betaFeedback

WorksPhysics and Astronomy

Baryonic feedback and intrinsic alignments in profile: a frequentist view of DES Y3 cosmic shear

Abstract

Current cosmic shear surveys primarily constrain S8S_8 and ΩmΩ_\mathrm{m}, while most other model parameters remain dominated by priors. Consequently, marginal posteriors are vulnerable to prior-volume effects, particularly through degeneracies among S8S_8, baryonic feedback, and intrinsic alignment (IA) model parameters. To isolate data-driven constraints from prior volume effects, we compare profile likelihoods with Bayesian posteriors for DES Y3 cosmic shear under the DES Y3+KiDS-1000 Hybrid model. For cosmic shear alone, the profile global best fit and the marginal posterior mean agree in S8S_8, but the profile likelihood reveals a strong degeneracy between feedback strength and S8S_8, leaving the upper 1σ1σ limit on S8S_8 determined by the assumed feedback range. External CMB information on parameters other than ΩmΩ_\mathrm{m} and S8S_8 breaks this degeneracy, suppressing prior-volume effects and reconciling the profiles with the posteriors. With CMB data included, small-scale cosmic shear data prefers moderate baryonic feedback strengths consistent with hydrodynamical simulations calibrated to observed gas fractions. Furthermore, additional unconstrained parameters in the tidal alignment and tidal torquing (TATT) model result in weaker constraints, shifting and broadening the S8S_8 profile relative to the nonlinear linear alignment (NLA) model. Adding DES Y3 shear ratios mitigates this degeneracy, bringing profile likelihood and marginal posteriors into close agreement.

Introduction

Bayesian inference remains the standard framework for cosmological analyses, constraining the parameters of interest while marginalizing over the many additional parameters needed to model astrophysical and observational systematics. Marginalization, however, is vulnerable to prior-volume or projection effects. When weakly constrained nuisance parameters are degenerate with the parameters of interest, the choice of priors can displace the peak of the marginal posterior away from the maximum-likelihood estimate, producing apparent tensions that may not reflect the constraining power of the data [1, 2].

Profile likelihood offers a complementary frequentist approach that replaces marginalization with optimization. At each fixed value of the parameter of interest, the likelihood is maximized over all remaining parameters, yielding the best achievable data fit independently of prior volume. Profile likelihoods have been applied to several cosmological analyses. The first systematic application to Planck data found close agreement with the Bayesian Λ\LambdaCDM constraints and used the Feldman–Cousins prescription for the neutrino mass bound [3, 4], an approach since revisited with newer data including DESI [5, 6]. Profile likelihoods have also shown that Bayesian constraints are shaped by the prior in analyses of the effective number of neutrino species [7], early dark energy [8, 9], decaying dark matter [10], EFTofLSS full-shape clustering [11], and w0waw_0w_aCDM fits to DESI full-shape data [12]. This comparison also serves as a robustness check for dynamical dark energy, showing strong agreement between the two methodologies when applied to DESI, CMB, and supernova datasets [13].

Cosmic shear, the weak gravitational lensing of background galaxies by large-scale structure, is particularly susceptible to prior-volume effects because it constrains few combinations of its model parameters, primarily S8≡σ8(Ωm/0.3)0.5S_8 \equiv\sigma_8(\Omega_{\mathrm{m}}/0.3)^{0.5}, while most other model parameters remain prior dominated. Stage-III surveys, including the Dark Energy Survey (DES) [14, 15], the Kilo-Degree Survey (KiDS) [16], and the Hyper Suprime-Cam (HSC) [17, 18], have generally found values of S8S_8 below the prediction from Planck CMB measurements [19]. More recent results are mixed. A joint analysis of DES Y3 and KiDS-1000 with a common “Hybrid” model is consistent with Planck at the 1.7σ1.7\sigma level [20], and the final KiDS-Legacy analysis agrees with Planck, a change attributed primarily to improved redshift calibration [21]. DES Year 6, in contrast, continues to find a lower S8S_8 [22].

Because the data constrain little beyond S8S_8, the marginal posteriors depend on the prior volume of the remaining parameters. In the Hybrid analysis, the marginal mean S8=0.790−0.014+0.018S_8 = 0.790^{+0.018}_{-0.014} lies below the maximum a posteriori (MAP) value of 0.8010.801, a difference the authors attribute to posterior skewness and to projection of the asymmetric neutrino mass prior; with Σmν\Sigma m_\nu fixed, the two agree [20]. Recent stage-III analyses therefore also report the MAP, which is noisy to estimate from chains, and projected joint highest posterior density (PJ-HPD) intervals [14, 15, 16, 20, 22, 23]. These measures reduce the problem within the Bayesian framework, whereas a profile likelihood removes the dependence on prior volume altogether (Sec. II and references therein).

Baryonic feedback is a prominent example of such a degeneracy. Stronger suppression of the small-scale matter power spectrum allows a higher S8S_8 for the same lensing signal [24, 25]; see [26] for a review of feedback effects on the matter power spectrum. Marginalizing over feedback with halo-model or principal-component parameterizations removes the resulting biases at the cost of weaker constraints [27, 28, 29, 30]. It also introduces projection effects, which in simulated LSST analyses are dominated by baryonic feedback [31]. The Hybrid model varies log⁡10(TAGN/K)\log_{10}(T_{\mathrm{AGN}}/\mathrm{K}) over [7.3,8.0][7.3, 8.0], and replacing it with a dark-matter-only power spectrum lowers S8S_8 by 0.4σ0.4\sigma [20]. In DES Year 6, S8S_8 increases as the upper prior bound on log⁡10(TAGN/K)\log_{10}(T_{\mathrm{AGN}}/\mathrm{K}) is raised [22]. The marginal S8S_8 may therefore depend on where the prior on TAGNT_{\mathrm{AGN}} ends.

Analyses that constrain the feedback strength with small scales or external data find that the inferred S8S_8 follows the preferred feedback strength. Including small scales in DES Y3 with a flexible baryon model raises S8S_8 by 0.2σ0.2\sigma [32]. Combining DES Y3 with kinetic Sunyaev–Zel’dovich measurements prefers stronger feedback and a higher S8S_8 [33], and calibrating the feedback with X-ray and kSZ data raises S8S_8 from 0.803 to 0.832, compared with scale cuts [34]. KiDS-Legacy and HSC Year 3, in contrast, find little evidence for strong feedback. In KiDS-Legacy, fixing the cosmology to Planck gives log⁡10(TAGN/K)≤7.71\log_{10}(T_{\mathrm{AGN}}/\mathrm{K}) \le7.71 at 68% confidence [21], and HSC Year 3 small scales allow at most 5% suppression of the matter power spectrum [35].1 The BAHAMAS simulations reproduce the observed halo gas fractions for log⁡10(TAGN/K)\log_{10}(T_{\mathrm{AGN}}/\mathrm{K}) between 7.6 and 8.0 [39, 40].

The choice of intrinsic alignment (IA) model also shifts S8S_8. Replacing the fiducial nonlinear alignment (NLA) model of the Hybrid analysis with the tidal alignment and tidal torquing (TATT) model lowers S8S_8 by 0.9σ0.9\sigma and increases its uncertainty by 35% [20]. The MAP values of the two models differ by 1.3σ1.3\sigma as well, so this shift is not only a projection effect [20]. In the DES Y3 cosmic shear analysis, which uses additional calibration data to break the IA–redshift uncertainty degeneracy, the NLA and TATT posteriors differ by 0.5σ0.5\sigma in S8S_8, and extending the TATT analysis to smaller scales removes this difference [15].

These degeneracies motivate a profile-likelihood analysis of cosmic shear. So far, related ideas have been used mainly to marginalize analytically over nuisance parameters for speed [41]. Public profile likelihood codes such as PINC [2], Procoli [42] and Prospect [43] rely on simulated annealing, while combining neural-network emulators with gradient-based optimizers reduces the cost further [44]. A full profile requires a global optimization at each value of the profiled parameter, on a likelihood with strong degeneracies, extended flat directions and multiple local minima. For cosmic shear, each optimization needs many evaluations of the nonlinear matter power spectrum, which dominate the cost.

In this work, we build a computationally efficient profile likelihood pipeline that couples CMA-ES optimization to a fast cosmic shear likelihood evaluation in CosmoSIS [45] using a matter power spectrum emulator. We validate the analysis on noiseless and noisy synthetic data. We then apply it to DES Y3 cosmic shear with the KiDS-overlapping area excised and the Hybrid model of [20], which allows a direct comparison with the published Bayesian results. We then test the robustness of the profile constraints to the range of TAGNT_{\mathrm{AGN}} and the inclusion of small scales, to the IA model (NLA or TATT, with a1a_1 free or restricted to positive values) and the addition of the DES Y3 shear ratios, which constrain the source redshifts and IA independently of cosmic shear [46], to the sum of neutrino masses, and to the addition of CMB likelihood.

Parameter inference: Bayesian vs frequentist

Bayesian inference

The quantity of interest is the posterior distribution of the model parameters θ=(Θ,n)\boldsymbol{\theta} = (\boldsymbol{\Theta}, \boldsymbol{n}), where Θ\boldsymbol{\Theta} are the parameters of interest and n\boldsymbol{n} denotes the nuisance parameters. It is given by Bayes’ theorem,

P(θ∣d)∝L(d∣θ)Π(θ),(1)\mathcal{P}(\boldsymbol{\theta}\mid\boldsymbol{d}) \propto\mathcal{L}(\boldsymbol{d}\mid\boldsymbol{\theta})\Pi(\boldsymbol{\theta}), \tag*{(1)}

where L(d∣θ)\mathcal{L}(\boldsymbol{d}\mid\boldsymbol{\theta}) is the likelihood of the data vector d\boldsymbol{d} and Π(θ)\Pi(\boldsymbol{\theta}) is the prior. Constraints on Θ\boldsymbol{\Theta} are obtained using the marginal posterior by integrating the full posterior over n\boldsymbol{n},

P(Θ∣d)∝∫L(d∣Θ,n)Π(Θ,n) dn.(2)\mathcal{P}(\boldsymbol{\Theta}\mid\boldsymbol{d}) \propto\int\mathcal{L}(\boldsymbol{d}\mid\boldsymbol{\Theta},\boldsymbol{n})\Pi(\boldsymbol{\Theta},\boldsymbol{n})\,\mathrm{d}\boldsymbol{n}. \tag*{(2)}

The posterior assigns a probability directly to Θ\boldsymbol{\Theta}. Therefore, the 1D intervals are quoted as regions of fixed posterior mass: the 1σ1\sigma interval on a parameter is the range containing 68.3% of the marginal posterior, and the 2σ2\sigma interval the range containing 95.4%.

Frequentist inference

The frequentist approach maximizes the likelihood as a function of the parameter of interest without introducing any priors. The profile likelihood for Θ\boldsymbol{\Theta} is obtained by maximizing the likelihood over n\boldsymbol{n},

LP(Θ)≡max⁡nL(d∣Θ,n)=L(d∣Θ,n^(Θ)),(3)\mathcal{L}_{\mathrm{P}}(\boldsymbol{\Theta}) \equiv\max_{\boldsymbol{n}} \mathcal{L}(\boldsymbol{d}\mid\boldsymbol{\Theta},\boldsymbol{n}) = \mathcal{L}(\boldsymbol{d}\mid\boldsymbol{\Theta},\hat{\boldsymbol{n}}(\boldsymbol{\Theta})), \tag*{(3)}

where n^(Θ)\hat{n}(\boldsymbol{\Theta}) denotes the value of nn that maximizes the likelihood at fixed Θ\boldsymbol{\Theta}. The confidence intervals are constructed from the likelihood ratio test statistic,

Δχ2(Θ)=−2ln⁡[Lp(Θ)Lmax⁡]=χ2(Θ,n^(Θ))−χmin⁡2,(4)\Delta\chi^2(\boldsymbol{\Theta})=-2\ln\left[\frac{\mathcal{L}_p(\boldsymbol{\Theta})}{\mathcal{L}_{\max}}\right]=\chi^2(\boldsymbol{\Theta},\hat{n}(\boldsymbol{\Theta}))-\chi^2_{\min}, \tag*{(4)}

where Lmax⁡≡L(d∣θ^)\mathcal{L}_{\max}\equiv\mathcal{L}(d\mid\hat{\boldsymbol{\theta}}) denotes the global maximum likelihood point at θ^\hat{\boldsymbol{\theta}}. For a flat prior on nn, the marginal posterior and the profile likelihood are related by

P(Θ)∝Lp(Θ)V(Θ),V(Θ)≡∫L(d∣Θ,n)Lp(Θ) dn,(5)\mathcal{P}(\boldsymbol{\Theta})\propto\mathcal{L}_p(\boldsymbol{\Theta})V(\boldsymbol{\Theta}),\qquad V(\boldsymbol{\Theta})\equiv\int\frac{\mathcal{L}(d\mid\boldsymbol{\Theta},n)}{\mathcal{L}_p(\boldsymbol{\Theta})}\,\mathrm{d}n, \tag*{(5)}

where V(Θ)V(\boldsymbol{\Theta}) is the volume of nuisance-parameter space that fits the data well at fixed Θ\boldsymbol{\Theta}. The posterior is the profile likelihood weighted by this volume. We illustrate the different scenarios comparing the Bayesian and Frequentist approaches in a toy model in App. VII.

