SIDES: Simulated Infrared Extragalactic Sky

• SIDES is an empirically calibrated simulation framework that models the dusty extragalactic sky across far-IR to mm wavelengths by connecting galaxy evolution, clustering, and observational selection effects.
• It utilizes halo abundance matching, redshift-dependent SEDs, and beam-convolved map-making to reproduce observational biases such as source confusion in single-dish surveys.
• Recent extensions like pySIDES integrate spectral line emission and AGN contributions, offering a versatile tool for forecasting surveys and analyzing intensity and polarization confusion.

Searching arXiv for SIDES and closely related simulated infrared extragalactic sky papers. SIDES is a public, empirically calibrated simulation framework for the dusty extragalactic sky, originally introduced as the “Simulated Infrared Dusty Extragalactic Sky” and later described in updated work as the “Simulated Infrared Extragalactic Dusty Sky.” Its defining objective is to connect halo-based galaxy populations, infrared spectral energy distributions, clustering, beam-convolved maps, and source-extraction procedures in the far-infrared to millimeter regime, where confusion and blending make observed catalogues differ substantially from intrinsic source populations (Bethermin et al., 2017, Vidal et al., 18 Sep 2025).

1. Original formulation and physical construction

The canonical SIDES formulation was introduced to address a concrete observational tension: single-dish far-IR/submm surveys and high-angular-resolution interferometric follow-up do not yield the same number counts, because limited angular resolution blends several galaxies into one source. The original simulation therefore aimed to model the entire chain from galaxy evolution prescriptions to clustered sky maps and extracted catalogues. It covers 2 deg$^2$ over $0 and spans 70 $\mu$m to 1.2 mm (Bethermin et al., 2017).

Its dark-matter backbone is the Bolshoi-Planck simulation, with box size $(250\, h^{-1}\,\mathrm{Mpc})^3$, particle mass $1.5\times 10^8\, h^{-1} M_\odot$, and halos identified with Rockstar. SIDES constructs a $1.4^\circ \times 1.4^\circ$ lightcone, corresponding to a comoving volume of $0.17\,\mathrm{Gpc}^3$, and uses periodic replication with a line of sight rotated by $10^\circ$ about the $y$-axis and $10^\circ$ about the $0subhalo abundance matching using the peak circular velocity $00.2 dex in stellar mass at fixed $0Bethermin et al., 2017).

On top of that halo lightcone, SIDES applies an updated 2SFM population model. The stellar mass function is redshift dependent and parameterized as a double Schechter form. Galaxies are first split into passive and star-forming populations through a mass- and redshift-dependent star-forming fraction. Star-forming systems are then divided into main-sequence and starburst modes, with the starburst fraction set to 1.5% at $0, rising linearly to 3% at $0 and remaining constant thereafter. The SFR distribution is lognormal, centered on the evolving main sequence for normal systems and on $0 for starbursts, with 0.3 dex scatter in practice for both populations. SIDES also imposes a hard cap,

$0

motivated by interferometric follow-up showing that apparently extreme single-dish sources frequently deblend into multiple components (Bethermin et al., 2017).

Infrared luminosities are assigned through the Kennicutt conversion adjusted to a Chabrier IMF,

$0

with $0SEDs are drawn from the Magdis et al. (2012) template library and depend on star-formation mode and the mean radiation field $\mu$0. For main-sequence galaxies SIDES adopts

$\mu$1

with $\mu$2 and $\mu$3, while starbursts use $\mu$4. A 0.2 dex scatter in $\mu$5 is added. Lensing is included probabilistically, with strong lensing drawn from Hezaveh et al. and weak lensing approximated from Hilbert et al., but not tied self-consistently to the simulated foreground mass distribution (Bethermin et al., 2017).

