Self-Calibration of Intrinsic Alignments

• Self-calibration of intrinsic alignments is a data-driven technique that distinguishes gravitational lensing from intrinsic alignment signals by exploiting differences in redshift and geometry.
• It employs galaxy shape–density cross-correlations and selective photometric binning to algebraically extract IA contributions without relying on external IA models.
• This method mitigates IA-induced biases in cosmic shear analyses, thereby enhancing cosmological parameter estimates for surveys such as LSST, Euclid, DES, and KiDS.

Self‐calibration of intrinsic alignments (IA) refers to a set of internally data-driven methodologies for isolating, modeling, and mitigating the astrophysical systematics from IA in weak lensing (cosmic shear) surveys. These techniques exploit the different statistical, geometric, and redshift separation properties of the IA and gravitational lensing signals, providing a way to measure and subtract (or jointly fit) IA contributions directly from observed data rather than relying on external modeling assumptions. Self-calibration strategies have been developed and implemented at both field level and in two- and three-point correlation function analyses in photometric redshift surveys. The importance of self-calibration lies in its ability to suppress or account for IA-induced parameter biases that would otherwise compromise the cosmological constraining power of current and next-generation weak lensing surveys such as LSST, Euclid, DES, and KiDS.

1. Physical Origin and Impact of Intrinsic Alignments

IA arises from the tendency of galaxies to form in and align with local large-scale tidal fields, leading to non-lensing-induced correlations in galaxy ellipticities. In the context of cosmic shear, the primary IA contaminants are the II (intrinsic-intrinsic) term, correlating physically close, similarly aligned galaxies, and the GI (gravitational-intrinsic) term, correlating the intrinsic shape of a foreground galaxy with the lensing shear experienced by a background galaxy. These terms are significant because they share—at least partially—the same scale dependence and redshift structure as the gravitational lensing signal, and if left uncorrected, can bias amplitude (e.g., σ₈) and structure growth inferences by several standard deviations (Yao et al., 2017, Yao et al., 2018). The redshift and geometric dependence of the alignment differs from that of shear: lensing is sensitive to source-lens geometry and only present when the source is behind mass in the line of sight, while IA effects are local and insensitive to redshift ordering. Self-calibration exploits this fact, among others, for separation.

2. Principles and Mathematical Structure of Self-Calibration

Self-calibration leverages the distinct dependence of II, GI, and GG signals on photometric redshift separation (Δz^P), binning, and geometric selection. In its most widely implemented form for two-point statistics, self-calibration proceeds as follows:

• Galaxy shape–density cross-correlation: In each tomographic bin, the observed cross-correlation C^{γg}_{ii}(ℓ) contains both lensing-induced (Gg) and IA-induced (Ig) contributions:

$C^{γg}_{ii}(ℓ) = C^{Gg}_{ii}(ℓ) + C^{Ig}_{ii}(ℓ)$

• Selection trick using photometric redshifts: By restricting pairs to those with z^P_γ < z^P_g or the reverse, one samples configurations in which lensing is suppressed (as the lensing signal depends on the source being behind the lens), but IA persists due to its locality. The key geometric ratio is

$$Q_i(ℓ) = \frac{C^{Gg}_{ii}|_S(ℓ)}{C^{Gg}_{ii}(ℓ)}, \quad 0 < Q_i < 1$$

• Closed-form separation: Two measurements (with and without selection) yield two equations for two unknowns, permitting the algebraic extraction of C^{Ig}_{ii} and C^{Gg}_{ii} without any IA model assumption (Yao et al., 2019, Pedersen et al., 2019):

$$C^{Ig}_{ii}(ℓ) = \frac{C^{γg}_{ii}|_S(ℓ) - Q_i(ℓ) C^{γg}_{ii}(ℓ)}{1 - Q_i(ℓ)}, \quad C^{Gg}_{ii}(ℓ) = \frac{C^{γg}_{ii}(ℓ) - C^{γg}_{ii}|_S(ℓ)}{1 - Q_i(ℓ)}$$