Frequentist confidence intervals are set by a threshold on the test statistic Δχ2\Delta\chi^2. For a single parameter of interest, the 1σ1\sigma interval is the range with Δχ2≤1\Delta\chi^2\leq1, and the 2σ2\sigma interval is the range with Δχ2≤4\Delta\chi^2\leq4. Joint two-dimensional regions are obtained by profiling over all remaining parameters on a grid in the two parameters of interest, and the corresponding thresholds are Δχ2≤2.30\Delta\chi^2\leq2.30 and Δχ2≤6.18\Delta\chi^2\leq6.18, since the test statistic then follows a χ2\chi^2 distribution with two degrees of freedom. In both cases, the number of degrees of freedom is set by the number of parameters of interest. This mapping relies on Wilks’ theorem [47], whose asymptotic assumptions can break down when the likelihood is strongly non-Gaussian, when parameters are poorly constrained by the data, or when the best fit lies at the edge of the allowed parameter range. In such scenarios, the thresholds must be calibrated explicitly, with a Neyman construction based on mock data sets [48] or, near a physical boundary, the Feldman–Cousins prescription [4], to guarantee correct frequentist coverage as shown in [2] in the context of CMB data.

Numerical implementation

For cosmic shear, the likelihood minimization is difficult because the full parameter space has 16 (NLA) or 19 (TATT) dimensions, several unconstrained parameters sit at the edges of their parameter ranges at the optimum, and the likelihood has multiple local minima. We use the covariance matrix adaptation evolution strategy (CMA-ES) [49, 50, 51]. At each step, it draws a population of trial points from a multivariate Gaussian, ranks them by χ2\chi^2, and moves the mean towards the best points while updating the step size and the full covariance of the Gaussian from the successful steps. The covariance therefore aligns with the degeneracy directions of the likelihood, approaching the global minimum efficiently. The CMA-ES algorithm is about nine times faster than the simulated-annealing code PROCOLI [42] (Fig. 9 in App. VIII A).

Comparison on synthetic data

We compare the Bayesian and frequentist approaches on a noiseless synthetic cosmic shear data vector generated for a DES Y3-like survey with Λ\LambdaCDM scale cuts in Figure 1. With all parameters free as shown in Panel (A1), the profile likelihood is substantially broader than the marginal posterior and the two peaks do not coincide exactly. The reason is provided in Panel (A2) where several parameters not constrained by cosmic shear are driven to the edge of the chosen parameter range. Since the prior volume is negligible at the edges, the posteriors shrink compared to the profiles. Once the weakly constrained parameters are fixed in Panel (B1), the profile likelihood and marginal posteriors are in good agreement. Panel (B2) shows that the best-fit values along the profile track the same degeneracy direction as the posterior.

Figure 1. Comparison of Bayesian and frequentist approaches on one noiseless synthetic DES Y3-like cosmic shear data vector with Λ\LambdaCDM-optimized scale cuts, generated with NLA. Thin black dotted lines mark the input values used to generate the data vector. First row: With all parameters free, the profile likelihoods yield weaker constraints than the marginal posteriors. Horizontal bars, in the line style of their curve, show the 1σ1\sigma intervals of the profile (Δχ2=1\Delta\chi^2 = 1) and the posterior. Second row: Best-fit values of all other free parameters along the S8S_8 (left) and Ωm\Omega_{\mathrm{m}} (right) profiles of the first row, shown as their position within the parameter range (0 and 1 are the prior edges; grey bands lie outside the prior). Filled symbols mark best fits for unconstrained parameters along the profile that are located at the edges of their parameter ranges. The negligible prior volume near the edges suppresses the marginal posteriors relative to the profile likelihoods in the first row. Third row: With parameters other than S8S_8, Ωm\Omega_{\mathrm{m}} and a1a_1 fixed, the profiles and posteriors agree closely. The small offset between the S8S_8 profile and posterior is a prior-volume effect: once the Laplace estimate of the prior-volume term is included (dotted), the volume-corrected profile and the posterior agree closely. Fourth row: Profile trajectories of the fixed-parameter analysis over 2D posteriors. Circles follow the S8S_8 profile and squares the Ωm\Omega_{\mathrm{m}} profile; both are shown in the middle panel. The trajectories follow the degeneracy direction of the posterior. The posterior volume is larger at low S8S_8, which explains the slight shift in the S8S_8 posterior relative to the profile in the third row.

We validate the pipeline on synthetic data in App. VIII B. We test five input cosmologies drawn from a Latin hypercube (Fig. 11) and ten noisy realizations of the synthetic data vector (Fig. 12). For both IA models, the profile best fit recovers the input cosmology, and the interval obtained by averaging over the ten noise realizations contains the true values of S8S_8 and Ωm\Omega_{\mathrm{m}}. Note that profile likelihoods are not generally expected to be broader than posteriors. In the case of cosmic shear, this broadening occurs because the best-fit parameters lie near the boundaries of the optimization parameter space, introducing edge effects into the profile construction (see App. VII for details).

Data and theoretical modeling

We adopt the DES Y3 + KiDS-1000 ‘Hybrid’ model of Ref. [20] as our baseline and therefore analyze the KiDS-excised DES Y3 cosmic shear data vector and covariance of that work. This allows us to compare our profile likelihood results directly with the published Hybrid posteriors. We describe the data in Sec. III A, and the baseline model, including the NLA IA model, in Secs. III B–III F. The TATT IA model, which we use only as a robustness test, is described in Sec. III D 1. We use the notation of [20] throughout, supplemented by [15, 52] for additional IA modeling details.

KiDS-excised DES Y3 cosmic shear data

To reduce the cross-covariance between the two surveys, Ref. [20] removes the region that overlaps with the KiDS-1000 footprint from the DES Y3 footprint. This removes 337 deg2^2 of unmasked area, or 8% of the DES Y3 data. The redshift distributions of the four tomographic source bins are assumed to be unchanged. The effective number densities, shape noise, and shear responses are re-measured on the reduced footprint, and the analytic covariance [53, 54, 55] is recomputed [20]. In the Hybrid analysis, removing the overlap region changes S8S_8 by 0.004 [20]. The data vector consists of the auto- and cross-correlation functions ξ±(ij)(θ)\xi^{(ij)}_{\pm}(\theta) of the four source bins in 20 logarithmic angular bins between 2.5′ and 250′. As in Ref. [20], the baseline analysis uses the DES Y3 ‘ΛCDM-optimized’ scale cuts [14, 15], which retain 273 data points (184 in ξ+\xi_+ and 89 in ξ−\xi_-) [15]. In Sec. IV C we also include all scales θ≥2.5′\theta\geq2.5'.

Modeling cosmic shear

The shear two-point correlation functions are related to the E- and B-mode cosmic shear power spectra, CϵϵEEC^{\mathrm{EE}}_{\epsilon\epsilon} and CϵϵBBC^{\mathrm{BB}}_{\epsilon\epsilon}, by [20]

ξ±(ij)(θ)=14π∑ℓ≥2(2ℓ+1)d2,±2ℓ(θ)[CϵϵEE,(ij)(ℓ)±CϵϵBB,(ij)(ℓ)](6)\xi^{(ij)}_{\pm}(\theta) = \frac{1}{4\pi}\sum_{\ell\geq2}(2\ell+1)d^{\ell}_{2,\pm2}(\theta)\left[C^{\mathrm{EE},(ij)}_{\epsilon\epsilon}(\ell)\pm C^{\mathrm{BB},(ij)}_{\epsilon\epsilon}(\ell)\right] \tag*{(6)}

where dm,nℓd^{\ell}_{m,n} are the reduced Wigner dd-matrices. We average the model over each angular bin [52]. Including intrinsic alignments (IA), the shear power spectra of tomographic bins ii and jj are [15]

CϵϵEE,(ij)(ℓ)=CGG(ij)(ℓ)+CGI(ij)(ℓ)+CGI(ji)(ℓ)+CII,EE(ij)(ℓ),C^{\mathrm{EE},(ij)}_{\epsilon\epsilon}(\ell)=C^{(ij)}_{\mathrm{GG}}(\ell)+C^{(ij)}_{\mathrm{GI}}(\ell)+C^{(ji)}_{\mathrm{GI}}(\ell)+C^{(ij)}_{\mathrm{II,EE}}(\ell),
CϵϵBB,(ij)(ℓ)=CII,BB(ij)(ℓ),(7)C^{\mathrm{BB},(ij)}_{\epsilon\epsilon}(\ell)=C^{(ij)}_{\mathrm{II,BB}}(\ell), \tag*{(7)}

where G denotes gravitational shear and I intrinsic shape. In the flat-sky Limber approximation, the lensing term is [20]

CGG(ij)(ℓ)=∫0χhordχ W(i)(χ)W(j)(χ)fK2(χ)Pδ(ℓ+1/2fK(χ),χ),(8)C^{(ij)}_{\mathrm{GG}}(\ell)=\int_{0}^{\chi_{\mathrm{hor}}}\mathrm{d}\chi\,\frac{W^{(i)}(\chi)W^{(j)}(\chi)}{f_K^2(\chi)}P_{\delta}\left(\frac{\ell+1/2}{f_K(\chi)},\chi\right), \tag*{(8)}

with PδP_{\delta} the nonlinear matter power spectrum, χ\chi the comoving distance, χhor\chi_{\mathrm{hor}} the comoving distance to the horizon, and fK(χ)f_K(\chi) the comoving angular diameter distance, which equals χ\chi in the flat models considered here. The lensing efficiency of source bin ii is

W(i)(χ)=3ΩmH022c2χa(χ)∫χχhordχ′ ni(χ′)χ′−χχ′,(9)W^{(i)}(\chi)=\frac{3\Omega_{\mathrm{m}}H_0^2}{2c^2}\frac{\chi}{a(\chi)}\int_{\chi}^{\chi_{\mathrm{hor}}}\mathrm{d}\chi'\,n_i(\chi')\frac{\chi'-\chi}{\chi'}, \tag*{(9)}

where cc is the speed of light, ni(χ)=ni(z) dz/dχn_i(\chi)=n_i(z)\,\mathrm{d}z/\mathrm{d}\chi is the source redshift distribution, normalized to unity. The IA terms are [15]

CGI(ij)(ℓ)=∫0χhordχ W(i)(χ)nj(χ)χ2PGI(ℓ+1/2χ,χ),C^{(ij)}_{\mathrm{GI}}(\ell)=\int_{0}^{\chi_{\mathrm{hor}}}\mathrm{d}\chi\,\frac{W^{(i)}(\chi)n_j(\chi)}{\chi^2}P_{\mathrm{GI}}\left(\frac{\ell+1/2}{\chi},\chi\right),
CII(ij)(ℓ)=∫0χhordχ ni(χ)nj(χ)χ2PII(ℓ+1/2χ,χ),(10)C^{(ij)}_{\mathrm{II}}(\ell)=\int_{0}^{\chi_{\mathrm{hor}}}\mathrm{d}\chi\,\frac{n_i(\chi)n_j(\chi)}{\chi^2}P_{\mathrm{II}}\left(\frac{\ell+1/2}{\chi},\chi\right), \tag*{(10)}

where the second line applies to both the E- and B-mode II spectra, and PGIP_{\mathrm{GI}} and PIIP_{\mathrm{II}} are given by the IA model (Sec. III D).

