CIGALE: a python Code Investigating GALaxy Emission

• The listed keywords place the work in data analysis, numerical and statistical methods, and general galaxy studies.Source: Context · p. 1

• A galaxy’s multi-wavelength SED records the interplay of its baryonic components and carries information about its formation and evolution, motivating broad-wavelength observations and models.The properties sought include star formation rate and history, stellar mass, attenuation, dust properties, and active-nucleus characteristics.Source: §1 · p. 1
• SED inference is challenging because galaxies with different properties can have similar spectra, especially when only restricted wavelength ranges are available.Dust-attenuated templates can estimate redshifts quickly but are limited for physical properties beyond stellar mass and can suffer age–metallicity and age–attenuation degeneracies.Source: §1 · p. 1
• Radiative-transfer models can provide detailed modelling, but their computational cost makes large model grids prohibitively expensive and has limited their observational use.The passage describes observational applications as largely confined to a handful of in-depth studies, generally of edge-on galaxies, with face-on modelling a more recent development.Source: §1 · pp. 1–2
• Energy-balance SED modelling links dust’s mid- and far-infrared emission to energy absorbed in the UV–optical, while Bayesian-like weighting accounts for observational uncertainties and parameter degeneracies.The passage attributes the approach’s success to using dust emission to break degeneracies while retaining relatively modest computing requirements.Source: §1 · p. 2
• The paper presents a completely new Python version of CIGALE, sharing the original code’s energy-balance principle and Bayesian-like strategy while addressing newer challenges and uses.The article describes the code as of version 2018.1.Source: §1 · p. 2
• The article introduces CIGALE’s architecture and modules, illustrates its applications, and leaves detailed Bayesian theory and precision and accuracy discussions to earlier work.It also notes that additional technical details, performance benchmarks, and examples are provided in appendices.Source: §1 · p. 2

2.1. Design principles

• CIGALE is designed around modularity, clarity, and efficiency, with independent stages and replaceable physical modules supporting adaptation, community extensions, and library use.The stated efficiency goal is to model thousands of galaxies across millions of models within a few hours on a typical multicore computer.Source: §2.1 · pp. 2–3

2.2. Choice of programming language

• Python was chosen for its accessible syntax and astronomy uptake, rich scientific libraries, fast numerical and parallel computing, and free licensing.The libraries handle numerical work, plotting, astronomy-specific tasks, and data input/output, letting developers focus on scientific challenges.Source: §2.2 · p. 3

2.3. High–level overview of CIGALE

• CIGALE generates theoretical models and can optionally fit them to observations to estimate galaxy physical properties.When used only for model generation, the analysis step is skipped.Source: §2.3 · p. 3, §2.3.1 · p. 3

2.3.1. The division of labour

• The pcigale executable computes models and, when requested, estimates galaxy properties; pcigale-plots and pcigale-filters provide plotting and filter-management utilities.The companion executables are provided for convenience and to facilitate interpretation of pcigale results.Source: §2.3.1 · p. 3
• Runs proceed through input handling, model generation, fitting and likelihood-based property estimation, then output; model-only generation skips analysis.Input data are processed and normalised before models are computed for parameter combinations.Source: §2.3.1 · p. 3

3. Model creation modules

• Models are computed by successive modules for star formation, stellar and nebular emission, attenuation and dust re-emission, active nuclei, and redshifting with intergalactic absorption.The model sequence can include a rest-frame module for quantities such as equivalent widths, UV slope, and colours; individual and combined spectra are stored to support later calculations and component retrieval.Source: §3 · p. 4

3.1. Star formation history

• CIGALE supports both parameterised analytical and arbitrary SFHs, but each run can use only one SFH type.The available prescriptions span different histories, while file-based input permits histories derived from simulations or semi-analytic models.Source: §3.1 · p. 4, §3.1.3 · p. 6, sfhfromfile · p. 6

3.1.1. Basic assumptions on the SFH

• SFHs are sampled every 1 Myr, with ten instantaneous 0.1 Myr-separated episodes per bin, and are normalised to 1 M⊙ of formed stars.The SFR is evaluated at each bin’s beginning and stellar contributions at its end; the normalisation excludes mass return and is later scaled to observations.Source: §3.1.1 · pp. 4–5

3.1.2. sfh2exp, sfhdelayed, and sfhperiodic modules

• The analytic modules provide double-exponential, delayed, and periodic SFHs, with the delayed form allowing a smoother rise and an optional burst.The double-exponential form can represent nearly constant or rising histories but is unsuitable for recent SFR drops; a delayed-SFH extension allows an instantaneous upward or downward change followed by a constant SFR, and was used to model KINGFISH galaxies.Source: §3.1.2 · p. 5, sfh2exp · p. 5, sfhdelayed · p. 5

