SLICS: LightCone Simulations for Weak Lensing

• Scinet LIghtCone Simulations (SLICS) are a set of dark-matter-only N-body simulations that create weak lensing and combined-probe mock data for large-scale structure analyses.
• The simulations use a past-light-cone framework with 18 mass sheets and dense pixel grids to generate cosmic shear, CMB-lensing maps, and survey-specific mock catalogues mimicking KiDS, LSST, and others.
• SLICS supports covariance estimation and systematic effect studies while acknowledging limitations like fixed cosmology, finite resolution, and broken three-dimensional correlations.

Scinet LIghtCone Simulations (SLICS) are a public suite of weak lensing mock data built from dark-matter-only $N$-body simulations and designed for covariance estimation, pipeline validation, and systematic-effect studies in large-scale-structure analyses. The suite was extended to support combined-probe analyses by providing KiDS-450- and LSST-like source catalogues, cosmic microwave background lensing maps, and spectroscopic survey mocks emulating GAMA, BOSS, and 2dFLenS, with 844 independent realizations containing complete galaxy and halo catalogues and 932 realizations for cosmic shear and CMB lensing products (Harnois-Deraps et al., 2018). Its scope is explicitly multi-probe: cosmic shear, galaxy-galaxy lensing, galaxy clustering, CMB-lensing cross-correlations, and downstream mock-based studies of selection effects and covariance structure are all supported within the same light-cone framework.

1. Numerical basis and fiducial cosmology

The fiducial SLICS cosmology is based on the WMAP9 + BAO + SN best fit of Hinshaw et al. 2013, with $\Omega_{\rm m} = 0.2905$, $\Omega_\Lambda = 0.7095$, $\Omega_{\rm b} = 0.0473$, $h = 0.6898$, $\sigma_8 = 0.826$, and $n_{\rm s} = 0.969$ (Harnois-Deraps et al., 2018). Each simulation uses a comoving box of side length $L_{\rm box} = 505\,h^{-1}\mathrm{Mpc}$ with $1536^3$ particles, giving a particle mass of $2.88\times10^9\,h^{-1}M_\odot$ and a minimum resolved halo mass of $\Omega_{\rm m} = 0.2905$0 for 20 particles per halo.

The numerical resolution is characterized in Fourier space. The three-dimensional power spectrum $\Omega_{\rm m} = 0.2905$1 agrees within $\Omega_{\rm m} = 0.2905$2 of the Cosmic Emulator for $\Omega_{\rm m} = 0.2905$3, while at $\Omega_{\rm m} = 0.2905$4–$\Omega_{\rm m} = 0.2905$5 the simulated $\Omega_{\rm m} = 0.2905$6 is $\Omega_{\rm m} = 0.2905$7–$\Omega_{\rm m} = 0.2905$8 lower because of mass-resolution limits. The suite also includes higher-resolution runs, denoted SLICS-HR, and runs with slightly different cosmological parameters for sensitivity tests.

Property	Value
Box size	$\Omega_{\rm m} = 0.2905$9
Particles per box	$\Omega_\Lambda = 0.7095$0
Particle mass	$\Omega_\Lambda = 0.7095$1
Realizations	932 (cosmic shear + CMB), 844 (all mocks)
Area per light cone	$\Omega_\Lambda = 0.7095$2

These specifications define the balance that SLICS adopts between ensemble size and per-realization fidelity. The large number of independent realizations is central to its covariance role, whereas the quoted small-scale suppression in $\Omega_\Lambda = 0.7095$3 fixes the scale range over which direct signal prediction is most reliable.

2. Light-cone architecture and lensing fields

SLICS constructs past-light-cone realizations by dividing the volume between the observer and $\Omega_\Lambda = 0.7095$4 into 18 slices, or “mass sheets,” each corresponding to a half-box of $\Omega_\Lambda = 0.7095$5 (Harnois-Deraps et al., 2018). At each lens redshift $\Omega_\Lambda = 0.7095$6, particles are projected onto a $\Omega_\Lambda = 0.7095$7 pixel grid using Cloud-In-Cell interpolation to form the lens plane. At high redshift, periodic boundary conditions fill the field of view, with random shifts and alternating projection axes used to minimize repeated structure.

Each light cone covers $\Omega_\Lambda = 0.7095$8 and is sampled on a $\Omega_\Lambda = 0.7095$9 pixel grid. Lens planes are placed at 18 discrete redshifts from $\Omega_{\rm b} = 0.0473$0 to $\Omega_{\rm b} = 0.0473$1. Convergence and shear maps are then computed in the Born approximation through weighted sums over the projected density planes. For a discrete set of lens planes and sources, the convergence is written as

$\Omega_{\rm b} = 0.0473$2

CMB-lensing products are generated by extending each light cone to $\Omega_{\rm b} = 0.0473$3 through the addition of linear-theory mass sheets over $\Omega_{\rm b} = 0.0473$4–$\Omega_{\rm b} = 0.0473$5. Both noise-free and Planck-like $\Omega_{\rm b} = 0.0473$6 maps are provided, and Planck-like noise is added following the Planck data release, enabling mock cross-correlation analyses with real Planck CMB-lensing data.