Matter power spectrum and baryon feedback

We compute PδP_{\delta} with HMCode2020 [40] from CAMB [56] linear power spectra; HMCode2020 includes massive neutrinos. Baryonic feedback, which suppresses the matter power spectrum on small scales, is described by one parameter, TAGNT_{\mathrm{AGN}}, the AGN heating temperature in the BAHAMAS hydrodynamical simulations [39] to which HMCode2020 is calibrated. We sample log⁡10(TAGN/K)\log_{10}(T_{\mathrm{AGN}}/\mathrm{K}). As in Ref. [20], the lower prior boundary, log⁡10(TAGN/K)=7.3\log_{10}(T_{\mathrm{AGN}}/\mathrm{K})=7.3, gives cosmic shear predictions close to a gravity-only model, and the upper boundary, 8.0, is the strongest feedback for which BAHAMAS matches the observed baryon fractions of galaxy groups. To speed up the profile likelihoods, we use the KiDS-Legacy CosmoPower [21, 57] emulator of HMCode2020 instead of CAMB+HMCode2020 [21] and validate the emulator against CAMB in Figure 13 and Figure 9. We compute angular power spectra and correlation functions with CosmoLike [53]. We adopt the COCOA [45] inference pipeline for likelihood evaluations and Bayesian inference.

Figure 9. CAMB and emulator profiles run using CMA-ES and PROCOLI for NLA (a1∈[−5,5]a_1 \in[-5, 5]) are consistent. The absolute difference between χ2\chi^2 values is ≤0.02\leq0.02.

Figure 13. Posterior comparison with public DES-KiDS (Σmν\Sigma m_\nu free), CAMB (Σmν\Sigma m_\nu free), and KiDS Legacy emulator (Σmν\Sigma m_\nu fixed) chains.

Intrinsic alignments: NLA baseline

As in Ref. [20], our baseline IA model is the nonlinear linear alignment model with power-law redshift evolution (NLA) [58, 59]. All results use NLA unless stated otherwise. In NLA, intrinsic shapes are linear in the tidal field, and the IA power spectra are proportional to the nonlinear matter power spectrum,

PGI(k,z)=A1(z)Pδ(k,z),P_{\mathrm{GI}}(k,z)=A_1(z)P_{\delta}(k,z),
PII,EE(k,z)=A12(z)Pδ(k,z),(11)P_{\mathrm{II,EE}}(k,z)=A_1^2(z)P_{\delta}(k,z), \tag*{(11)}

with PII,BB=0P_{\mathrm{II,BB}}=0. The amplitude is [15, 20]

A1(z)=−a1Cˉ1ρcritΩmD(z)(1+z1+zpivot)η1,(12)A_1(z)=-a_1\bar{C}_1\frac{\rho_{\mathrm{crit}}\Omega_{\mathrm{m}}}{D(z)}\left(\frac{1+z}{1+z_{\mathrm{pivot}}}\right)^{\eta_1}, \tag*{(12)}

where D(z)D(z) is the linear growth factor, ρcrit\rho_{\mathrm{crit}} the critical density, Cˉ1=5×10−14 M⊙−1h−2Mpc3\bar{C}_1=5\times10^{-14}\,M_{\odot}^{-1}h^{-2}\mathrm{Mpc}^3 a normalization constant fixed by convention, and zpivot=0.62z_{\mathrm{pivot}}=0.62 [20]. The free parameters are the amplitude a1a_1 and the power-law index η1\eta_1, denoted AIAA_{\mathrm{IA}} and ηIA\eta_{\mathrm{IA}} in Ref. [20]; we use a1a_1 and η1\eta_1 so that the notation is the same as for TATT (Sec. III D 1). With this sign convention, a1>0a_1>0 means that galaxies are aligned with the stretching direction of the tidal field, as expected for tidal alignment. We restrict our baseline analysis to a1≥0a_1\geq0, based on hydrodynamical simulations, direct observations and the shear ratios (Sec. IV A).

TATT extension

We use the ‘tidal alignment and tidal torquing’ (TATT) model [60], the baseline IA model of the DES Y3 analyses [15, 52], only to test how our results depend on the IA model (Sec. IV E). TATT expands the intrinsic shape field in the density and tidal field,

γ~α,IA=A1sα+A1δδsα+A2(s×s)α+⋯ .(13)\widetilde{\gamma}_{\alpha,\mathrm{IA}} = A_1s_\alpha+ A_{1\delta}\delta s_\alpha+ A_2(s \times s)_\alpha+ \cdots. \tag*{(13)}

The first term is the NLA term, the second term accounts for the weighting of the shape field by the source density, and the third term, quadratic in the tidal field, describes tidal torquing. Suppressing the redshift dependence of the coefficients, the IA power spectra are [15, 60]

PGI=A1Pδ+A1δP0∣0E+A2P0∣E2,PII,EE=A12Pδ+2A1A1δP0∣0E+A1δ2P0E∣0E+A22PE2∣E2+2A1A2P0∣E2+2A1δA2P0E∣E2,PII,BB=A1δ2P0B∣0B+A22PB2∣B2+2A1δA2P0B∣B2,(14)\begin{aligned} P_{\mathrm{GI}} &= A_1P_\delta+ A_{1\delta}P_{0|0\mathrm{E}} + A_2P_{0|\mathrm{E}2},\\ P_{\mathrm{II,EE}} &= A_1^2P_\delta+ 2A_1A_{1\delta}P_{0|0\mathrm{E}} + A_{1\delta}^2P_{0\mathrm{E}|0\mathrm{E}} + A_2^2P_{\mathrm{E}2|\mathrm{E}2}\\ &\quad+ 2A_1A_2P_{0|\mathrm{E}2} + 2A_{1\delta}A_2P_{0\mathrm{E}|\mathrm{E}2},\\ P_{\mathrm{II,BB}} &= A_{1\delta}^2P_{0\mathrm{B}|0\mathrm{B}} + A_2^2P_{\mathrm{B}2|\mathrm{B}2} + 2A_{1\delta}A_2P_{0\mathrm{B}|\mathrm{B}2}, \tag*{(14)} \end{aligned}

where the kk-dependent terms are one-loop integrals over the linear matter power spectrum, which we compute with FAST-PT [61, 62]. Unlike NLA, TATT produces B-mode power. The amplitude A1A_1 is given by Eq. (12), and [15]

A2(z)=5a2C‾1ρcritΩmD2(z)(1+z1+zpivot)η2,A_2(z) = 5a_2\overline{C}_1\frac{\rho_{\mathrm{crit}}\Omega_{\mathrm{m}}}{D^2(z)}\left(\frac{1+z}{1+z_{\mathrm{pivot}}}\right)^{\eta_2},
A1δ(z)=bTAA1(z),(15)A_{1\delta}(z) = b_{\mathrm{TA}} A_1(z), \tag*{(15)}

where bTAb_{\mathrm{TA}} is the linear bias of the source galaxies, taken to be constant in redshift. TATT has five free parameters: the tidal alignment amplitude a1a_1 and its redshift evolution η1\eta_1, the tidal torquing amplitude a2a_2 and its redshift evolution η2\eta_2, and bTAb_{\mathrm{TA}}. We use the DES Y3 priors for these (Table I).

NLA and TATT are nested models: for a2→0a_2 \to0 and bTA→0b_{\mathrm{TA}} \to0, TATT reduces to NLA, and η2\eta_2 has no effect [15, 20]. Both limits are needed: for fixed non-zero bTAb_{\mathrm{TA}}, TATT does not reduce to NLA as a2→0a_2 \to0, because the density-weighting term A1δ=bTAA1A_{1\delta} = b_{\mathrm{TA}}A_1 still adds the P0∣0EP_{0|0\mathrm{E}}, P0E∣0EP_{0\mathrm{E}|0\mathrm{E}}, and P0B∣0BP_{0\mathrm{B}|0\mathrm{B}} contributions in Eq. (14), including B-mode power.

Observational systematics

As in Ref. [20], we use the DES Y3 treatment of calibration uncertainties: For each source bin, the uncertainty in the redshift calibration is modeled with a shift parameter Δzi\Delta z_i, such that ni(z)→ni(z−Δzi)n_i(z) \to n_i(z-\Delta z_i), i.e. Δzi=⟨z⟩itrue−⟨z⟩iestimate\Delta z_i = \langle z\rangle_i^{\mathrm{true}} - \langle z\rangle_i^{\mathrm{estimate}} [14, 63]2, and the multiplicative shear calibration uncertainty is modeled as

ξ±(ij)→(1+mi)(1+mj)ξ±(ij)\xi_{\pm}^{(ij)} \to(1+m_i)(1+m_j)\xi_{\pm}^{(ij)}

[14, 64]. Both Δzi\Delta z_i and mim_i are calibrated with data and simulations external to the cosmic shear two-point analysis: the shear calibration with image simulations [64], and the redshift calibration with self-organizing-map photometric redshifts combined with clustering redshifts [63, 65]. The Hybrid model [20] does not include the DES Y3 shear-ratio likelihood [46], which in the DES Y3 analyses additionally constrains the redshift and IA parameters [14, 15]. We follow the Hybrid model for our baseline analysis, but add the shear-ratio likelihood to analysis variants that profile IA parameters (Sec. III F 1). Additive shear biases are removed at the catalog level by subtracting the mean ellipticity in each tomographic bin [66].

Likelihood and parameters

We assume a Gaussian likelihood,

−2ln⁡L(d∣θ)=[d−M(θ)]TC−1[d−M(θ)]+const.,(16)-2\ln\mathcal{L}(\boldsymbol{d}\mid\boldsymbol{\theta}) = [\boldsymbol{d}-\boldsymbol{M}(\boldsymbol{\theta})]^{\mathrm{T}}\boldsymbol{C}^{-1}[\boldsymbol{d}-\boldsymbol{M}(\boldsymbol{\theta})] + \mathrm{const.}, \tag*{(16)}

where M(θ)\boldsymbol{M}(\boldsymbol{\theta}) is the model data vector of Eq. (6) including systematics, and C\boldsymbol{C} is the analytic covariance of the KiDS-esque data vector [20]. As in Ref. [20], we sample the cosmological parameters {S8,ωc,ωb,h,ns}\{S_8, \omega_{\mathrm{c}}, \omega_{\mathrm{b}}, h, n_{\mathrm{s}}\}, where S8≡σ8Ωm/0.3S_8 \equiv\sigma_8\sqrt{\Omega_{\mathrm{m}}/0.3}, ωc≡Ωch2\omega_{\mathrm{c}} \equiv\Omega_{\mathrm{c}}h^2, ωb≡Ωbh2\omega_{\mathrm{b}} \equiv\Omega_{\mathrm{b}}h^2, and hh is the dimensionless Hubble parameter. The parameters Ωm\Omega_{\mathrm{m}} and σ8\sigma_8 are derived. While [20] also samples the sum of neutrino masses, Σmν\Sigma m_\nu, in our baseline analyses we instead fix Σmν=0.06 eV\Sigma m_\nu= 0.06\,\mathrm{eV}, the value at which the power spectrum emulator was trained, as this enables an order-of-magnitude computational speed-up (cf. Figure 9). In App. IX A, we vary Σmν\Sigma m_\nu at the cost of running CAMB instead of the emulator for the matter power spectrum. All other priors are those of Ref. [20] and are listed in Table I. For the profile likelihoods, the uniform priors set the parameter boundaries of the maximization. The redshift and shear calibration parameters are informed by external calibration data, which in a Bayesian analysis enter as Gaussian priors. A profile likelihood has no priors, and the consistent frequentist approach would be to fit the calibration data jointly with the shear data, but this is computationally not feasible. Instead, we approximate each calibration result as a Gaussian measurement of the corresponding nuisance parameter, with mean and width from Table I, and add it to the likelihood,

χnuisance2=∑i=14(Δzi−μΔz,iσΔz,i)2+∑i=14(mi−μm,iσm,i)2.(17)\chi^2_{\mathrm{nuisance}} = \sum_{i=1}^{4}\left(\frac{\Delta z_i-\mu_{\Delta z,i}}{\sigma_{\Delta z,i}}\right)^2 + \sum_{i=1}^{4}\left(\frac{m_i-\mu_{m,i}}{\sigma_{m,i}}\right)^2 . \tag*{(17)}

This is the standard treatment of auxiliary measurements in frequentist analyses [67, 68]. Since each Gaussian term adds one measurement and one parameter, it introduces no prior-volume effect and does not change the number of degrees of freedom of Δχ2\Delta\chi^2 (Sec. II). Profile analyses of galaxy clustering have treated external constraints the same way, for EFT nuisance parameters [11] and for BBN and nsn_s [12]. The approximation assumes that the calibration results are dominated by the calibration data rather than by the priors of those analyses.

