Sym-EFT: Accelerating Effective Field Theory of Large Scale Structure with Symbolic Regression

arXiv:2511.05093 · astro-ph.CO · Submitted 2025-11-07 · Read on arXiv

Listen

Transcript

Introduction to the show: ident: Astrophysics Radio. Generated commentary on the latest astrophysics papers.

Vera: Today's paper: "Sym-EFT: Accelerating Effective Field Theory of Large Scale Structure with Symbolic Regression".

Jocelyn: An emulator suite for one- and two-loop cold dark matter power spectra from the Effective Field Theory of Large Scale Structures (EFTofLSS) has been developed using symbolic regression to provide…

Vera: First, who's behind it and why it matters.

Paper summary: Vera: We’ve covered a bit about how this paper aims to solve computational bottlenecks in structure formation calculations by introducing an emulator suite for one- and two-loop cold dark matter power spectra using EFTofLSS <ref:2511.05093#pg0>.

Jocelyn: That sounds like it’s tackling the problem of needing computationally cheap ways to model the non-linear regime where linear perturbation theory fails, which is a big deal for observational cosmology <ref:2511.05093#pg1>.

Subrahmanyan: The thesis here is that by using symbolic regression, they can generate explicit mathematical expressions that allow for ultra-fast and highly accurate predictions suitable for parameter estimation in cosmological surveys <ref:2511.05093#pg4>.

Vera: So, the core claim is that this technique allows for the efficient computation of complex non-linear structure formation models beyond linear perturbation theory, overcoming the time-costly nature of traditional N-body simulations <ref:2511.05093#pg0>.

Jocelyn: Why does this matter practically? Because it offers a robust framework for testing various cosmological extensions, which is crucial when we’re trying to see if our models match the actual universe we observe <ref:2511.05093#pg4>.

Subrahmanyan: This method essentially extends standard perturbation theory by including non-linearities at smaller scales while treating short-wavelength perturbations by splitting fields using a top-hat filter at a cut-off scale <ref:2511.05093#pg1>.

Vera: And they capture the effect of integrating out these non-linearities through counterterms, which are time-dependent functions that aren't computable within the EFT itself <ref:2511.05093#pg0>.

Jocelyn: That means they’ve created a system where we can use the EFT structure to guide our calculations while using symbolic regression to efficiently handle the non-linear parts <ref:2511.05093#pg4>.

Subrahmanyan: It's essentially creating an intermediate regime between large and small scales where this theoretical framework provides additional insights into how the Universe is evolving <ref:2511.05093#pg1>.

Vera: So, to summarize, the paper presents a suite of emulators that use symbolic regression within the EFTofLSS framework to predict power spectra accurately and quickly for one- and two-loop contributions <ref:2511.05093#pg0>.

Jocelyn: It’s about providing a practical computational tool that lets us explore cosmological models much more thoroughly than was possible before <ref:2511.05093#pg4>.

Subrahmanyan: The significance is that it provides a means to efficiently compute structure formation models that are necessary for testing various cosmological extensions <ref:2511.05093#pg0>.

Vera: This moves us closer to being able to run more complex analyses on large datasets by bypassing the limitations of traditional, time-consuming simulation methods <ref:2511.05093#pg4>.

Jocelyn: It seems like a major step in making advanced non-linear structure formation modeling accessible for broader use in cosmology <ref:2511.05093#pg4>.

Conclusion: Vera: Thinking about the paper "Sym-EFT: Accelerating Effective Field Theory of Large Scale Structure with Symbolic Regression," it’s clear that the authors have successfully developed a method to accelerate our understanding of non-linear structure formation by coupling EFTofLSS with symbolic regression <ref:2511.05093#pg0>.

Jocelyn: It really speaks to how powerful combining theoretical physics frameworks, like EFT and machine learning techniques, can be when the goal is practical application in cosmology <ref:2511.05093#pg4>.

Subrahmanyan: The overall implication is that we now have a tool capable of ultra-fast testing of CDM in the mildly non-linear regime, allowing us to probe parameter space more efficiently than previously possible <ref:2511.05093#pg4>.

Vera: So, in simpler terms, it means we can get much faster and more accurate answers when trying to figure out the properties of dark matter clustering at intermediate scales <ref:2511.05093#pg4>.

Jocelyn: That efficiency translates directly into being able to test a wider variety of cosmological models against observational constraints from surveys <ref:2511.05093#pg4>.

Subrahmanyan: The work suggests that we can now use these techniques to gain more insight into how non-linearities affect the evolution of the Universe across different scales <ref:2511.05093#pg4>.

Vera: And with the potential link to CMB lensing, this opens up new ways for us to use data from other sources like the CMB to constrain these models <ref:2511.05093#pg4>.

Jocelyn: It’s about moving beyond what we can calculate by hand or with slow simulations and instead having a powerful, fast engine to explore those difficult parameter spaces <ref:2511.05093#pg4>.