The resulting data model is explicitly two-dimensional and lens-plane based. One consequence, noted in the published limitations, is that direct three-dimensional correlations across mass planes are broken (Harnois-Deraps et al., 2018).

3. Mock catalogues for lensing and galaxy surveys

The public data products include source galaxy catalogues, spectroscopic lens catalogues, halo catalogues, mass and shear planes, CMB-lensing maps, and random catalogues (Harnois-Deraps et al., 2018). The source catalogues are designed to emulate specific surveys. For KiDS-450-like mocks, galaxies are placed randomly in the light cone with $\Omega_{\rm b} = 0.0473$7 and an $\Omega_{\rm b} = 0.0473$8 matching the KiDS-450 data using the DIR method for redshift calibration. Photometric redshifts $\Omega_{\rm b} = 0.0473$9 are assigned by sampling the matched spectroscopic–photo-$h = 0.6898$0 joint PDF, and random shape noise is drawn per ellipticity component from a truncated Gaussian with $h = 0.6898$1. Tomographic bins follow the $h = 0.6898$2 ranges used in KiDS-450 analyses.

The LSST-like source catalogues use the same general methodology but adopt $h = 0.6898$3, a forecasted redshift distribution, and 10 tomographic bins (Harnois-Deraps et al., 2018). These mocks are therefore deeper and denser, which is directly relevant for studies of blending and neighbour-exclusion systematics.

Spectroscopic lens catalogues are generated through Halo Occupation Distribution prescriptions. For CMASS and LOWZ, intended to emulate BOSS and 2dFLenS samples, central and satellite occupation numbers are assigned using best-fit HOD models, and satellites follow an NFW profile with concentration taken from Bullock et al. or Macciò et al. Redshift selection and downsampling are used to match observed $h = 0.6898$4 and number densities, and corresponding random catalogues are produced for clustering estimators. For GAMA, the mock construction uses a conditional luminosity function HOD in which centrals and satellites are assigned luminosities to match the observed stellar-mass and magnitude distributions as closely as possible.

Hybrid catalogues extend these survey-specific constructions. KiDS-HOD extends the GAMA HOD to $h = 0.6898$5, while LSST-HOD extends it to $h = 0.6898$6 and to fainter magnitudes, for use in cross-probe systematics studies and selection-bias calibration (Harnois-Deraps et al., 2018). Random catalogues are constructed to reproduce the three-dimensional step-function nature of the SLICS light-cone geometry together with the target survey $h = 0.6898$7.

4. Covariance estimation and combined-probe analyses

A central application of SLICS is empirical covariance estimation for weak-lensing and clustering observables. For a data vector $h = 0.6898$8, the covariance is estimated from the ensemble as

$h = 0.6898$9

with $\sigma_8 = 0.826$0 typically 844 for the full galaxy-and-halo mock set (Harnois-Deraps et al., 2018).

The suite was optimized for combined-probe covariance estimation and was used to demonstrate joint covariance matrices for cosmic shear $\sigma_8 = 0.826$1, galaxy clustering $\sigma_8 = 0.826$2, and galaxy-galaxy lensing $\sigma_8 = 0.826$3, thereby supporting “3×2-point” likelihood analyses such as KiDS+BOSS and KiDS+GAMA (Harnois-Deraps et al., 2018). Because the observables are measured on matched realizations, shape noise, sample variance, and probe-to-probe cross-covariances are included naturally.

The published caveat is that SLICS covariances are evaluated at a fixed cosmology and underestimate super-sample covariance because of the finite simulation box; larger simulations or analytic corrections are therefore recommended for complete analyses (Harnois-Deraps et al., 2018). In comparison with jackknife estimates from data, SLICS-based errors are reported to be consistent on small scales and more reliable on large scales, where jackknife methods underestimate the variance.

Subsequent work used the mock ensemble in more specialized estimators. In the study of galaxy troughs and ridges with KiDS, SLICS was employed both for covariance estimation and for deriving optimal signal-to-noise weights for stacked lensing profiles, using 349 independent realizations. The resulting weighted trough detections reached $\sigma_8 = 0.826$4, $\sigma_8 = 0.826$5, $\sigma_8 = 0.826$6, and $\sigma_8 = 0.826$7 for aperture radii $\sigma_8 = 0.826$8 arcmin, and the optimal weighting increased the signal-to-noise by about $\sigma_8 = 0.826$9 for troughs and about $n_{\rm s} = 0.969$0 for ridges relative to a hard-cut percentile approach (Brouwer et al., 2018). In that same analysis, the analytic covariance agreed well with the SLICS covariance up to moderate separations, $n_{\rm s} = 0.969$1 arcmin.