• Scaling relation for GI: The cross-bin gravitational shear–intrinsic alignment (GI) term is related to the self-calibrated Ig measurement through an analytic kernel that depends solely on redshift distributions, lensing geometry, and galaxy bias:

$$C^{IG}_{ij}(ℓ) \simeq \frac{W_{ij}\, Δ_i}{b_i} C^{Ig}_{ii}(ℓ)$$

where W_{ij} encapsulates the lensing geometry, Δ_i the bin width, and b_i is the galaxy bias in bin i (Yao et al., 2017, Pedersen et al., 2019, Zhang, 2010).

• Higher-order (3-point) generalization: Similar approaches are applied to the bispectrum, where the line-of-sight dependence allows the separation of B^{GGI}, B^{GII}, and B^{III} from the lensing bispectrum by measuring the redshift-separation response in fine photometric bins (Troxel et al., 2012, Troxel et al., 2011).

A key advantage is the model independence: all scaling relations derive from geometry and statistics rather than an external IA model, although practical application relies on accurate photo-z distributions and knowledge of galaxy bias.

3. Field-Level and Bayesian Hierarchical Extensions

Recent advances generalize self-calibration to field-level inference. In "Field-level inference of cosmic shear with intrinsic alignments and baryons," IA parameters are inferred as part of a high-dimensional Bayesian hierarchical model incorporating the initial density field, cosmological parameters, baryon feedback, and a TATT (tidal alignment and tidal torquing) IA model (Porqueres et al., 2023). The approach jointly samples:

• The initial density contrast δ^{ic},
• Cosmological parameters θ = {Ω_m, σ_8},
• IA parameters χ = {A_1, b_{TA}, A_2},
• Baryonic parameters ξ = {β, γ},

with a field-level likelihood for the observed ellipticity maps. The Bayesian framework enables self-calibration by effectively "learning" the IA parameters from the data distributions themselves. Posterior distributions of key IA parameters (A_1, A_2) are strongly data-driven (narrower than priors), showing that these parameters are tightly constrained by the lensing data alone. End-to-end validation with synthetic data confirms that the true IA values are recovered within the 68% credible intervals, illustrating that self-calibration operates successfully in a forward-modeling, hierarchical Bayesian context (Porqueres et al., 2023).

4. Applications, Performance, and Empirical Results

Self-calibration methods have been implemented on several current weak lensing data sets, with robust measurement and mitigation of IA. Key results include:

• In KiDS-450 and KV450, SC robustly separated the lensing and IA contributions in shape–density correlations, with significance levels of 3.5σ–3.7σ for Ig and GI correlations, and measured IA amplitudes in agreement with best-fit cosmologies (Yao et al., 2019, Pedersen et al., 2019).
• Application to the DES Y1 sample (via the LSST-DESC pipeline) produced marginal but consistent IA detections in the mid- to high-z bins, with non-detections at low redshift primarily due to photo-z limitations (Pedersen et al., 15 Jan 2026).
• The DECaLS DR3 analysis yielded a ∼14σ detection of total IA, revealing strong dependence on galaxy color (red galaxies show high IA amplitude, blue galaxies no significant IA), and a clear redshift evolution of A_{IA} (Yao et al., 2020).
• In all these cases, SC was less sensitive to cosmological assumptions and rivaled template-marginalization ("marginalize over IA model") in terms of precision loss, but provided direct empirical access to IA systematics.
• Performance forecasts for future surveys (LSST, Euclid, WFIRST) show that SC reduces residual GI and II biases by up to an order of magnitude, inflating parameter contours by ≲2× compared to no mitigation. With photo-z uncertainties marginalized, the resultant biases on Ω_m, σ_8, w_0 are reduced from several σ to <1σ (Yao et al., 2017, Yao et al., 2018).