When combining cosmic shear with the CMB, we add a compressed CMB Gaussian likelihood for (ωb,ns,h)(\omega_b, n_s, h) derived with mean and covariance taken from the Planck 2018 high-ℓ\ell TT,TE,EE chains [19]. Cosmic shear alone is not sensitive to these parameters; however, they set the shape of the linear matter power spectrum, so constraining them with CMB breaks their degeneracies with Ωm\Omega_m and S8S_8 and increases the constraining power of the analysis.

Shear ratios

For some analysis variants that profile the intrinsic-alignment parameters, we go beyond the Hybrid setup and add the DES Y3 small-scale shear-ratio (SR) likelihood of Ref. [46] for the MagLim lens sample, as in the DES Y3 cosmic shear analysis [14, 15]. The shear ratios constrain the source redshifts and intrinsic alignments independently of cosmic shear and so reduce the degeneracies between photo-zz and IA parameters [46]. The likelihood consists of nine ratios of galaxy–galaxy lensing measurements around the same lens bin (lens bins 1–3) for different source bins, on scales of 22–6 h−1Mpc6\,h^{-1}\mathrm{Mpc}, with a Gaussian likelihood and fixed covariance. Since the ratios are independent of the two-point measurements [46], their χSR2\chi^2_{\mathrm{SR}} is added to that of cosmic shear and the nuisance parameters (Eq. (18) below). The ratios are modelled as in DES Y3 [46, 52], with cosmology, intrinsic alignments, source redshifts and shear calibration shared with cosmic shear. The lens galaxies in bin ii have a linear bias bib_i, and lens magnification enters with fixed coefficients cilc_i^l [69]. Uncertainties in the lens redshift distributions are described by a shift Δzil\Delta z_i^l and a stretch sis_i about the mean redshift ⟨z⟩i\langle z\rangle_i, nil(z)→nil((z−Δzil−⟨z⟩i)/si+⟨z⟩i)/sin_i^l(z) \rightarrow n_i^l((z-\Delta z_i^l-\langle z\rangle_i)/s_i+\langle z\rangle_i)/s_i [70], with priors calibrated with clustering redshifts [71]. These nine lens parameters enter only the SR likelihood; we internally minimize over them at each evaluation of χSR2\chi^2_{\mathrm{SR}}, including the external calibration likelihood for the lens photo-zz parameters (analogous to Eq. (17), using the priors in Table II). We evaluate the SR likelihood with a fast re-implementation of the DES Y3 model. It agrees with the DES Y3 model to better than 1% in χSR2\chi^2_{\mathrm{SR}} for χSR2<100\chi^2_{\mathrm{SR}}<100 for both NLA and TATT. The SR data vector and covariance are measured on the full DES Y3 footprint; no KiDS-excised version exists, so the SR likelihood does not exactly match the excised area of the cosmic shear data vector.

Collecting the terms above, the total χ2\chi^2 minimized at each point of a profile is

χtot2=χcosmic shear2+χnuisance2+χCMB2+χSR2,(18)\chi^2_{\mathrm{tot}}=\chi^2_{\mathrm{cosmic\ shear}}+\chi^2_{\mathrm{nuisance}}+\chi^2_{\mathrm{CMB}}+\chi^2_{\mathrm{SR}}, \tag*{(18)}

where χcosmic shear2=−2ln⁡L\chi^2_{\mathrm{cosmic\ shear}}=-2\ln\mathcal{L} from Eq. (16) and χnuisance2\chi^2_{\mathrm{nuisance}} is given by Eq. (17). The last two terms are included only where stated: χCMB2\chi^2_{\mathrm{CMB}} in the analyses labelled “+CMB”, and χSR2\chi^2_{\mathrm{SR}} in those labelled “+ shear ratios”.

Data analysis: DES Y3 (KiDS excised)

Restriction of NLA parameter space

Our main results use the a1≥0a_1 \ge0 branch of the NLA profiles. When a1a_1 is allowed to be negative, we find that the a1<0a_1<0 part of the S8S_8 profile (S8<0.78S_8<0.78 in the top panel of Figure 6, discussed in Sec. IV E) has a local minimum at a1≈−2.7a_1\approx-2.7 and η1≈5\eta_1\approx5 (the prior edge). This implies a large negative alignment amplitude that grows quickly with redshift: the effective amplitude a1eff(z)≡a1[(1+z)/(1+zpivot)]η1a^{\mathrm{eff}}_{1}(z) \equiv a_{1}[(1+z)/(1+z_{\mathrm{pivot}})]^{\eta_{1}} reaches −1.8-1.8 at z=0.5z = 0.5 and −4.6-4.6 at z=0.8z = 0.8. In linear alignment theory, hydrodynamical simulations, and measurements of density–shape correlations [58, 72, 73, 74, 75, 76, 77], red galaxies are radially aligned (a1>0a_{1} > 0). For blue galaxies, which dominate the source sample, the sign is less certain. Hydrodynamical simulations disagree on the sign for disc galaxies [73, 78], and spiral galaxies in FLAMINGO are tangentially aligned, with a1=−0.76±0.16a_{1} = -0.76 \pm0.16 at z=0.5z = 0.5 [79]. Measured alignment amplitudes of blue galaxies are consistent with zero, with uncertainties of σ(a1)≃0.4\sigma(a_{1}) \simeq0.4–1 and ∣a1∣<0.78|a_{1}| < 0.78 (95% CL) for emission-line galaxies at z≈0.8z \approx0.8 [76, 77, 80, 81, 82]. A small negative amplitude is possible, but not one as large as the a1effa^{\mathrm{eff}}_{1} found on the a1<0a_{1} < 0 branch. In lensing data, the alignment amplitude is degenerate with the amplitude of the lensing kernel and thus with the source redshift calibration. The DES Y3 shear ratios reduce this degeneracy [46]. With them, the best-fit a1a_{1} is non-negative at every point of the NLA S8S_{8} profile. In TATT, a1a_{1} is degenerate with the tidal-torquing amplitude a2a_{2}, and the best fit has a small negative a1a_{1} and positive a2a_{2}, as in DES Y3 [15] (Sec. IV E). We also show profiles for the full range a1∈[−5,5]a_{1} \in[-5, 5] for comparison (Figure 6 and Figure 15 in App. IX B).

Figure 6. Profile comparison between NLA and TATT models with a1a_1 in range [−5,5][-5,5]. Thick lines mark points along the profile where the best fit has a1≥0a_1 \ge0, thin lines where it has a1<0a_1 < 0. Top: for cosmic shear alone, the NLA profile has a secondary peak on the a1<0a_1 < 0 branch at low S8S_8. The TATT best fit lies on the a1<0a_1 < 0 branch, and its profile peaks at lower S8S_8 and is much broader. Bottom: DES Y3 shear ratios remove the NLA a1<0a_1 < 0 branch and bring the NLA and TATT S8S_8 profiles and the marginal posteriors (dashed) into close agreement.

Figure 15. Both NLA and TATT profiles show bimodality in a1a_1. The best fit for NLA (TATT) lies in the a1≥0a_1 \ge0 (a1<0a_1 < 0) region. The addition of shear ratios helps constrain strong a1<0a_1 < 0 regions, bringing the posteriors and profiles for both models into agreement.

ParameterPrior
Lens redshift shift
Δz11\Delta z^{1}_{1}N(−0.009,0.007)\mathcal{N}(-0.009, 0.007)
Δz21\Delta z^{1}_{2}N(−0.035,0.011)\mathcal{N}(-0.035, 0.011)
Δz31\Delta z^{1}_{3}N(−0.005,0.006)\mathcal{N}(-0.005, 0.006)
Lens redshift stretch
s1s_{1}N(0.975,0.062)\mathcal{N}(0.975, 0.062)
s2s_{2}N(1.306,0.093)\mathcal{N}(1.306, 0.093)
s3s_{3}N(0.870,0.054)\mathcal{N}(0.870, 0.054)
Linear galaxy bias
bib_{i}U[0.8,3.0]\mathcal{U}[0.8, 3.0]
Lens magnification
C11,C21,C31C^{1}_{1}, C^{1}_{2}, C^{1}_{3}0.42, 0.30, 1.76 (fixed)

Table II. Lens parameters of the shear-ratio likelihood for MagLim lens bins i=1i = 1–3, with the DES Y3 priors [69, 71]. All other parameters are shared with cosmic shear (Table I).

ParameterPrior
Cosmology
S8S_8U[0.1, 1.3]\mathcal{U}[0.1,\,1.3]
ωc\omega_cU[0.051, 0.255]\mathcal{U}[0.051,\,0.255]
ωb\omega_bU[0.019, 0.026]\mathcal{U}[0.019,\,0.026]
hhU[0.64, 0.82]\mathcal{U}[0.64,\,0.82]
nsn_sU[0.84, 1.1]\mathcal{U}[0.84,\,1.1]
∑mν\sum m_\nu [eV]0.06 (fixed)a^{a}
Ωm,σ8\Omega_m, \sigma_8derived
Baryon feedback
log⁡10(TAGN/K)\log_{10}(T_{\mathrm{AGN}}/\mathrm{K})U[7.3, 8.0]b\mathcal{U}[7.3,\,8.0]^{b}
Intrinsic alignment (NLA, baseline)
a1a_1U[−5, 5]\mathcal{U}[-5,\,5] or U[0, 5]c\mathcal{U}[0,\,5]^{c}
η1\eta_1U[−5, 5]\mathcal{U}[-5,\,5]
Intrinsic alignment (TATT extension, in addition)
a2, η2a_2,\,\eta_2U[−5, 5]\mathcal{U}[-5,\,5]
bTAb_{\mathrm{TA}}U[0, 2]\mathcal{U}[0,\,2]
Redshift calibration
Δz1\Delta z_1N(0, 0.018)\mathcal{N}(0,\,0.018)
Δz2\Delta z_2N(0, 0.015)\mathcal{N}(0,\,0.015)
Δz3\Delta z_3N(0, 0.011)\mathcal{N}(0,\,0.011)
Δz4\Delta z_4N(0, 0.017)\mathcal{N}(0,\,0.017)
Shear calibration
m1m_1N(−0.0063, 0.0091)\mathcal{N}(-0.0063,\,0.0091)
m2m_2N(−0.0198, 0.0078)\mathcal{N}(-0.0198,\,0.0078)
m3m_3N(−0.0241, 0.0076)\mathcal{N}(-0.0241,\,0.0076)
m4m_4N(−0.0369, 0.0076)\mathcal{N}(-0.0369,\,0.0076)

Table I. TABLE I. Model parameters and priors. U[a,b]\mathcal{U}[a,b] denotes a uniform prior and N(μ,σ)\mathcal{N}(\mu,\sigma) a Gaussian prior with mean μ\mu and standard deviation σ\sigma. All priors are those of the DES Y3 + KiDS-1000 Hybrid analysis [20], except for the fixed neutrino mass. The TATT and calibration priors are those of DES Y3 [15], which Ref. [20] also uses. In the profile likelihoods, the Gaussian calibration priors enter as likelihood terms (Eq. (17)). a^{a} Hybrid prior U[0.055,0.6]\mathcal{U}[0.055,0.6] used in App. IX A. b^{b} Upper limit extended to 8.3 in Sec. IV C. c^{c} The restricted prior is used for the ‘a1≥0a_1 \ge0’ analyses.