5. Systematic effects: neighbour-exclusion bias

SLICS was also used to quantify neighbour-exclusion bias in tomographic cosmic shear measurements (Harnois-Deraps et al., 2018). The origin of this bias is straightforward: when galaxies with close projected neighbours are rejected because of blending or measurement failures, overdense lines of sight lose more sources, which biases shear two-point statistics low.

In the mock implementation, close-pair sources are identified with a KDTree and filtered using exclusion radii of 1, 2, 3.7, and 5 arcsec. Two removal strategies are considered: “FAINT,” which removes the faintest galaxy in a close pair, and “BOTH,” which removes both members. The effect is then quantified after correcting for the induced change in $n_{\rm s} = 0.969$2, and an empirical fit is supplied in the form

$n_{\rm s} = 0.969$3

with nuisance parameters $n_{\rm s} = 0.969$4 and $n_{\rm s} = 0.969$5 calibrated from the mocks (Harnois-Deraps et al., 2018).

For KiDS- and DES-like survey depth, with an exclusion radius of about 2 arcsec, the measured $n_{\rm s} = 0.969$6 is biased low by less than one percent on the angular scales usually retained in standard analyses, namely $n_{\rm s} = 0.969$7 arcmin for KiDS and $n_{\rm s} = 0.969$8 arcmin for DES. For $n_{\rm s} = 0.969$9, the bias can reach a few percent on the smallest scales probed. In LSST-like data, the neighbour-exclusion bias roughly doubles, affecting $L_{\rm box} = 505\,h^{-1}\mathrm{Mpc}$0 by up to $L_{\rm box} = 505\,h^{-1}\mathrm{Mpc}$1–$L_{\rm box} = 505\,h^{-1}\mathrm{Mpc}$2 for typical selection strategies. The effect is stronger in higher-redshift and denser tomographic bins.

The effective exclusion scale is pipeline dependent. DES SVA1 data and KiDS image simulations were used to infer the relevant exclusion radius for specific shape-measurement pipelines, namely lensfit for KiDS and SExtractor flagging for DES ngmix, indicating that the practical bias depends on the blend definition and thresholding adopted in a given survey analysis (Harnois-Deraps et al., 2018).

6. Extensions, validation against other methods, and limitations

Although SLICS is dark-matter only, later work showed that it can be augmented for baryonic observables through learned mappings from matter density to gas pressure. In “Painting with baryons,” variational auto-encoders and generative adversarial networks were trained on paired matter-density and pressure slices from the BAHAMAS hydrodynamical simulation and then applied to 100 SLICS lines of sight to generate thermal Sunyaev–Zeldovich maps (Tröster et al., 2019). The generated maps were found to be statistically consistent with those from BAHAMAS, and the angular cross-power spectrum $L_{\rm box} = 505\,h^{-1}\mathrm{Mpc}$3 between weak-lensing convergence and the Compton-$L_{\rm box} = 505\,h^{-1}\mathrm{Mpc}$4 field, together with its variance, showed excellent agreement between BAHAMAS and the SLICS-based predictions. This established a route to tSZ covariance estimation without rerunning hydrodynamical simulations at SLICS scale.

SLICS has also served as a validation target for analytic covariance formalisms. A later study on the analytical cross-covariance between second- and third-order aperture-mass statistics used SLICS shear catalogues and convergence maps as the numerical benchmark (Wielders et al., 24 Sep 2025). In that framework, the cross-covariance separates into three terms governed by the power spectrum, bispectrum, and tetraspectrum, with the tetraspectrum term dominating. The analytic model qualitatively reproduced the SLICS covariance; the reported figure of merit was $L_{\rm box} = 505\,h^{-1}\mathrm{Mpc}$5 of the numerical case and rose to $L_{\rm box} = 505\,h^{-1}\mathrm{Mpc}$6 when small-scale information was excluded.

The suite’s limitations are stated explicitly. SLICS does not include neutrino mass, baryon feedback, or subhalo information; it has finite resolution and finite box size; and direct three-dimensional correlations across mass planes are broken (Harnois-Deraps et al., 2018). A common misconception is therefore to treat SLICS as a complete physical forward model of all late-time observables. It is more accurately described as a dark-matter light-cone ensemble optimized for weak-lensing and combined-probe covariance work, with documented extensions for specific baryonic applications and with known finite-volume and fixed-cosmology limitations.

All products described in the main release—mass and shear planes, CMB-lensing maps, halo catalogues, mock galaxy catalogues, and random catalogues—are made public at http://slics.roe.ac.uk/ (Harnois-Deraps et al., 2018).