2. Map-making, confusion, and the reinterpretation of observed counts

A distinguishing feature of SIDES is that it does not stop at intrinsic galaxy catalogues. It produces beam-convolved maps in Herschel/PACS, Herschel/SPIRE, SCUBA-2-like, AzTEC-like, and NIKA2-like bands, and then runs source extraction in a manner intended to mimic observational practice. Herschel/PACS/SPIRE catalogues are extracted with FASTPHOT using 24 $\mu$6m positions as priors, retaining only the brightest 24 $\mu$7m source within 0.5 FWHM to limit degeneracies, whereas 850 $\mu$8m and 1.2 mm sources are extracted blindly above $\mu$9 and deboosted following Geach et al. The SPIRE confusion noise measured in the simulated maps is 6.0 mJy at 250 $(250\, h^{-1}\,\mathrm{Mpc})^3$0m, 6.5 mJy at 350 $(250\, h^{-1}\,\mathrm{Mpc})^3$1m, and 6.0 mJy at 500 $(250\, h^{-1}\,\mathrm{Mpc})^3$2m, consistent within $(250\, h^{-1}\,\mathrm{Mpc})^3$3 with Nguyen et al. (Bethermin et al., 2017).

The central scientific result is that limited angular resolution fundamentally reshapes observed continuum counts. SIDES shows that at 350 and 500 $(250\, h^{-1}\,\mathrm{Mpc})^3$4m, Herschel number counts between 5 and 50 mJy are biased high by about a factor 2, and that the corresponding redshift distributions are biased toward low $(250\, h^{-1}\,\mathrm{Mpc})^3$5. At 500 $(250\, h^{-1}\,\mathrm{Mpc})^3$6m, the brightest galaxy contributes only about 58–60% of the measured flux, whereas in a NIKA2-like 1.2 mm survey the brightest galaxy contributes about 95%. This is why intrinsic counts match interferometric measurements while source-extracted counts from beam-smoothed maps match single-dish measurements: the two observing modes are probing different effective source definitions (Bethermin et al., 2017).

SIDES also quantifies how clustering propagates into standard analysis products. The relative stacking excess,

$(250\, h^{-1}\,\mathrm{Mpc})^3$7

is 13 ± 1% at 250 $(250\, h^{-1}\,\mathrm{Mpc})^3$8m, 21 ± 1% at 350 $(250\, h^{-1}\,\mathrm{Mpc})^3$9m, and 34 ± 1% at 500 $1.5\times 10^8\, h^{-1} M_\odot$0m. In the SPIRE pixel histogram $1.5\times 10^8\, h^{-1} M_\odot$1, clustering changes the histogram by less than about 5% at 250 $1.5\times 10^8\, h^{-1} M_\odot$2m and by up to 15% at 350 and 500 $1.5\times 10^8\, h^{-1} M_\odot$3m. The shot-noise level is written as

$1.5\times 10^8\, h^{-1} M_\odot$4

and SIDES shows that previously reported discrepancies between measured CIB shot noise and shot noise inferred from observed SPIRE counts can be resolved once one uses intrinsic rather than blend-boosted counts (Bethermin et al., 2017).

The same framework also reinterprets “red Herschel sources.” For the colour and flux cuts used by Asboth et al., the intrinsic clustered 2 deg$1.5\times 10^8\, h^{-1} M_\odot$5 SIDES catalogue contains no such source, while a larger 274 deg$1.5\times 10^8\, h^{-1} M_\odot$6 simplified unclustered catalogue contains only 18 intrinsically selected objects, all strongly lensed. After beam smoothing and noise, however, the simulation yields $1.5\times 10^8\, h^{-1} M_\odot$7 red detections in 2 deg$1.5\times 10^8\, h^{-1} M_\odot$8, scaling to $1.5\times 10^8\, h^{-1} M_\odot$9 over 274 deg$1.4^\circ \times 1.4^\circ$0, consistent within $1.4^\circ \times 1.4^\circ$1 with the observed 477 detections. The implication stated in the paper is that the apparent excess can be an observational artifact produced by noise, confusion, weak lensing, and steep underlying colour and flux distributions (Bethermin et al., 2017).