Cosmic shear alone and combined with CMB

Our main result from DES Y3 (KiDS excised) cosmic shear data, analyzed with the NLA amplitude restricted to a1≥0a_{1} \ge0 and Σmν=0.06\Sigma m_{\nu} = 0.06 eV, is shown in Figure 2. Both the 2D profile likelihood constraints and the marginal posteriors enclose the maximum-likelihood point. We find the profile likelihood constraints to be weaker than the marginal posteriors because, away from the maximum-likelihood point, the likelihood is maximized at the edges of the parameter range. As described

Figure 2. 2D constraints from Bayesian (dashed contours) and profile likelihood (solid contours) analysis are consistent. Away from the best fit, the profile likelihood is maximized at the edges of the parameter ranges for poorly constrained parameters (see Figure 1 for intuition). These edges hold little prior volume, so they artificially shrink the marginal posteriors compared to the profiles.

in Figure 1, the prior volume is negligible at the boundaries, which makes the posteriors compact compared to the profiles.

Figure 3 shows the 1D constraints. For cosmic shear alone, the S8S_{8} and Ωm\Omega_{\mathrm{m}} profiles are broad and extend to higher values than the posteriors. We find,

Figure 3. Addition of compressed CMB Planck 2018 TT,TE,EE Gaussian likelihood (in ωb\omega_{b}, nsn_{s}, and hh) to cosmic shear brings profiles and posteriors into agreement.

S8=0.809−0.022+0.043(0.813±0.020),S_{8} = 0.809^{+0.043}_{-0.022}\quad(0.813 \pm0.020),
Ωm=0.301−0.083+0.075(0.278−0.042+0.051),(19)\Omega_{\mathrm{m}} = 0.301^{+0.075}_{-0.083}\quad(0.278^{+0.051}_{-0.042}), \tag*{(19)}

where the first value is the profile best fit with its Δχ2≤1\Delta\chi^{2} \le1 interval and the value in parentheses is the marginal posterior mean with its 68% credible interval. Adding the compressed Planck 2018 high-ℓ\ell TT,TE,EE Gaussian likelihood in ωb\omega_b, nsn_s and hh increases the constraining power and brings the profiles and posteriors into agreement,

S8=0.808±0.019 (0.805±0.017),S_8 = 0.808 \pm0.019\ (0.805 \pm0.017),
Ωm=0.309−0.036+0.043 (0.309−0.035+0.042).(20)\Omega_m = 0.309^{+0.043}_{-0.036}\ (0.309^{+0.042}_{-0.035}). \tag*{(20)}

In the noiseless synthetic test, the best-fit values of nsn_s, hh and ωb\omega_b jump between the edges of their parameter range along the profile (Figure 1, A2). When all the unconstrained parameters were fixed, the profile and posterior agree closely (Figure 1, B1). After adding the CMB likelihood, the same happens in Figure 3, where the best-fit parameters along the profile path no longer reach the parameter boundaries, bringing profiles and posteriors into agreement.

Robustness to baryon feedback modeling

The physical degeneracy between S8S_8 and baryonic feedback (TAGNT_{\mathrm{AGN}}) arises because both affect the matter power spectrum similarly on non-linear scales. In Figure 4, we find that for cosmic shear alone, strong feedback is degenerate with S8S_8 in the high-S8S_8 tail. When the upper limit on log⁡10(TAGN/K)\log_{10}(T_{\mathrm{AGN}}/\mathrm{K}) is extended from 8.0 to 8.3, the best fit remains unchanged, but the profile becomes effectively flat at high S8S_8, and its upper 1σ1\sigma limit moves to higher values. The shift in the S8S_8 posterior is therefore driven by the log⁡10(TAGN/K)\log_{10}(T_{\mathrm{AGN}}/\mathrm{K}) prior. The addition of the CMB likelihood removes this dependence, bringing the profiles and posteriors into agreement. The flatness in the S8S_8 profile with high log⁡10(TAGN/K)\log_{10}(T_{\mathrm{AGN}}/\mathrm{K}) in the top panel of Figure 4 is seen in the 2D profiles in Figure 5. For cosmic shear alone, the 2D contours rise towards higher S8S_8 as feedback is increased. The addition of the CMB likelihood breaks this degeneracy, making the contours nearly horizontal and S8S_8 constraints insensitive to the upper limit on log⁡10(TAGN/K)\log_{10}(T_{\mathrm{AGN}}/\mathrm{K}).

Figure 4. Profiles (solid) and posterior (dashed) comparison for two upper prior bounds on baryonic feedback strength. Top: For cosmic shear alone, the degeneracy between S8S_8 and feedback makes the profile flat at high S8S_8, and the flat part extends with the upper bound. Bottom: With the CMB likelihood, profiles and posteriors agree and do not depend on the upper bound on log⁡10(TAGN/K)\log_{10}(T_{\mathrm{AGN}}/\mathrm{K}).

Figure 5. 1D profiles of log⁡10(TAGN/K)\log_{10}(T_{\mathrm{AGN}}/\mathrm{K}) (top) and 2D profiles in log⁡10(TAGN/K)\log_{10}(T_{\mathrm{AGN}}/\mathrm{K}) and S8S_8 (bottom) for Λ\LambdaCDM-optimized scale cuts (left) and all scales ≥2.5′\geq2.5' (right), with the upper prior bound extended to 8.3. Cosmic shear alone is unable to break high S8S_8-high feedback degeneracy. With small scales and the CMB, the profile prefers moderate feedback. The global best-fit S8S_8 value is unaffected by the choice of scale cuts.

Constraining baryonic feedback with cosmic shear on small scales and CMB

In Figure 5, Λ\LambdaCDM-optimized scale cuts combined with the CMB give only an upper limit on log⁡10(TAGN/K)\log_{10}(T_{\mathrm{AGN}}/\mathrm{K}). Including scales down to 2.5′2.5' closes the 1σ1\sigma contour, and the profiles give,

log⁡10(TAGN/K)≤8.0(ΛCDM scale cuts+CMB),\log_{10}(T_{\mathrm{AGN}}/\mathrm{K}) \leq8.0 \qquad(\Lambda\mathrm{CDM\ scale\ cuts} + \mathrm{CMB}),
log⁡10(TAGN/K)=7.80−0.30+0.23(≥2.5′ scales+CMB).(21)\log_{10}(T_{\mathrm{AGN}}/\mathrm{K}) = 7.80^{+0.23}_{-0.30} \qquad(\geq2.5'\ \mathrm{scales} + \mathrm{CMB}). \tag*{(21)}

The best fit agrees with the calibrated BAHAMAS value, log⁡10(TAGN/K)=7.8\log_{10}(T_{\mathrm{AGN}}/\mathrm{K}) = 7.8 [39], and the interval largely overlaps the range 7.6–8.0 over which BAHAMAS reproduces the observed halo gas fractions. The data prefer moderate feedback, consistent with the upper limit obtained with Λ\LambdaCDM-optimized scale cuts and the CMB. With the CMB likelihood, the global best-fit S8S_8 is unaffected by the choice of scales. The Λ\LambdaCDM-optimized scale cuts therefore mitigate the impact of baryons without biasing S8S_8.

Robustness to IA modeling

The degeneracy between the IA parameters and S8S_8 arises because the IA terms CGIC_{\mathrm{GI}} and CIIC_{\mathrm{II}} in Eq. (7) add to the lensing term, CGG∝S82C_{\mathrm{GG}} \propto S_8^2. With our sign convention (Eq. (12)), a negative alignment amplitude a1a_1 makes CGIC_{\mathrm{GI}} positive, so a lower S8S_8 can be compensated by a more negative a1a_1. Extending NLA to TATT adds higher-order contributions to CGIC_{\mathrm{GI}} and CIIC_{\mathrm{II}}, which open further degenerate directions in the space of S8S_8, Ωm\Omega_{\mathrm{m}} and the IA parameters. The comparison between the NLA and TATT models is given in Figure 6. For cosmic shear alone (top row), the NLA best fit has a1≥0a_1 \ge0. The a1<0a_1 < 0 best-fit solution appears only at low S8S_8, outside the 1σ1\sigma interval. The TATT best fit has a1<0a_1 < 0 and lies at lower S8S_8. Cosmic shear alone constrains the additional TATT parameters only weakly, resulting in TATT profiles being much less constrained than NLA (as also seen for noiseless synthetic data in Figure 10). Adding shear ratios removes the NLA solutions with a1<0a_1 < 0 and brings the NLA and TATT S8S_8 profiles and posteriors into close agreement. We investigate IA models in more detail with 2D profiles in Figure 15 (App. IX B).

Figure 10. Same as the first row of Figure 1, showing comparison of NLA and TATT intrinsic-alignment models fitted to the same noiseless data vector, which was generated with NLA. The three additional TATT parameters add extra freedom, which shows up in the profile becoming broader, but is largely suppressed in the posterior.

Forecast for Roman (LSST wide) survey

To test whether the differences between profile likelihoods and marginal posteriors persist at Stage-IV precision, we perform a forecast on noiseless cosmic-shear data for a wide-area Roman survey covering the LSST Y10 footprint at Roman depth [83]. We assume the surface density of source galaxies neff=43 arcmin−2n_{\mathrm{eff}} = 43\,\mathrm{arcmin}^{-2}, split equally among ten tomographic bins, with shape noise σϵ=0.37\sigma_{\epsilon} = 0.37 (total). We compute ξ±\xi_{\pm} in 15 logarithmic angular bins between 2.5′2.5' and 250′250' with a Gaussian covariance computed with CosmoCov [54]. The mock data vector is computed assuming the NLA IA model and no photometric-redshift or shear-calibration bias. In simulated LSST cosmic shear analyses, the bias in marginal posteriors is caused by asymmetric prior boundaries relative to true parameter values [31]. Profile likelihood analyses demonstrate that this bias arises because the likelihood is maximized at the prior boundaries for poorly constrained parameters; asymmetric bounds truncate the high-likelihood volume non-uniformly, inducing a systematic shift in the marginal posterior away from the true values (see Figure 8 for a toy illustration). With higher statistical precision, we find that most parameters are well-constrained in the prior range, yielding similar constraining power for profile likelihoods and marginal posteriors (Figure 7).

Figure 7. Forecast for a wide Roman survey over the LSST Y10 footprint. Top: profile likelihoods (solid) and marginal posteriors (dashed), shown relative to the fiducial values (dotted line). Horizontal bars show the 1σ1\sigma intervals. Bottom: best-fit values of the other parameters along each profile, shown as their position within the prior range, as in Figure 1. The profiles recover the input values. Unlike the DES Y3-like case of Figure 1, most parameters are well constrained within the parameter boundaries, and profiles and posteriors agree more closely. The slight bias in the posteriors is caused by an asymmetric prior in weakly constrained parameters hh and ωb\omega_b around the true value.

Figure 8. Effect of prior (parameter) boundaries on marginal posterior (profile likelihood). We assume a Gaussian likelihood (Eq. (22)) with σ=1\sigma= 1 and σ~=0.3\tilde{\sigma} = 0.3.

Conclusion

We test the robustness of parameter inference by comparing Bayesian posteriors with frequentist profile likelihood for DES Y3 cosmic shear only and in combination with CMB. Our main findings are:

  • Cosmic shear is only sensitive to S8S_8 and Ωm\Omega_{\mathrm{m}}. Along the profile, the best-fit values of the remaining cosmological and nuisance parameters lie at the edges of their parameter ranges where prior volume is negligible. As a result, for DES Y3 cosmic shear alone, Bayesian posteriors are tighter than profile likelihoods.

  • Cosmic-shear-only constraints on S8S_8 are driven by the parameter range of the baryonic feedback model (log⁡10(TAGN/K)\log_{10}(T_{\rm AGN}/\mathrm{K})). Stronger feedback suppresses the small-scale power and can compensate for a higher S8S_8.

  • Extended intrinsic alignment models like TATT with additional parameters open new degenerate directions compared to NLA and are weakly constrained by cosmic shear data alone. Once shear ratios are incorporated, both Bayesian and profile likelihood constraints come in agreement.

  • External information from CMB likelihood on ωb\omega_b, nsn_s, and hh helps constrain the shape of the matter power spectrum, bringing profiles and posteriors into agreement. Cosmic shear data on small scales with CMB favors moderate feedback, consistent with hydrodynamical simulations calibrated to observed gas fractions.

  • For DES Y3 cosmic shear alone, the upper end of the S8S_8 profile depends on the assumed upper limit on the feedback strength (Sec. IV C), which limits what cosmic shear alone can say about the S8S_8 tension. With the CMB likelihood, profile and posterior agree with each other and are consistent with the Planck S8S_8 value, within 1.0σ1.0\sigma and 1.3σ1.3\sigma, respectively.