3.1.3. sfhfromfile module

• CIGALE also supports rotation-velocity-dependent SFHs and file-based arbitrary histories that can be evaluated at selected ages.File inputs can be ASCII, FITS, or VO-table formats, with optional normalisation to 1 M⊙ of formed stars.Source: sfh_buat08 · p. 5, sfhfromfile · p. 6

3.2. Stellar populations

• Composite stellar spectra are computed from SFHs and Bruzual & Charlot (2003) or Maraston (2005) SSP libraries, with young and old stars stored separately for differential attenuation.The SSP libraries offer multiple metallicities and IMFs; the age separating young and old populations defaults to 10 Myr and is configurable.Source: §3.2 · p. 6

3.3. Nebular emission

• Nebular templates based on Inoue (2011) model 124 emission lines and hydrogen continuum, with line emission corrected for Lyman-continuum escape and dust absorption.The templates are parameterised by ionisation parameter and metallicity, assume an electron density of 100 cm−3, and omit metal and CO lines from photo-dissociation regions and molecular clouds.Source: §3.3 · pp. 6–7

3.4.1. The dustatt_modified_CF00 module

• To represent diverse galaxy attenuation curves, CIGALE implements the Charlot & Fall (2000) model and flexible laws inspired by the starburst curve.In the Charlot & Fall implementation, stars younger than 10 Myr receive birth-cloud and ISM attenuation, while older stars receive ISM attenuation only; the default power-law slopes are −1.3 and −0.7, respectively.Source: §3.4 · p. 7, §3.4.1 · p. 7
• CIGALE offers distinct attenuation modules, including a modified Charlot–Fall model and a flexible starburst curve with adjustable slope and UV bump.The starburst curve builds on Calzetti et al. (2000), extended below 150 nm using Leitherer et al. (2002); emission-line attenuation can use Milky Way or Magellanic Cloud curves, with lines dimmed more than stars. The separate dustatt_powerlaw module uses one power law for young and old stars, with different absolute attenuation.Source: §3.4.2 · p. 8, §3.4.1 · p. 8

3.5. Dust emission

• CIGALE includes three dust-emission model sets: Dale et al. (2014) empirical templates, Draine & Li (2007, 2014) models, and the Casey (2012) analytic model.Dust emission spans PAH-dominated mid-IR wavelengths, warmer small grains, and colder large grains beyond about 100 µm; heating, composition, metallicity, and radiation-field properties affect the SED. The Dale templates use a single parameter α linked to 60-to-100 µm colour, but PAH variation is limited, which can be problematic for metal-poor galaxies.Source: §3.5 · p. 9, §3.5.1 · p. 9

3.5.2. dl2007 and dl2014 modules

• The dl2007 and dl2014 modules separate diffuse dust heated by Umin from star-forming-region dust exposed to radiation fields spanning Umin to Umax.They also parameterize the star-forming dust fraction γ and PAH mass fraction qPAH; dl2014 expands radiation-field and PAH ranges and makes α free. Their flexibility comes with greater processing and memory costs than the Dale et al. (2014) templates.Source: §3.5.2 · p. 9

3.5.3. casey2012 module

• The Casey (2012) module combines a single-temperature modified black body in the far-IR with a mid-IR power law, storing both components separately.It is flexible for local and high-redshift galaxies, but includes no PAH emission and excludes AGN heating because its infrared emission follows energy balance.Source: §3.5.3 · p. 9

3.6. Synchrotron radio emission

• CIGALE’s synchrotron module estimates 21 cm luminosity density from infrared emission and qIR, then scales a power-law spectrum with a free slope α.Radio data can also help estimate infrared emission when other data in that range are unavailable.Source: §3.6 · p. 9

3.7. Active galactic nuclei

• CIGALE offers rapid AGN modeling with Casey (2012) for IR-only fits or Dale et al. (2014) quasar templates from UV to IR, alongside slower, detailed Fritz et al. (2006) radiative-transfer models.Fritz models explicitly include primary torus-source, dust-scattered, and thermal dust emission.Source: §3.7 · p. 10

3.8. Measuring physical properties from the rest–frame spectrum

• Before redshifting, CIGALE derives UV slope β, Dn4000, IRX, rest-frame equivalent widths, luminosity densities, and colours from the full spectrum.These quantities are computed from wavelength-range fits or integrations over user-defined ranges and available filters.Source: §3.8 · pp. 10–11