Subrahmanyan: The authors have provided a versatile technique that is adaptable to fitting any type of counterterm parametrization, which suggests this method will be useful across many different theoretical scenarios <ref:2511.05093#pg0>.

Vera: Ultimately, it’s about providing a concrete way to handle the computational demands of modern structure formation physics without sacrificing the precision we need for cosmological measurements <ref:2511.05093#pg4>.

Jocelyn: The Sym-EFT approach is a significant advancement in how we can use theoretical tools to inform and test our observational data <ref:2511.05093#pg4>.

Despoina Farakou, Constantinos Skordis

CEICO—FZU, Institute of Physics of the Czech Academy of Sciences · Institute of Theoretical Physics, Faculty of Mathematics and Physics, Charles University · Department of Physics, University of Oxford

astro-ph.CO

Submitted: 2025-11-07

Updated: 2025-11-07

Comments: 21 pages, 17 figures, 4 tables

Journal ref: Mon Not R Astron Soc (2026)

DOI: 10.1093/mnras/stag1663

Code: https://github.com/heal-research/operon

License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

Importance score: 92/100

The gist: An emulator suite for one- and two-loop cold dark matter power spectra from the Effective Field Theory of Large Scale Structures (EFTofLSS) has been developed using symbolic regression to provide

Key concepts

EFTofLSS: This framework extends standard perturbation theory to include non-linear effects at smaller scales. It splits fields into short and long wavelengths using a filter and uses counterterms to capture the integrated out non-linearities, making it useful where linear theory fails.
Symbolic Regression: A supervised machine learning method used here to automatically generate explicit mathematical expressions directly usable in computer code. It minimizes a loss function to fit data, allowing for an interpretable and differentiable way to interpolate complex power spectra.
Emulator Construction: The process of building fast computational models by training them on existing calculations. The researchers trained emulators using 200 cosmologies sampled across a parameter space, focusing on the k-range where EFTofLSS is relevant (0.05 < k < 1 hMpc⁻¹).
Power Spectra Terms: The suite includes ten specific emulators corresponding to different mathematical terms in the EFTofLSS power spectrum, such as [k²P¹¹(k)]∥⁰ and [P²-loop(k)]∥⁰. These cover various orders of loop corrections and are designed to be computationally efficient.

Terminology

Symbolic Regression — 40× in this paper · explained in 5 episodes
Accuracy — 21× in this paper · explained in 1 episode
Matter Power Spectrum — 17× in this paper · explained in 2 episodes
Core — 8× in this paper · explained in 2 episodes
N-body simulations — 7× in this paper · explained in 2 episodes
Clustering — 3× in this paper · explained in 2 episodes
Generalization — 2× in this paper · explained in 13 episodes
Benchmark — 1× in this paper · explained in 2 episodes
Boltzmann Equation — 1× in this paper · explained in 2 episodes
Consistency — 1× in this paper · explained in 3 episodes
Gravitational Lensing — 1× in this paper · explained in 2 episodes
Loss Function — 1× in this paper · explained in 2 episodes
Measurements of the Cosmic Microwave Background — 1× in this paper
Root Mean Square Error — 1× in this paper
Simons Observatory — 1× in this paper
South Pole Telescope — 1× in this paper

Summary

An emulator suite for one- and two-loop cold dark matter power spectra from the Effective Field Theory of Large Scale Structures (EFTofLSS) has been developed using symbolic regression to provide ultra-fast, highly accurate predictions suitable for parameter estimation in cosmological surveys. This technique is significant because it allows for the efficient computation of complex non-linear structure formation models beyond linear perturbation theory, overcoming the time-costly nature of traditional N-body simulations and offering a robust framework for testing various cosmological extensions.

EFTofLSS Framework

The Effective Field Theory of Large Scale Structure (EFTofLSS) extends standard perturbation theory by including the effects of non-linearities at smaller scales, making it valuable in the intermediate regime between large and small scales where linear theory breaks down. This framework treats short-wavelength perturbations by splitting fields into short- and long-wavelength parts using a top-hat filter at a cut-off scale, leading to equations involving long wavelength fields similar to those used in perturbation theory. The effect of integrating out the non-linearities is captured through counterterms which are time-dependent functions that are otherwise not computable within the EFT.

Symbolic Regression for Emulators

The researchers employ symbolic regression, a supervised machine learning method, to generate explicit mathematical expressions that can be used directly in computer code. This allows for an interpretable and fully differentiable way of interpolating the desired power spectra. The process involves minimizing a loss function to fit the data, with a multi-objective strategy applied to select expressions of smaller length to avoid over-fitting and enable generalization, often utilizing Genetic Programming inspired by natural selection.