  • With the enhanced statistical precision of Stage-IV surveys, the constraining power of profile likelihoods is expected to improve significantly. Our forecast for Roman (LSST wide) survey indicates that profile likelihoods will deliver constraints comparable to those obtained from Bayesian posteriors.

Our results demonstrate that profile likelihoods provide an essential cross-check to assess the impact of parameter degeneracies and prior-volume effects on Bayesian constraints. For cosmic shear-only analyses, calibrating 1σ1\sigma profile intervals requires larger sets of noisy realizations, which we leave to future work.

Acknowledgments

We thank Elisa Ferreira, Tim Eifer, Jiachuan Xu, and Yu-Hsiu Huang for insightful discussions. AG, EK, and VM were supported in part by the Roman Project Infrastructure Team “Maximizing Cosmological Science with the Roman High Latitude Imaging Survey” (NASA contracts 80NM0018D0004-80NM0024F0012), NASA ATP “Modeling Large-Scale Structure - Interfacing Theory, Simulations, and ML- accelerated Inference” (award 23-ATP23-94) and Department of Energy grant DE-SC0025993. During this work, we used large language model assistants (Claude and Gemini) for code translation (from C to Python), code optimization, figure styling suggestions, and language editing of author-written text. All AI-assisted code was rigorously tested and verified by the authors.

Toy model: Bayesian vs frequentist inference

We compare Bayesian and frequentist parameter inference methods in a toy model with two parameters, Θ\Theta: parameter of interest, and nn: a nuisance parameter. We assume the likelihood to be a two-dimensional Gaussian centered at Θ=0=n\Theta= 0 = n,

−2ln⁡L(Θ,n)=Θ2σ2+(n−Θ)2σ~2,(22)-2\ln\mathcal{L}(\Theta,n)=\frac{\Theta^2}{\sigma^2}+\frac{(n-\Theta)^2}{\tilde{\sigma}^2}, \tag*{(22)}

where the second term introduces correlation between the two parameters. At each Θ\Theta, the profile likelihood Lp\mathcal{L}_p takes the maximum of L\mathcal{L} over the allowed range of nn. The posterior P\mathcal{P} integrates L\mathcal{L} over the same range.

The posterior is a weighted average of the likelihood over the prior volume, whereas the profile is only sensitive to the likelihood peak in that volume. We compare the two statistics for different choices of nuisance parameter volume V(Θ)V(\Theta) (see Eq. (5) for definition) in Figure 8. This range is the prior range for the posterior and bounds the maximization for the profile. When V(Θ)V(\Theta) spans the entire high-likelihood region, i.e., the priors are wide relative to the constraining power of the dataset, the two methods yield identical constraints. If V(Θ)V(\Theta) is asymmetric around the maximum likelihood point, the posterior is biased and skewed toward regions of larger prior volume. The profile is also skewed but always peaks at the maximum likelihood point. If V(Θ)V(\Theta) is symmetric, the posterior remains unbiased, but prior-volume effects shrink it relative to the profile.

Analysis validation

Emulator and profile likelihood code

The comparison of our analysis with public DES Y3 (Hybrid pipeline) is provided in Figure 13. The profile like-

likelihood code validation is provided in Figure 9.

Noiseless and noisy synthetic data

We validate our profile-likelihood pipeline on synthetic data vectors. We compare NLA and TATT models on noiseless datavector in Figure 10. We first test noiseless realizations (to verify recovery of the input cosmology) in Figure 11 and then noisy realizations (to check the behavior of the frequentist intervals) in Figure 12.

Figure 11. Validation on noiseless synthetic cosmic-shear data at five cosmologies drawn from a Latin hypercube. Each data vector is generated with NLA (wine) or TATT (olive) and fitted with the same IA model. For every cosmology and both IA models, the profile best fit recovers the input cosmology.

Figure 12. Validation on ten synthetic noisy realizations of cosmic shear data for NLA and TATT models. The interval constructed by averaging over ten noisy realizations contains the true value (black dotted line) at which noisy realizations were generated. More noise realizations are needed to calibrate the true coverage of the profile likelihood intervals.

Extended analysis

Freeing neutrino mass

Our baseline analysis fixes ∑mν=0.06 eV\sum m_\nu= 0.06\,\mathrm{eV}, whereas the Hybrid analysis of Ref. [20] varies it with a uniform prior U[0.055,0.6] eVU[0.055, 0.6]\,\mathrm{eV}. Figure 14 compares the two cases for cosmic shear alone with the NLA model and a1≥0a_1 \ge0. Massive neutrinos suppress the matter power spectrum below their free-streaming scale, acting as an additional parameter that is degenerate with S8S_8 and Ωm\Omega_{\mathrm{m}}. Freeing ∑mν\sum m_\nu leaves S8S_8 almost unchanged but slightly skews Ωm\Omega_{\mathrm{m}} profiles towards higher values.

Figure 14. Allowing ∑mν\sum m_\nu to vary opens another degeneracy direction, skewing the profiles towards higher S8S_8 and Ωm\Omega_{\mathrm{m}} values.

Intrinsic alignment: Removing a1≥0a_1 \ge0 restriction

Figure 15 compares the profile likelihood (top) with the posteriors (bottom) for the full range a1∈[−5,5]a_1 \in[-5, 5]. For cosmic shear alone, the profile is much broader than the posterior, by about a factor of 1.5 in S8S_8 for NLA and 2 for TATT. The secondary maximum in profiles with large a1<0a_1 < 0 values is nearly as good as the a1≥0a_1 \ge0 region. However, a1<0a_1 < 0 regions in the profile prefer best-fit values of other weakly constrained cosmological and IA parameters close to the edges of their priors. The posterior gives little weight to prior edges and only explores the a1≥0a_1 \ge0 region. The shear ratios constrain a1a_1 independently of cosmic shear and remove the a1<0a_1 < 0 fits. For both models, the global best-fit S8S_8 agree, and the profile is only moderately broader than the posterior.

