How far can a two-loop galaxy power spectrum model extend cosmological scales?

A complete two-loop EFT model fits galaxy power spectra to smaller scales, but σ8 constraints barely improve over one-loop.

Direct answer

A complete two-loop effective field theory model of the redshift-space galaxy power spectrum pushes the maximum usable wavenumber from about 0.2 to 0.34 h Mpc⁻¹ for a luminous red galaxy sample at z = 0.5, and to at least 0.4 h Mpc⁻¹ for an emission-line galaxy sample at z = 1.1. The catch is that most of the new small-scale information is absorbed by 49 free coefficients, so constraints on σ8 are not significantly improved over the one-loop fit.

1sources cited

This article was generated with WisPaper-powered search and paper analysis.

What two-loop order actually adds

The paper builds the first complete two-loop model of the redshift-space galaxy power spectrum for biased tracers, extending earlier real-space two-loop models. One-loop perturbation theory describes the two-point statistic down to intermediate scales, roughly k ≈ 0.2 h Mpc⁻¹, and then breaks down at larger wavenumbers. Going to two-loop order means modelling every contribution to the galaxy density field up to fifth order in the initial conditions, including galaxy bias, redshift-space distortions and stochasticity, while handling both infrared and ultraviolet physics.

The model is derived in the Effective Field Theory of Large Scale Structure. In this framework, the galaxy overdensity is written as a perturbative expansion in the initial conditions, and uncertain small-scale physics is absorbed by renormalization into free coefficients rather than computed directly. The two-loop redshift-space model contains 49 such Wilson coefficients — biases, counterterms and stochasticity amplitudes — compared with 29 in real space. It also encodes all deterministic contributions to the one-loop bispectrum and trispectrum, plus the tree-level quadspectrum and pentaspectrum, many of which share the same parameters.

Why small scales were out of reach

The power spectrum is the two-point statistic of galaxy positions and remains a leading tool for analysing spectroscopic survey data. Its usefulness is limited by how accurately theory can describe non-linear dark matter collapse and galaxy formation. The one-loop EFT model has become the backbone of full-shape analyses, but it fails beyond intermediate wavenumbers, which blocks extraction of cosmological information from small-scale clustering.

Pushing to smaller scales is technically demanding because the model must self-consistently include bias, redshift-space distortions and stochasticity up to fifth order. Before this work, the complete two-loop model existed for galaxies in real space, and redshift-space two-loop results existed only for dark matter. This paper supplies the missing redshift-space galaxy version and, importantly, a numerical implementation fast enough to be used in fits.

A likelihood that runs in milliseconds

The loop integrals are precomputed numerically with the vegas algorithm, so the full likelihood can be evaluated in milliseconds. That speed matters because a two-loop fit varies many more nuisance parameters than a one-loop fit: the discussion notes that 12 nuisance parameters must be explicitly varied rather than 3. Without a fast implementation, the model would be impractical for repeated likelihood evaluations.

The model also includes Alcock–Paczyński corrections, which account for the fiducial cosmology used to convert observed redshifts and angles into physical positions. These are treated by rescaling the parallel and perpendicular coordinates, so the theory multipoles can be compared with data binned in wavenumber. Odd multipoles are generated only by relativistic and wide-angle effects and are ignored here.

How far the fit holds in simulations

The approach is validated against power spectra measured from high-resolution AbacusSummit simulations, covering both dark matter and realistic galaxy samples. The paper reports that the two-loop theory provides an excellent fit to both dark matter and galaxy power spectra, and that the higher-loop corrections substantially increase the range of scales over which the power spectrum can be modelled, including in the presence of non-linear fingers-of-God effects.

The headline scale comparison is a maximum wavenumber rising from 0.2 h Mpc⁻¹ to 0.34 h Mpc⁻¹ for a luminous red galaxy sample at z = 0.5. For an emission-line galaxy sample at z = 1.1, the two-loop theory returns unbiased constraints on σ8 up to at least 0.4 h Mpc⁻¹, while the one-loop theory breaks down beyond 0.2 h Mpc⁻¹, with a somewhat larger reach for emission-line galaxies. These conclusions hold for the mean of 25 realizations and for each simulation individually, up to slight prior-volume shifts of about 0.4σ in the latter.

Why σ8 gains stay small

The wider scale range does not translate into a much tighter constraint on σ8. The paper finds similar constraints on σ8 from the one-loop theory at k max ≈ 0.2 h Mpc⁻¹ as from the two-loop result at k max ≈ 0.3–0.4 h Mpc⁻¹. Much of the additional information is absorbed by the numerous fifth-order biases and counterterms.

The author offers an explanation: on small scales, variations of σ8 primarily change the broadband shape of the power spectrum, which is degenerate with a variety of non-linear phenomena and therefore with the EFT coefficients. This is a statement about σ8 specifically, not a claim that the two-loop model carries no new information. The paper suggests that tighter constraints may appear for other parameters, or when the power spectrum is combined with higher-order statistics, but states that this remains to be investigated.

Where the two-loop model does not yet transfer

The validation varies only σ8 and the Wilson coefficients. To assess the full information content, the predictions must be computed as a function of all relevant cosmological parameters, adding for example H0, Ωm and ns, and the full cosmology fits must be validated against simulations. The paper estimates this would require roughly ten times more loop integral templates, which it describes as computationally feasible.

The model has also not yet been applied to observational data. The paper expects this to be a straightforward extension of current pipelines, but notes it is more expensive because 12 nuisance parameters must be varied rather than 3. Until those steps are done, the demonstrated result is a validated model on simulations, not a new cosmological parameter measurement.

What the scale gain is worth

The paper's own priority is clear: the two-loop model buys scale reach, not an immediate gain in σ8 precision. For a luminous red galaxy sample at z = 0.5, the usable range extends to 0.34 h Mpc⁻¹; for an emission-line sample at z = 1.1, it extends to at least 0.4 h Mpc⁻¹. Those thresholds, rather than a generic claim of improved cosmology, are the concrete outcome of this work.

The natural next test is whether combining the two-loop power spectrum with the one-loop bispectrum or trispectrum, which rest on the same deterministic model, produces tighter parameter constraints than the power spectrum alone. The paper flags that combination, along with full cosmological-parameter fits, as the place where the two-loop model's value will actually be decided.

Sources used in this answer

1

The Redshift-Space Galaxy Power Spectrum at Two-Loop Order

A complete two-loop EFT model of the redshift-space galaxy power spectrum extends the maximum modelled wavenumber from 0.2 to 0.34 h Mpc⁻¹ for an LRG sample at z = 0.5 (and to at least 0.4 h Mpc⁻¹ for an ELG sample at z = 1.1), but σ8 constraints are not significantly improved because fifth-order biases and counterterms absorb much of the new information.