Emulator Construction and Training

The emulators are constructed by computing the terms of the EFTofLSS power spectrum in (14) and (16) using codes like CosmoEFT C++ and ResumEFT, passing the result through ResumEFT for IR resummation. The training set consists of 200 cosmologies drawn from a specified parameter space, sampled with a Latin hypercube. The researchers kept fixed k-values between 0.01 hMpc−1 and 3.3 hMpc−1, focusing on the region where EFTofLSS is relevant (0.05 < k < 1 hMpc−1).

Emulator Performance and Accuracy

The resulting emulators achieve errors better than 0.5% within the k-range of validity of EFT and maintain ultra-fast computational evaluation of less than ∼ 5 × 10−4 s on a single core. The performance is benchmarked against existing tools, showing that the Sym-EFT emulator is closer to codes like Pybird than CLASS-PT, with mean errors within 0.5% for 1σ and barely over for 2σ across 1000 cosmologies.

Specific Emulator Components

The suite includes ten emulators for the different terms in the EFTofLSS power spectra:

The term [k2P11(k)]∥0, which is emulated by a function whose functional form is close to the bare linear spectrum multiplied by k2.
The term [P11(k)]∥1, which is emulated by a complex function involving numerous cosmological parameters and k-dependence.
The term [P1−loop(k)]∥0, which is split into three overlapping regions in k-space (0.01–0.3 hMpc−1, 0.2–1 hMpc−1, and 0.9–3.3 hMpc−1) with separate emulators for each region joined by an error function to form the full function.
The term [P2-loop(k)]∥0, which is split into two overlapping regions in k-space (0.01–0.5 hMpc−1 and 0.4–3.3 hMpc−1) with separate emulators for each region joined by an error function to form the full function.
The term [P(cs)1-loop]∥0, which is emulated using a model of length 78 across all k-space regions.
The term [P(quad)1-loop]∥0, which is emulated using a model of length 72 across all k-space regions.

Applications and Future Directions

The primary application is the ultra-fast MCMC based testing of ΛCDM in the mildly non-linear regime. A direct application is to use these emulators with CMB data, as CMB lensing is sensitive to larger scales than galaxy lensing surveys and has a wider redshift range, suggesting a complementary effect of two-loop EFTofLSS in CMB lensing.

Improvements for AI systems

As an excellent, fastidious, and diligent researcher, I have analyzed this paper titled Sym-EFT: Accelerating Effective Field Theory of Large Scale Structure with Symbolic Regression. The core innovation is a symbolic regression-based emulator suite for calculating the one- and two-loop cold dark matter power spectrum from EFTofLSS.

Here are the specific improvements that can be made to AI systems, and what the resulting improved system can achieve:

)

Improve the speed and accuracy of cosmological parameter estimation in weakly non-linear regimes by integrating this emulator suite directly into MCMC pipelines.
Enable rapid, model-independent exploration of the parameter space for extensions beyond ΛCDM using symbolic regression functions as a highly efficient surrogate model for complex N-body simulations or high-order perturbation theory calculations.
Develop a hybrid inference framework that leverages the ultra-fast computation of the emulators to perform large parameter scans or Bayesian inference on cosmological models, significantly reducing computational overhead compared to running full EFTofLSS codes (CosmoEFT) or expensive N-body simulations for every step.
Create a tool for systematic error characterization and model comparison by allowing users to directly compare the performance of different theoretical approaches (e.g., comparing Sym-EFT against CosmoEFT or CLASS-PT) based on their predictive accuracy across various scales and redshift regimes, leveraging the provided error benchmarks.
Build a system capable of predicting observable quantities like CMB lensing angular power spectra by seamlessly integrating the 2-loop EFTofLSS emulator with Boltzmann solvers (like CLASS) via the provided user-friendly interface.

)

The improved AI system can perform:

Perform cosmological parameter inference in the mildly non-linear regime with significantly reduced computational cost, allowing for larger datasets and more complex model testing than currently feasible using standard MCMC techniques on full EFTofLSS codes.
Conduct high-throughput sensitivity studies for new physics by rapidly generating power spectrum predictions across vast regions of the parameter space, identifying promising regions for targeted, expensive N-body simulations or detailed analytical calculations (e.g., those involving 2-loop corrections).
Execute complex, multi-stage Bayesian inference workflows where the emulator acts as a fast likelihood evaluator for thousands of model iterations, drastically accelerating the convergence time of MCMC chains used to constrain cosmological parameters.
Serve as an advanced diagnostic tool that benchmarks the accuracy of various theoretical frameworks (EFTofLSS vs. other codes) by providing quantifiable error metrics across different scales and redshift regimes, ensuring that any adopted theoretical model is validated against empirical data constraints before being used in final inference pipelines.
Generate high-precision predictions for cosmological observables, such as CMB lensing signals, by efficiently coupling the symbolic emulator with cosmological background solvers to produce statistically significant results across a wide range of redshifts and angular scales.