Suppression factors for GI and GGI contaminations are generally ≥10× except for auto- and adjacent-bin spectra, for which they remain at ≥3× (Troxel et al., 2011, Yao et al., 2018).

5. Limitations, Systematic Uncertainties, and Survey Requirements

Self-calibration efficacy is contingent upon:

• Photo-z accuracy: High-precision, object-level photo-z PDF knowledge is critical, as the method exploits redshift ordering. Systematic photo-z biases or catastrophic outliers degrade estimator performance (e.g., DES Y1 results are limited by photo-z uncertainty) (Pedersen et al., 15 Jan 2026, Yao et al., 2017).
• Galaxy bias knowledge: Uncertainties in the galaxy bias b_i propagate directly into IA subtraction accuracy. Joint or independent clustering analyses to constrain b_i are recommended (Pedersen et al., 2019).
• Redshift binning: For two-point SC, bin widths Δz^P ≲ 0.2 are optimal; for three-point SC, fine sub-binning (Δz^P ≈ 0.01) within a thick bin is necessary for bispectrum decomposition (Troxel et al., 2012, Zhang, 2010).
• Survey area and density: Sufficient area (f_sky ≳0.4) and source density (n_g ≳ 20 arcmin⁻²) are needed to produce well-populated fine redshift bins and minimize shot noise.
• Modeling assumptions: While SC is largely model-independent, it relies on the Limber approximation and galaxy bias linearity. Higher-order or highly nonlinear IA effects, and small-scale baryonic feedback, may limit ultimate mitigation (Yao et al., 2018, Porqueres et al., 2023).

6. Extensions and Future Prospects

The self-calibration framework is being further extended to:

• Auto-spectra and close-bin inclusion: Newer estimators now admit II-correction and inclusion of auto- and adjacent-bin cross-spectra, preserving the full constraining power of cosmic shear (Yao et al., 2018).
• Field-level and likelihood-based inference: The use of full-field Bayesian inference enables joint, model-agnostic posterior estimation of IA, baryonic, and cosmological parameters directly from the data. This is expected to be the standard for Stage IV surveys (Porqueres et al., 2023).
• Higher-order statistics: 3-point and multi-point generalizations using redshift-separation tomography further enhance IA detection and removal at the bispectrum level (Troxel et al., 2011, Troxel et al., 2012).
• Data-driven IA physics: Extracted empirical IA signals serve as direct input to the physical modeling of galaxy–halo–tidal field connections, feeding back into improved theoretical templates.

Continued validation on simulated data and diverse survey modalities—including CMB lensing cross-correlations—are anticipated. For LSST, Euclid, and other deep imaging surveys, self-calibration is projected to be integral to the cosmological exploitation of cosmic shear, potentially reducing IA-driven systematic uncertainties below the survey’s statistical floor.

7. Comparative Framework and Survey Recommendations

Self-calibration is complementary to, and in some scenarios competitive with, IA template-marginalization approaches:

• Model independence: SC does not require a specific IA functional prescription, enabling robust, non-parametric measurement and subtraction of IA.
• Data usage: SC uniquely exploits the full information in shape–density (and related) cross-correlations, rather than discarding contaminated bins or relying solely on lensing-only spectra.
• Synergies: Joint application—measuring SC-based IA residuals, fitting them with improved physical models, and using the results to inform or cross-check template marginalization—enhances both systematics control and physical insight into galaxy formation.
• Implementation guidance: For upcoming surveys, high-fidelity per-galaxy photo-z PDFs, robust source-density characterizations, controlled galaxy-bias measurements, and flexible tomographic binning infrastructure are critical prerequisites for maximal SC efficacy (Pedersen et al., 15 Jan 2026, Yao et al., 2017).

Ongoing work focuses on refining estimator robustness to real-world systematics, extending the approach to broader weak lensing observables, and integrating SC methodologies within standard cosmological parameter inference pipelines.