3. Spectroscopic SIDES and the pySIDES implementation

The 2022 CONCERTO-oriented extension transformed SIDES from a continuum-focused dusty-sky simulation into a spectroscopic sub-mm/mm framework that includes the dominant lines around 1 mm: [$1.4^\circ \times 1.4^\circ$2] 158 $1.4^\circ \times 1.4^\circ$3m, the CO rotational ladder, and [CI]. It uses the original SIDES halo lightcone and semi-empirical population model, but adds line-emission prescriptions, a spectral-cube generator, and a full public Python rewrite, pySIDES (Bethermin et al., 2022).

For [$1.4^\circ \times 1.4^\circ$4], the default model adopts the De Looze et al. relation,

$1.4^\circ \times 1.4^\circ$5

with an intrinsic scatter of 0.2 dex and no duty cycle, so every star-forming SIDES galaxy emits according to the adopted relation. An alternative Lagache et al. prescription is also implemented. For CO, the simulation derives $1.4^\circ \times 1.4^\circ$6 from $1.4^\circ \times 1.4^\circ$7 using the Sargent et al. relation, adds 0.2 dex scatter, and lets the CO SLED vary from galaxy to galaxy through the radiation field parameter $1.4^\circ \times 1.4^\circ$8. For [CI], empirical relations are calibrated from the Valentino et al. compilation, with 0.2 dex scatter for CI and 0.19 dex scatter for CI (Bethermin et al., 2022).

Mock CONCERTO-like cubes are generated over 125–305 GHz, with 1 GHz spectral channels, 5 arcsec angular pixels, and the underlying $1.4^\circ \times 1.4^\circ$9 deg$0.17\,\mathrm{Gpc}^3$0 SIDES field. The total cube is the sum of continuum and all line cubes. The released products are designed for both intensity mapping and sub-mm continuum and line surveys, and the paper is explicit that the code and products are public (Bethermin et al., 2022).

Validation is broad. The earlier SIDES continuum model was already shown to reproduce mid-IR to mm number counts, redshift distributions, Herschel map pixel histograms, and CIB power spectra. The spectroscopic extension adds validation against CO luminosity functions from COLDz and ASPECS/PdBI, [CI] luminosity functions from ASPECS, [$0.17\,\mathrm{Gpc}^3$1] luminosity-function constraints from ALPINE, and mmIME shot noise near 100 GHz. The paper states that SIDES is always within the 1-$0.17\,\mathrm{Gpc}^3$2 observational ranges for the CO luminosity functions shown, and predicts an average total CO+[CI] shot noise of

$0.17\,\mathrm{Gpc}^3$3

consistent with the mmIME measurements quoted in the paper (Bethermin et al., 2022).

The main forecasting result is that [$0.17\,\mathrm{Gpc}^3$4] intensity mapping at $0.17\,\mathrm{Gpc}^3$5 is highly uncertain and strongly foreground limited. Depending on the adopted $0.17\,\mathrm{Gpc}^3$6-SFR relation and high-$0.17\,\mathrm{Gpc}^3$7 star-formation history, the predicted $0.17\,\mathrm{Gpc}^3$8 [$0.17\,\mathrm{Gpc}^3$9] power spectrum varies by about two orders of magnitude. The continuum is the brightest component by a factor of 3 to 100 depending on frequency, and at 300 GHz the CO foreground power spectrum is higher than the [$10^\circ$0] one for the base scenario. Masking galaxies detectable in deep surveys can reduce foregrounds to about 20% of the [$10^\circ$1] power spectrum up to $10^\circ$2, but the paper states that this strategy will not be sufficient at higher redshift (Bethermin et al., 2022).

4. JWST/MIRI-era SIDES: stellar light, quiescent galaxies, and obscured AGN

A 2025 update explicitly reworks SIDES for the JWST/MIRI regime. The paper keeps the original population backbone of pySIDES—Bolshoi-Planck halos, abundance matching via $10^\circ$3, and the 2SFM division into quiescent, main-sequence, and starburst populations—but modifies the SED and AGN treatment so that the simulation can reproduce MIRI number counts, redshift distributions, and AGN fractions (Vidal et al., 18 Sep 2025).