References

References

  1. [1]Adrià Gómez-Valent. Fast test to assess the impact of marginalization in Monte Carlo analyses and its application to cosmology. Phys. Rev. D, 106(6):063506, September 2022. arXiv:2203.16285, [DOI], [ADS].
  2. [2]Laura Herold, Elisa G. M. Ferreira, and Lukas Heinrich. Profile likelihoods in cosmology: When, why, and how illustrated with ΛCDM, massive neutrinos, and dark energy. Phys. Rev. D, 111(8):083504, April 2025. arXiv:2408.07700, [DOI], [ADS].DOI
  3. [3]Planck Collaboration, P. A. R. Ade, N. Aghanim, M. Arnaud, et al. Planck intermediate results. XVI. Profile likelihoods for cosmological parameters. A&A, 566:A54, June 2014. arXiv:1311.1657, [DOI], [ADS].
  4. [4]G. J. Feldman and R. D. Cousins. Unified approach to the classical statistical analysis of small signals. Phys. Rev. D, 57:3873, 1998. arXiv:physics/9711021, [DOI].DOI
  5. [5]Daniel Naredo-Tuero, Miguel Escudero, Enrique Fernandez-Martinez, Xabier Marcano, and Vivian Poulin. Critical look at the cosmological neutrino mass bound. Phys. Rev. D, 110(12):123537, December 2024. arXiv:2407.13831, [DOI], [ADS].
  6. [6]D. Chebat, C. Yèche, E. Armengaud, N. Schöneberg, et al. Cosmological neutrino mass: a frequentist overview in light of DESI. J. Cosmology Astropart. Phys., 2026(1):041, January 2026. arXiv:2507.12401, [DOI], [ADS].arxiv.org/abs/2507.12401
  7. [7]Jan Hamann. Evidence for extra radiation? Profile likelihood versus Bayesian posterior. J. Cosmology Astropart. Phys., 2012(3):021, March 2012. arXiv:1110.4271, [DOI], [ADS].DOI
  8. [8]Laura Herold, Elisa G. M. Ferreira, and Eiichiro Komatsu. New Constraint on Early Dark Energy from Planck and BOSS Data Using the Profile Likelihood. ApJ, 929(1):L16, April 2022. arXiv:2112.12140, [DOI], [ADS].DOI
  9. [9]Laura Herold and Elisa G. M. Ferreira. Resolving the Hubble tension with early dark energy. Phys. Rev. D, 108(4):043513, August 2023. arXiv:2210.16296, [DOI], [ADS].DOI
  10. [10]Emil Brinch Holm, Laura Herold, Steen Hannestad, Andreas Nygaard, and Thomas Tram. Decaying dark matter with profile likelihoods. Phys. Rev. D, 107(2):L021303, January 2023. arXiv:2211.01935, [DOI], [ADS].arxiv.org/abs/2211.01935
  11. [11]Emil Brinch Holm, Laura Herold, Théo Simon, Elisa G. M. Ferreira, Steen Hannestad, Vivian Poulin, and Thomas Tram. Bayesian and frequentist investigation of prior effects in EFT of LSS analyses of full-shape BOSS and eBOSS data. Phys. Rev. D, 108(12):123514, December 2023. arXiv:2309.04468, [DOI], [ADS].arxiv.org/abs/2309.04468
  12. [12]J. Morawetz, H. Zhang, M. Bonici, W. J. Percival, and DESI Collaboration. Frequentist cosmological constraints from full-shape clustering measurements in DESI DR1. J. Cosmology Astropart. Phys., 2026(8):001, August 2026. arXiv:2508.11811, [DOI], [ADS].DOI
  13. [13]Laura Herold and Tanvi Karwal. Bayesian and frequentist perspectives agree on dynamical dark energy. Phys. Rev. D, 113(12):123551, June 2026. arXiv:2506.12004, [DOI], [ADS].DOI
  14. [14]A. Amon, D. Gruen, M. A. Troxel, N. MacCrann, and DES Collaboration. Dark Energy Survey Year 3 results: Cosmology from cosmic shear and robustness to data calibration. Phys. Rev. D, 105(2):023514, January 2022. arXiv:2105.13543, [DOI], [ADS].DOI
  15. [15]L. F. Secco, S. Samuroff, E. Krause, B. Jain, and DES Collaboration. Dark Energy Survey Year 3 results: Cosmology from cosmic shear and robustness to modeling uncertainty. Phys. Rev. D, 105(2):023515, January 2022. arXiv:2105.13544, [DOI], [ADS].DOI
  16. [16]Marika Asgari, Chieh-An Lin, Benjamin Joachimi, Benjamin Giblin, et al. KiDS-1000 cosmology: Cosmic shear constraints and comparison between two point statistics. A&A, 645:A104, January 2021. arXiv:2007.15633, [DOI], [ADS].DOI
  17. [17]Xiangchong Li, Tianqing Zhang, Sunao Sugiyama, Roohi Dalal, and HSC Collaboration. Hyper Suprime-Cam Year 3 results: Cosmology from cosmic shear two-point correlation functions. Phys. Rev. D, 108(12):123518, December 2023. arXiv:2304.00702, [DOI], [ADS].arxiv.org/abs/2304.00702
  18. [18]Roohi Dalal, Xiangchong Li, Andrina Nicola, Joe Zuntz, and HSC Collaboration. Hyper Suprime-Cam Year 3 results: Cosmology from cosmic shear power spectra. Phys. Rev. D, 108(12):123519, December 2023. arXiv:2304.00701, [DOI], [ADS].DOI
  19. [19]Planck Collaboration, N. Aghanim, Y. Akrami, M. Ashdown, et al. Planck 2018 results. VI. Cosmological parameters. A&A, 641:A6, September 2020. arXiv:1807.06209, [DOI], [ADS].
  20. [20]Dark Energy Survey and KiLo-Degree Survey Collaboration and T. M. C. Abbott et al. DES Y3 + KiDS-1000: Consistent cosmology combining cosmic shear surveys. The Open Journal of Astrophysics, 6:36, October 2023. arXiv:2305.17173, [DOI], [ADS].DOI
  21. [21]Angus H. Wright, Benjamin Stolzner, Marika Asgari, Maciej Bilicki, et al. KiDS-Legacy: Cosmological constraints from cosmic shear with the complete KiLo-Degree Survey. A&A, 703:A158, November 2025. arXiv:2503.19441, [DOI], [ADS].DOI
  22. [22]DES Collaboration, T. M. C. Abbott, M. Aguena, A. Alarcon, et al. Dark Energy Survey Year 6 Results: Cosmological Constraints from Cosmic Shear. arXiv e-prints, page arXiv:2602.10065, February 2026. arXiv:2602.10065, [DOI], [ADS].
  23. [23]B. Joachimi, C.-A. Lin, M. Asgari, T. Tröster, C. Heymans, H. Hildebrandt, F. Köhlinger, A. G. Sánchez, A. H. Wright, M. Bilicki, C. Blake, J. L. van den Busch, M. Crocce, A. Dvornik, T. Erben, F. Getman, B. Giblin, H. Hoekstra, A. Kannawadi, K. Kuijken, N. R. Napolitano, P. Schneider, R. Scoccimarro, E. Sellentin, H. Y. Shan, M. von Wietersheim-Kramsta, and J. Zuntz. KiDS-1000 methodology: Modelling and inference for joint weak gravitational lensing and spectroscopic galaxy clustering analysis. A&A, 646:A129, 2021. arXiv:2007.01844, [DOI], [ADS].DOI
  24. [24]Alexandra Amon and George Efstathiou. A non-linear solution to the S₈ tension? MNRAS, 516(4):5355–5366, November 2022. arXiv:2206.11794, [DOI], [ADS].DOI
  25. [25]Calvin Preston, Alexandra Amon, and George Efstathiou. A non-linear solution to the S₈ tension - II. Analysis of DES Year 3 cosmic shear. MNRAS, 525(4):5554–5564, November 2023. arXiv:2305.09827, [DOI], [ADS].arxiv.org/abs/2305.09827
  26. [26]Nora Elisa Chisari, Alexander J. Mead, Shahab Joudaki, Pedro G. Ferreira, Aurel Schneider, Joseph Mohr, Tilman Tröster, David Alonso, Ian G. McCarthy, Sergio Martin-Alvarez, and et al. Modelling baryonic feedback for survey cosmology. The Open Journal of Astrophysics, 2(1):4, June 2019. arXiv:1905.06082, [DOI], [ADS].DOI
  27. [27]Elisabetta Semboloni, Henk Hoekstra, and Joop Schaye. Effect of baryonic feedback on two- and three-point shear statistics: prospects for detection and improved modelling. MNRAS, 434(1):148–162, September 2013. arXiv:1210.7303, [DOI], [ADS].DOI
  28. [28]A. J. Mead, J. A. Peacock, C. Heymans, S. Joudaki, and A. F. Heavens. An accurate halo model for fitting non-linear cosmological power spectra and baryonic feedback models. MNRAS, 454(2):1958–1975, December 2015. arXiv:1505.07833, [DOI], [ADS].DOI
  29. [29]Tim Eifler, Elisabeth Krause, Scott Dodelson, Andrew R. Zentner, Andrew P. Hearin, and Nicolay Y. Gnedin. Accounting for baryonic effects in cosmic shear tomography: determining a minimal set of nuisance parameters using PCA. MNRAS, 454(3):2451–2471, December 2015. arXiv:1405.7423, [DOI], [ADS].DOI
  30. [30]Hung-Jin Huang, Tim Eifler, Rachel Mandelbaum, and Scott Dodelson. Modelling baryonic physics in future weak lensing surveys. MNRAS, 488(2):1652–1678, September 2019. arXiv:1809.01146, [DOI], [ADS].DOI
  31. [31]Naomi Clare Robertson, C. Heymans, J. Zuntz, P. Burger, C. D. Leonard, I. G. McCarthy, J. C. Paine, J. Salcido, N. Šarčević, M. Schaller, J. Schaye, and M. P. van Daalen. Impact and interplay of ΛCDM analysis choices for LSST cosmic shear. The Open Journal of Astrophysics, 9:68762, August 2026. arXiv:2602.21863, [DOI], [ADS].DOI
  32. [32]Giovanni Aricò, Raul E. Angulo, Matteo Zennaro, Sergio Contreras, Angela Chen, and Carlos Hernández-Monteagudo. DES Y3 cosmic shear down to small scales: Constraints on cosmology and baryons. A&A, 678:A109, October 2023. arXiv:2303.05537, [DOI], [ADS].
  33. [33]L. Bigwood, A. Amon, A. Schneider, J. Salcido, I. G. McCarthy, C. Preston, D. Sanchez, D. Sijacki, E. Schaan, S. Ferraro, and et al. Weak lensing combined with the kinetic Sunyaev-Zel’dovich effect: a study of baryonic feedback. MNRAS, 534(1):655–682, October 2024. arXiv:2404.06098, [DOI], [ADS].arxiv.org/abs/2404.06098
  34. [34]Leah Bigwood, Jamie McCullough, Jared Siegel, Alexandra Amon, George Efstathiou, David Sanchez-Cid, Elisa Legnani, Daniel Gruen, Jonathan Blazek, Cyrille Doux, and et al. Confronting cosmic shear astrophysical uncertainties: DES Year 3 revisited. arXiv e-prints, page arXiv:2512.04209, December 2025. arXiv:2512.04209, [DOI], [ADS].arxiv.org/abs/2512.04209
  35. [35]Ryo Terasawa, Xiangchong Li, Masahiro Takada, Takahiro Nishimichi, Satoshi Tanaka, Sunao Sugiyama, Toshiki Kurita, Tianqing Zhang, Masato Shirasaki, Ryuichi Takahashi, and et al. Exploring the baryonic effect signature in the Hyper Supreme-Cam Year 3 cosmic shear two-point correlations on small scales: The S₈ tension remains present. Phys. Rev. D, 111(6):063509, March 2025. arXiv:2403.20323, [DOI], [ADS].arxiv.org/abs/2403.20323
  36. [36]Hung-Jin Huang, Tim Eifler, Rachel Mandelbaum, Gary M. Bernstein, Anqi Chen, Ami Choi, Juan García-Bellido, Dragan Huterer, Elisabeth Krause, Eduardo Rozo, and et al. Dark energy survey year 1 results: Constraining baryonic physics in the Universe. MNRAS, 502(4):6010–6031, April 2021. arXiv:2007.15026, [DOI], [ADS].DOI
  37. [37]Jiachuan Xu, Tim Eifler, Vivian Miranda, Xiao Fang, Evan Saranaiov, Elisabeth Krause, Hung-Jin Huang, Karim Benabed, and Kunhao Zhong. Constraining baryonic physics with DES Y1 and Planck data: Combining galaxy clustering, weak lensing, and CMB lensing. Phys. Rev. D, 110(6):063532, September 2024. arXiv:2311.08047, [DOI], [ADS].arxiv.org/abs/2311.08047
  38. [38]Jiachuan Xu, Tim Eifler, Elisabeth Krause, Vivian Miranda, Jaime Salcido, and Ian McCarthy. Constraining baryonic feedback and cosmology from DES Y3 and Planck PR4 6×2pt data. I. ACMD models. arXiv e-prints, page arXiv:2510.25596, October 2025. arXiv:2510.25596, [DOI], [ADS].arxiv.org/abs/2510.25596
  39. [39]Ian G. McCarthy, Joop Schaye, Simeon Bird, and Amandine M. C. Le Brun. The BAHAMAS project: calibrated hydrodynamical simulations for large-scale structure cosmology. MNRAS, 465(3):2936–2965, March 2017. arXiv:1603.02702, [DOI], [ADS].DOI
  40. [40]A. J. Mead, S. Briedén, T. Tröster, and C. Heymans. HMcode-2020: improved modelling of non-linear cosmological power spectra with baryonic feedback. MNRAS, 502(1):1401–1422, March 2021. arXiv:2009.01858, [DOI], [ADS].DOI
  41. [41]Boryana Hadzhyiska, Kevin Wolz, Susanna Azzoni, David Alonso, Carlos García-García, Jaime Ruiz-Zapatero, and Anže Slosar. Cosmology with 6 parameters in the Stage-IV era: efficient marginalisation over nuisance parameters. The Open Journal of Astrophysics, 6:23, July 2023. arXiv:2301.11895, [DOI], [ADS].
  42. [42]Tani Karwal, Yashvi Patel, Alexa Bartlett, Vivian Poulin, Tristan L. Smith, and Daniel N. Pfeffer. Procoli: Profiles of cosmological likelihoods. arXiv e-prints, page arXiv:2401.14225, January 2025. arXiv:2401.14225, [DOI], [ADS].arxiv.org/abs/2401.14225
  43. [43]Emil Brinch Holm, Andreas Nygaard, Jeppe Dakin, Steen Hannestad, and Thomas Tram. PROSPECT: a profile likelihood code for frequentist cosmological parameter inference. MNRAS, 535(4):3686–3699, December 2024. arXiv:2312.02972, [DOI], [ADS].arxiv.org/abs/2312.02972
  44. [44]Andreas Nygaard, Emil Brinch Holm, Steen Hannestad, and Thomas Tram. Fast and effortless computation of profile likelihoods using CONNECT. J. Cosmology Astropart. Phys., 2023(11):064, November 2023. arXiv:2308.06379, [DOI], [ADS].DOI