3.9. Redshifting

• The redshifting module scales wavelengths by 1 + z, dims the spectrum by 1 + z, and applies Meiksin (2006) IGM absorption.If redshifts are not specified, CIGALE derives a list from input fluxes, rounding to two decimal places by default, while computing physical quantities at the exact input redshift.Source: §3.9 · p. 11

4.1. Computing physical quantities

• Intrinsic intensive and extensive properties are computed in physical modules, whereas fluxes are measured after model computation and depend on the observer and measurement type.Flux computation differs for bandpasses and low- or high-resolution spectra.Source: §4.1 · p. 11

4.1.1. Bandpasses

• Bandpass fluxes are computed by integrating the model spectrum through each filter, interpolating spectrum and transmission curves to preserve sampling, then converting to frequency units using the pivot wavelength.Filters are rescaled so integration yields flux density in mJy.Source: §4.1.1 · pp. 11–12

4.1.2. Emission lines

• For low-resolution spectra, continuum-subtracting filters use positive transmission over the line and negative transmission off-line; for high-resolution data, line fluxes come from nebular templates.Low-resolution filters subtract the inferred continuum, while the high-resolution template method applies Lyman-continuum normalisation and dust extinction.Source: §4.1.2 · p. 12

4.2. savefluxes module

• savefluxes computes spectra, fluxes, and physical properties for parameter-grid or explicitly specified models, then saves integrated outputs and optionally full spectra and SFHs.The explicit-list option suits simulated galaxies whose properties do not follow a grid; outputs can be saved as ASCII or FITS tables.Source: §4.2 · p. 12

4.3. pdf_analysis module

• pdf_analysis fits a fixed grid of models and uses likelihood-weighted properties to estimate galaxy parameters and uncertainties, with marginalised PDFs for more complex distributions.The fixed grid is computed once for all objects; it must be sufficiently sampled and remain in memory, trading memory use for computational efficiency.Source: §4.3 · pp. 12–13
• pdf_analysis compares observations with models by scaling them, computing χ2 and likelihoods, then saving likelihood-weighted estimates, uncertainties, and the best-fitting model outputs.The module can also save the best-fit spectrum, χ2 distribution, and marginalised PDFs; extensive properties are scaled with a multiplicative factor.Source: §4.3 · p. 13
• For fluxes with only upper limits, CIGALE adopts Sawicki (2012), separates χ2 into regular and upper-limit terms, and numerically solves ∂χ2/∂α = 0 for each model.After χ2 is computed, subsequent steps no longer depend on whether upper limits are present. Invalid values (<0 or “nan”) are omitted, and available data and band counts are handled per object.Source: §4.3 · pp. 13–14
• A mock-catalogue analysis tests whether properties are reliably estimated by comparing known mock values with likelihood-based estimates after adding observation-scale Gaussian noise.The mock is built from each object's best fit and analysed like the original observations.Source: §4.3 · p. 14

5.1. Example of CIGALE as an SED generation tool

• Mid-to-near-IR flux-ratio variation is typically under 20% for nearby bands but reaches a factor of two at WISE 22 µm or MIPS 24 µm, with stronger SFH dependence.The largest differences occur for early-type galaxies with small τ; galaxies around 1 Gyr old have bluer colours, while ongoing star formation yields weaker colour dependence and permits a constant colour for late-type galaxies.Source: §5.1 · p. 14, Figure 8 · p. 15
• Generated model grids can test assumptions and connect simulations to observables, including SFR estimators from 23 high-resolution simulations and AGN-property studies.The example grid varies the age and timescale of a delayed SFH; similar investigations can vary the SFH parametrisation, IMF, stellar-population models, metallicity, dust, or an active nucleus.Source: §5.1 · p. 14

5.2. Example of CIGALE as a library

• CIGALE’s modular design supports library-based applications, from an interactive dust-attenuation explorer to uniform access to base models without custom format-reading code.The example interface explores attenuation effects on the FUV-to-far-IR spectrum and could be extended to parameters across all modules.Source: §5.2 · p. 14, Figure 9 · p. 16

5.3. Example of CIGALE to estimate the physical properties of star–forming galaxies

• CIGALE shows excellent consistency with MAGPHYS and grasil for 100-Myr-average SFR, stellar mass, FUV attenuation, and dust luminosity.A separate effort focused on higher-redshift samples also reports excellent performance (Pacifici et al., in prep.).Source: §5.3 · pp. 15–16
• Mock-catalogue tests recover SFR, Ldust, and IRX with scatter below 0.1 dex, β with scatter ≤0.217, and A(FUV) with scatter 0.157 dex.The test compares estimates with known input values; its uncertainties are lower limits because it does not include uncertainty in the models themselves. The Brown et al. (2014) sample comprised 78 selected star-forming galaxies and excluded pure AGN; its six-property grid contained 8,164,800 models.Source: §5.3 · p. 16, Figure 12 · p. 18
• The SFH helps explain IRX–β offsets: active galaxies follow the Kong et al. (2004) relation, while increasingly quiescent galaxies move toward the Boquien et al. (2012) relation.CIGALE’s flexibility allows investigation across galaxies ranging from intense starbursts to ellipticals.Source: §5.3 · p. 18, Figure 13 · p. 18