The most important change is that main-sequence and starburst templates no longer consist only of dust emission. The updated model adds stellar emission using SWIRE templates: Sd for main-sequence galaxies and M82 for starbursts. Quiescent galaxies, which were omitted from the original far-IR SIDES because they were assumed irrelevant for far-IR counts, are now added explicitly using the Ell5 template from the SWIRE library, normalized through a redshift-dependent stellar mass-to-$10^\circ$4-band light ratio. AGN and composite systems are introduced using the Kirkpatrick et al. (2015) template library, binned by the mid-IR AGN fraction $10^\circ$5, with additional SWIRE-based stellar components stitched onto the templates at wavelengths chosen not to alter the $10^\circ$6–$10^\circ$7 normalization (Vidal et al., 18 Sep 2025).

The AGN contribution is assigned probabilistically. The total IR AGN fraction is defined as

$10^\circ$8

with the accretion-disk luminosity written as

$10^\circ$9

using $y$0, and the IR AGN luminosity parameterized as $y$1. The paper tests $y$2 and finds that $y$3 gives the best agreement with observed MIRI AGN fractions, which is then adopted as the fiducial model. The AGN fraction is capped at $y$4 (Vidal et al., 18 Sep 2025).

This revised SIDES reproduces observed integral number counts across 8 of the 9 MIRI imaging filters, with F1130W omitted because constraints were sparse. It also preserves SIDES’s earlier successes at Herschel wavelengths. The scientific picture reported in the paper is that main-sequence galaxies dominate the faint end of the MIRI counts, while AGN dominate at bright flux densities $y$5. The paper further notes that the brightest MIRI AGN at roughly $y$6–$y$7 and above are likely AGN-dominated, whereas fainter AGN-selected sources are more often composites. The abstract also states that surveys below $y$8 suffer from $y$9 uncertainty in bright AGN counts because of cosmic variance (Vidal et al., 18 Sep 2025).

5. SIDES as a survey-forecasting platform: PRIMAger and polarized confusion

A 2024 PRIMAger study uses the 2 deg$10^\circ$0 SIDES realization as the astrophysical sky model for a baseline assessment of confusion in both intensity and polarization. In this application SIDES supplies the source positions, flux densities, redshifts, stellar masses, and clustering; polarization is then added through a phenomenological extension rather than a native SIDES prediction (Béthermin et al., 2024).

The simulated PRIMAger maps include only confusion noise, with no instrumental noise, no photon noise from diffuse backgrounds, and no Galactic cirrus in the baseline case. Gaussian beams are adopted, with FWHM from 4.1 arcsec at 25 $10^\circ$1m to 27.6 arcsec at 235 $10^\circ$2m. The classical confusion variance is written as

$10^\circ$3

but the paper estimates the confusion level directly from maps by iterative clipping and then defines the detection threshold as

$10^\circ$4

For polarization, each SIDES galaxy is assigned an integrated polarization fraction drawn from a Gaussian with mean $10^\circ$5 and standard deviation $10^\circ$6, with a uniform polarization angle $10^\circ$7, and the Stokes parameters are defined by

$10^\circ$8

The maps are built in $10^\circ$9 and $0Béthermin et al., 2024).

The resulting confusion limits rise rapidly with wavelength in intensity: 0.021 mJy at 25.0 $0, 4.6 mJy at 96.3 $0, 12 mJy at 126 $0, 28 mJy at 172 $0, and 46 mJy at 235 $0. In polarization, assuming mean $00.03 mJy at 96.3 $0, 0.078 mJy at 126 $0, 0.17 mJy at 172 $0, and 0.25 mJy at 235 $0. The paper states that polarization confusion is therefore more than 100x lower than intensity confusion and that simply multiplying intensity confusion by a mean polarization fraction is not adequate (Béthermin et al., 2024).