  45. [45]Vivian Miranda and the CoCoA team. CoCoA: the Cobaya–CosmoLike joint architecture. https://github.com/CosmoLike/cocoa, 2026. Version 4.06, commit 76c08b4.
  46. [46]C. Sánchez, J. Prat, et al. Dark Energy Survey Year 3 Results: Exploiting small-scale information with lensing ratios. Phys. Rev. D, 105(8):083529, April 2022. arXiv:2105.13542, [DOI].DOI
  47. [47]S. S. Wilks. The large-sample distribution of the likelihood ratio for testing composite hypotheses. Ann. Math. Stat., 9:60, 1938. [DOI].DOI
  48. [48]J. Neyman. Outline of a theory of statistical estimation based on the classical theory of probability. Phil. Trans. R. Soc. Lond. A, 236:333, 1937. [DOI].
  49. [49]N. Hansen and A. Ostermeier. Completely derandomized self-adaptation in evolution strategies. Evol. Comput., 9(2):159–195, 2001. [DOI].pubmed.ncbi.nlm.nih.gov/11382355
  50. [50]N. Hansen. The CMA evolution strategy: A tutorial, 2016. arXiv:1604.00772.arxiv.org/abs/1604.00772
  51. [51]N. Hansen, Y. Akimoto, and P. Baudis. CMA-ES/pycma on Github. Zenodo, 2019. [DOI].
  52. [52]E. Krause, X. Fang, S. Pandey, L. F. Secco, and DES Collaboration. Dark Energy Survey Year 3 Results: Multi-Probe Modeling Strategy and Validation. arXiv e-prints, page arXiv:2105.13548, May 2021. arXiv:2105.13548, [DOI], [ADS].arxiv.org/abs/2105.13548
  53. [53]Elisabeth Krause and Tim Eifler. CosmoLike – cosmological likelihood analyses for photometric galaxy surveys. MNRAS, 470(2):2100–2112, September 2017. arXiv:1601.05779, [DOI], [ADS].DOI
  54. [54]X. Fang, T. Eifler, and E. Krause. 2D-FFTLog: efficient computation of real-space covariance matrices for galaxy clustering and weak lensing. Mon. Not. R. Astron. Soc., 497:2699, 2020. arXiv:2004.04833.DOI
  55. [55]O. Friedrich et al. Dark Energy Survey year 3 results: covariance modelling and its impact on parameter estimation and quality of fit. MNRAS, 508(3):3125–3165, December 2021. arXiv:2012.08568, [DOI], [ADS].DOI
  56. [56]Antony Lewis, Anthony Challinor, and Anthony Lasenby. Efficient Computation of Cosmic Microwave Background Anisotropies in Closed Friedmann-Robertson-Walker Models. ApJ, 538(2):473–476, August 2000. arXiv:astro-ph/9911177, [DOI], [ADS].
  57. [57]Alessio Spurio Mancini, Davide Piras, Justin Alsing, Benjamin Joachimi, and Michael P. Hobson. CosmoPower: emulating cosmological power spectra for accelerated Bayesian inference from next-generation surveys. MNRAS, 511(2):1771–1788, April 2022. arXiv:2106.03846, [DOI], [ADS].DOI
  58. [58]Christopher M. Hirata and Uroš Seljak. Intrinsic alignment-lensing interference as a contaminant of cosmic shear. Phys. Rev. D, 70(6):063526, September 2004. arXiv:astro-ph/0406275, [DOI], [ADS].arxiv.org/abs/astro-ph/0406275
  59. [59]Sarah Bridle and Lindsay King. Dark energy constraints from cosmic shear power spectra: impact of intrinsic alignments on photometric redshift requirements. New Journal of Physics, 9(12):444, December 2007. arXiv:0705.0166, [DOI], [ADS].DOI
  60. [60]Jonathan A. Blazek, Niall MacCrann, M. A. Troxel, and Xiao Fang. Beyond linear galaxy alignments. Phys. Rev. D, 100(10):105306, November 2019. arXiv:1708.09247, [DOI], [ADS].DOI
  61. [61]Joseph E. McEwen, Xiao Fang, Christopher M. Hirata, and Jonathan A. Blazek. FAST-PT: a novel algorithm to calculate convolution integrals in cosmological perturbation theory. J. Cosmology Astropart. Phys., 2016(9):015, September 2016. arXiv:1603.04826, [DOI], [ADS].DOI
  62. [62]Xiao Fang, Jonathan A. Blazek, Joseph E. McEwen, and Christopher M. Hirata. FAST-PT II: an algorithm to calculate convolution integrals of general tensor quantities in cosmological perturbation theory. J. Cosmology Astropart. Phys., 2017(2):030, February 2017. arXiv:1609.05978, [DOI], [ADS].DOI
  63. [63]J. Myles, A. Alarcon, A. Amon, C. Sánchez, and DES Collaboration. Dark Energy Survey Year 3 results: redshift calibration of the weak lensing source galaxies. MNRAS, 505(3):4249–4277, August 2021. arXiv:2012.08566, [DOI], [ADS].DOI
  64. [64]N. MacCrann, M. R. Becker, J. McCullough, A. Amon, and DES Collaboration. Dark Energy Survey Y3 results: blending shear and redshift biases in image simulations. MNRAS, 509(3):3371–3394, January 2022. arXiv:2012.08567, [DOI], [ADS].DOI
  65. [65]M. Gatti, G. Giannini, et al. Dark Energy Survey Year 3 Results: clustering redshifts - calibration of the weak lensing source redshift distributions with redMaGiC and BOSS/eBOSS. MNRAS, 510(1):1223–1247, February 2022. arXiv:2012.08569, [DOI], [ADS].DOI
  66. [66]M. Gatti, E. Sheldon, et al. Dark energy survey year 3 results: weak lensing shape catalogue. MNRAS, 504(3):4312–4336, July 2021. arXiv:2011.03408, [DOI], [ADS].DOI
  67. [67]G. Cowan, K. Cranmer, E. Gross, and O. Vitells. Asymptotic formulae for likelihood-based tests of new physics. Eur. Phys. J. C, 71:1554, 2011. arXiv:1007.1727, [DOI].DOI
  68. [68]Kyle Cranmer. Practical Statistics for the LHC, 2015. arXiv:1503.07622.
  69. [69]J. Elvin-Poole, N. MacCrann, et al. Dark Energy Survey Year 3 results: magnification modelling and impact on cosmological constraints from galaxy clustering and galaxy-galaxy lensing. MNRAS, 523(3):3649–3670, August 2023. arXiv:2209.09782, [DOI], [ADS].
  70. [70]A. Porredon et al. Dark Energy Survey Year 3 results: Cosmological constraints from galaxy clustering and galaxy-galaxy lensing using the MagLim lens sample. Phys. Rev. D, 106(10):105300, November 2022. arXiv:2105.13546, [DOI], [ADS].DOI
  71. [71]R. Cawthon et al. Dark Energy Survey Year 3 results: calibration of lens sample redshift distributions using clustering redshifts with BOSS/eBOSS. MNRAS, 513(4):5517–5539, July 2022. arXiv:2012.12826, [DOI], [ADS].DOI
  72. [72]Paolo Catelan, Marc Kamionkowski, and Roger D. Blandford. Intrinsic and extrinsic galaxy alignment. MNRAS, 320(1):L7–L13, January 2001. arXiv:astro-ph/0005470, [DOI], [ADS].DOI
  73. [73]N. Chisari, C. Codis, C. Laigle, Y. Dubois, C. Pichon, J. Devriendt, A. Slyz, L. Miller, R. Gavazzi, and K. Ben-Abd. Intrinsic alignments of galaxies in the Horizon-AGN cosmological hydrodynamical simulation. MNRAS, 454(3):2736–2753, December 2015. arXiv:1507.07843, [DOI], [ADS].DOI
  74. [74]B. Joachimi, R. Mandelbaum, F. B. Abdalla, and S. L. Bridle. Constraints on intrinsic alignment contamination of weak lensing surveys using the MegaZ-LRG sample. A&A, 527:A26, March 2011. arXiv:1008.3491, [DOI], [ADS].DOI
  75. [75]Sukhdeep Singh, Rachel Mandelbaum, and Surhud More. Intrinsic alignments of SDSS-III BOSS LOWZ sample galaxies. MNRAS, 450(2):2195–2216, June 2015. arXiv:1411.1755, [DOI], [ADS].DOI
  76. [76]Harry Johnston, Christos Georgiou, Benjamin Joachimi, Henk Hoekstra, Nora Elisa Chisari, Daniel Farrow, Maria Cristina Fortuna, Catherine Heymans, Shahab Joudaki, Konrad Kuijken, and Angus Wright. KiDS+GAMA: Intrinsic alignment model constraints for current and future weak lensing cosmology. A&A, 624:A30, April 2019. arXiv:1811.09598, [DOI], [ADS].DOI
  77. [77]S. Samuroff, R. Mandelbaum, J. Blazek, A. Campos, N. MacCrann, G. Zacharegkas, A. Amon, J. Prat, S. Singh, J. Elvin-Poole, A. J. Ross, A. Alarcon, E. Baxter, K. Bechtol, M. R. Becker, G. M. Bernstein, A. Carnero Rosell, M. Carrasco Kind, R. Cawthon, C. Chang, R. Chen, A. Choi, M. Crocce, C. Davis, J. DeRose, S. Dodleson, C. Doux, A. Drlica-Wagner, K. Eckert, S. Everett, A. Ferté, M. Gatti, G. Giannini, D. Gruen, R. A. Gruendl, I. Harrison, K. Herner, E. M. Huff, M. Jarvis, N. Kuropatkin, P.-F. Leget, P. Lemos, J. McCullough, J. Myles, A. Navarro-Alsina, S. Pandey, A. Porredon, M. Raveri, M. Rodriguez-Monroy, R. P. Rollins, A. Roodman, G. Rossi, E. S. Rykoff, C. Sánchez, L. F. Secco, I. Sevilla-Noarbe, E. Sheldon, T. Shin, M. A. Troxel, I. Tutsauus, N. Weaverdyck, B. Yanny, B. Yin, Y. Zhang, J. Zuntz, M. Aguena, O. Alves, J. Annis, D. Bacon, E. Bertin, S. Bocquet, D. Brooks, D. L. Burke, J. Carretero, M. Costanzi, L. N. da Costa, M. E. S. Pereira, J. De Vicente, S. Desai, H. T. Diehl, J. P. Dietrich, P. Doel, I. Ferrero, B. Flaugher, J. Frieman, J. García-Bellido, S. R. Hinton, D. L. Hollowood, K. Honscheid, D. J. James, K. Kuehn, O. Lahav, J. L. Marshall, P. Melchior, J. Mena-Fernández, F. Menanteau, R. Miquel, J. Newman, A. Palmese, A. Pierres, A. A. Plazas Malagón, E. Sanchez, V. Scarpine, M. Smith, E. Suchyta, M. E. C. Swanson, G. Tarle, C. To, and DES Collaboration. The Dark Energy Survey Year 3 and eBOSS: constraining galaxy intrinsic alignments across luminosity and colour space. MNRAS, 524(2):2195–2223, September 2023. arXiv:2212.11319, [DOI], [ADS].
  78. [78]Nora Elisa Chisari. A rising tide: intrinsic alignments since the turn of the millennium. A&A Rev., 33(1):5, October 2025. arXiv:2510.15738, [DOI], [ADS].
  79. [79]Nickolas Kokron, Mikhail M. Ivanov, and James M. Sullivan. Field-level modeling of halo and galaxy alignments. arXiv e-prints, page arXiv:2610.36027, October 2026. arXiv:2610.36027.arxiv.org/abs/2610.36027
  80. [80]Rachel Mandelbaum, Chris Blake, Sarah Bridle, Filipe B. Abdalla, Sarah Brough, Matthew Colless, Warrick Couch, Scott Croom, Tamara Davis, Michael J. Drinkwater, Karl Forster, Karl Glazebrook, Ben Jelliffe, Russell J. Jurek, I.-Hui Li, Barry Madore, Chris Martin, Kevin Pimbblet, Gregory B. Poole, Michael Pracy, Rob Sharp, Emily Wisnioski, David Woods, and Ted Wyder. The WiggleZ Dark Energy Survey: direct constraints on blue galaxy intrinsic alignments at intermediate redshifts. MNRAS, 410(2):844–859, January 2011. arXiv:0911.5347, [DOI], [ADS].DOI
  81. [81]Christos Georgiou, Nora Elisa Chisari, Maciej Bilicki, Francesco La Barbera, Nicola R. Napolitano, Nivya Roy, and Crescenzo Tortora. Intrinsic galaxy alignments in the KiDS-1000 bright sample: Dependence on colour, luminosity, morphology, and galaxy scale. A&A, 699:A252, July 2025. arXiv:2502.09452, [DOI], [ADS].DOI
  82. [82]J. Siegel, J. McCullough, A. Amon, C. Lammman, N. Jeffrey, B. Joachimi, H. Hoekstra, S. Heydenreich, A. J. Ross, J. Aguilar, S. Ahlen, D. Bianchi, C. Blake, D. Brooks, F. J. Castander, T. Claybaugh, A. de la Macorra, J. DeRose, P. Doel, N. Emas, S. Ferraro, A. Font-Ribera, J. E. Forero-Romero, E. Gaztañaga, S. Gontcho A. Gontcho, G. Gutierrez, K. Honscheid, M. Ishak, S. Joudaki, R. Kehoe, D. Kirkby, T. Kisner, A. Krolewski, O. Lahav, A. Lambert, M. Landriau, L. Le Guillou, M. E. Levi, M. Manera, A. Meisner, R. Miquel, J. Moustakas, S. Nadathur, J. A. Newman, G. Niz, N. Palanque-Delabrouille, W. J. Percival, A. Porredon, F. Prada, I. Pérez-Ráfols, G. Rossi, E. Sanchez, C. Saulder, D. Schlegel, M. Schubnell, A. Semenaite, J. Silber, D. Sprayberry, Z. Sun, G. Tarle, B. A. Weaver, R. Zhou, and H. Zou. Intrinsic alignment demographics for next-generation lensing: revealing galaxy property trends with DESI Y1 direct measurements. MNRAS, 552(2):stag1691, October 2026. arXiv:2507.11530, [DOI], [ADS].DOI
  83. [83]Tim Elfer, Melanie Simet, Elisabeth Krause, Christopher Hirata, Hung-Jin Huang, Xiao Fang, Vivian Miranda, Rachel Mandelbaum, Cyrille Doux, Chen Heinrich, Eric Huff, Hironao Miyatake, Shoubaneh Hemmati, Jiachuan Xu, Paul Rogozenski, Peter Capak, Ami Choi, Olivier Doré, Bhuvnesh Jain, Mike Jarvis, Jeffrey Kruk, Niall MacCrann, Dan Masters, Eduardo Rozo, David N. Spergel, Michael Troxel, Anja von der Linden, Yun Wang, David H. Weinberg, Lukas Wenzl, and Hao-Yi Wu. Cosmology with the Roman Space Telescope: synergies with the Rubin Observatory Legacy Survey of Space and Time. MNRAS, 507(1):1514–1527, October 2021. arXiv:2004.04702, [DOI], [ADS].DOI

Paper details

vlnc:W_uhd6x9pv
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 4

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Claude, Gemini (multiple versions)

View statement
Looking for related papers…
Contents