The Fisher matrix analysis suggests that while not all 8 counterterms are tightly constrained by current data (some are degenerate), including all of them is theoretically necessary for a consistent UV-insensitive prediction. The analysis indicates that a subset (e.g., $c_{\delta,1}, c_{\delta^2,1}, c_{\partial^2\delta,1}$ ) captures the dominant constraints, but the full set is required for robustness.
Convergence: The perturbative expansion around the reference cosmology converges rapidly. The leading term (reference cosmology) accounts for $\sim 60\%$ of the correction, with the first-order correction in $\Delta P_{\text{lin}}$ dominating the rest. The second-order correction is small but necessary for sub-percent precision.

5. Significance

This work represents a major step forward in cosmological data analysis:

Enabling Full-Shape Analysis at Two Loops: It removes the computational bottleneck that prevented the use of two-loop EFT predictions in global fits of cosmological parameters.
Theoretical Rigor: By explicitly deriving the need for 8 counterterms and handling non-local-in-time effects, it sets a new standard for the theoretical consistency of EFTofLSS calculations.
Future-Proofing: The method is generalizable to other tracers (e.g., galaxies in redshift space, Ly- $\alpha$ forest) and higher-order statistics (bispectrum), paving the way for precision cosmology with upcoming missions like Euclid and the Vera C. Rubin Observatory.

In summary, Anastasiou et al. have transformed the two-loop EFTofLSS from a theoretically complex, computationally intractable problem into a fast, robust, and publicly available tool for analyzing the next generation of cosmological surveys.