Because SIDES contains the full underlying source population, the PRIMAger study can identify the main counterpart as the brightest galaxy within the beam half-light radius. It finds that in intensity the brightest galaxy contributes more than 91% of the measured flux in the PHI bands, but only 0.72 in PPI4, so other objects contribute about $0 at 235 $0. Clustering has a mild impact on intensity confusion, up to 25%, and negligible impact in polarization under the assumptions of random polarization angles and no intrinsic alignments. The paper also states that a basic blind extraction is sufficient to detect galaxies at the knee of the luminosity function up to $0 and $0 main-sequence galaxies up to $0. In polarization, for a conservative sensitivity, the forecast is $0 detections up to $0 (Béthermin et al., 2024).

6. Related frameworks, boundaries of the SIDES concept, and persistent limitations

Several adjacent simulated-sky frameworks are often compared with SIDES, but those comparisons are sometimes explicitly inferential because the papers do not always mention SIDES directly. The central distinction is whether the simulation is designed primarily as a dusty far-IR/submm extragalactic sky or for a different wavelength regime and science case (OpenUniverse et al., 10 Jan 2025, Bisigello et al., 2020, Stein et al., 2020, Wilman et al., 2010).

Framework	Main regime	Relation to SIDES
OpenUniverse2024	Optical to near-IR Rubin/Roman sky	Matched multi-survey synthetic sky, not a far-IR/submm dusty-sky simulation
SPRITZ	Panchromatic UV-to-radio/X-ray, IR-LF anchored	Conceptually similar in goal, but luminosity-function-first and strongly AGN-aware
Websky	Full-sky microwave/CIB with tSZ, kSZ, lensing	Complementary CMB-oriented framework with a CIB component, not a detailed IR-galaxy mock
Wilman et al. (2010) IR-radio simulation	Mid-IR to sub-mm extension of a radio sky	Precursor and related semi-empirical mock, not the canonical SIDES model

OpenUniverse2024 is particularly important as a boundary case. The paper itself states that it should be understood as a matched Rubin/LSST and Roman synthetic sky with improved optical-to-near-infrared galaxy realism, not as a dedicated far-IR/submm extragalactic background simulation in the style of SIDES. SPRITZ, by contrast, is explicitly SIDES-adjacent: it starts from Herschel population-separated IR luminosity functions and builds a panchromatic spectro-photometric mock including AGN, low-resolution IR spectroscopy, radio properties, and X-ray properties, but it is luminosity-function-first rather than halo-lightcone-first. Websky is complementary rather than equivalent: it simulates the full extragalactic microwave sky for CMB analyses, including a CIB/DSFG component, but emphasizes full-sky maps and cross-correlations with tSZ, kSZ, and lensing rather than detailed infrared galaxy-population realism. The Wilman et al. infrared-radio simulation is best regarded as a precursor in the lineage of semi-empirical extragalactic sky simulations rather than as SIDES itself (OpenUniverse et al., 10 Jan 2025, Bisigello et al., 2020, Stein et al., 2020, Wilman et al., 2010).

Within SIDES proper, several limitations recur across releases. In the original continuum framework, the clustered simulation covers only 2 deg$0, environmental quenching is not modeled explicitly, lensing is probabilistic rather than spatially self-consistent, high-$0$0 are absent, and the original continuum model omitted AGN mid-IR contributions. The spectroscopic cubes omit instrumental noise and detailed line profiles in the main forecast products. The MIRI update imports AGN fractions statistically from SC SAM rather than modeling black-hole growth within pySIDES itself. The PRIMAger polarization forecasts rely on a simple integrated polarization prescription with no wavelength dependence, no AGN-specific treatment, and no intrinsic alignment of polarization angles. These limitations do not negate the utility of SIDES; rather, they define the framework’s proper domain as an empirically anchored, survey-oriented family of mock skies in which map-making, confusion, and selection effects are treated as first-order scientific variables rather than post-processing details (Bethermin et al., 2017, Bethermin et al., 2022, Vidal et al., 18 Sep 2025, Béthermin et al., 2024).