6. Summary

• CIGALE is designed around modularity, clarity, and efficiency for both developers and users.Modules can be added or swapped, the code is intended to be easy to use and understand, and parallelization exploits multi-core processors.Source: §6 · p. 18
• Its modular FUV-to-radio models use energy balance, re-emitting dust-absorbed UV-to-near-IR energy in the mid- and far-infrared.The models account for flexible star-formation histories and stellar populations, ionised gas, dust attenuation and re-emission, active nuclei, and the intergalactic medium.Source: §6 · p. 18
• Parallel computation builds model grids reaching several hundred million elements, which can be saved for theoretical studies or used to evaluate observed objects.Physical-property estimates use likelihood-weighted means and standard deviations while accounting for upper limits.Source: §6 · pp. 18–19
• CIGALE can also serve as a library for building new applications.The paper presents a snapshot of a code that continues to evolve, with future papers planned to describe major developments for changing panchromatic-modelling challenges.Source: §6 · p. 19
• The code and its documentation are publicly available.The paper describes its current state as a snapshot and anticipates further major evolutions.Source: §6 · p. 19

Appendix A: General workflow

• Figure A.1 schematically presents the general CIGALE workflow, including important choices detailed in Appendix B.Source: Appendix A · p. 19

Appendix B: Important implementation choices

• CIGALE’s SED-fitting workflow comprises input handling, model generation, analysis, and output handling.Model generation and fitting run in parallel, while shared data remain in memory.Source: Figure A.1 · p. 20

Appendix B.1: Parallelisation strategy

• CIGALE parallelises model generation, object analysis, and plot generation using Python’s process-based multiprocessing module.A user-defined number of child processes each computes one model or analyses one object on demand from the main process.Source: Appendix B.1 · p. 20
• Shared arrays hold model fluxes, physical properties, and analysis results, avoiding repeated copies and reducing communication costs.Sub-processes access these arrays directly, improving computation speed and memory usage.Source: Appendix B.1 · p. 20
• Parallel speed gains initially grow with physical-core count but become sub-linear as memory-to-processor communication becomes a bottleneck.Launching sub-processes adds slight start-up delay, and using them also incurs a small memory cost.Source: Appendix B.1 · p. 20

Appendix B.2: Cache system

• CIGALE caches model-creation modules so each is instantiated and its parameter-independent data computed only once per run.This avoids repeating potentially expensive module initialisation whenever that module is applied to an SED.Source: Appendix B.2 · p. 20
• CIGALE caches intermediate SEDs so later model steps can reuse earlier computations, then discards intermediates no longer needed.It orders model computation to retain only a small number of useful intermediate SEDs, balancing cache speed and memory use.Source: Appendix B.2 · pp. 20–21

Appendix B.3: Computation by blocks of models

• CIGALE computes large grids in blocks, estimates target properties from each block, and combines block results into overall estimates.The overall best fit is the model with the lowest χ2 across blocks; likelihood-weighted means and standard deviations are combined using a variance-parallel algorithm.Source: Appendix B.3 · p. 21

• Table C.13 is identified as listing the bc03 module’s input parameters.Source: Table C.13 · p. 26
• The listed outputs include the initial-mass-function choice and the age separating young and old populations.Source: Appendix C.2.1 · p. 25, Table C.14
• The output list includes young- and old-population masses, Lyman-continuum photon counts and luminosities, population luminosities, and total Lyman-continuum counts and luminosity.Source: Appendix C.2.1 · p. 25, Table C.14

Appendix C.2.2: m2005

• Table C.15 lists the input parameters of the m2005 module.Source: Table C.15 · p. 25
• Table C.16 lists the output parameters of the m2005 module.Source: Table C.16 · p. 27

Appendix C.3.1: nebular

• Table C.17 lists input parameters for the nebular module.Source: Table C.17 · p. 26
• Table C.18 lists output parameters for the nebular module.Source: Table C.18 · p. 26

Appendix C.4.1: dustatt_modified_CF00

• Tables C.19 and C.20 are captioned as listing input parameters for the dustatt_modified_CF00 module.The supplied passages contain the table captions but no parameter names or values.Source: Table C.20 · p. 27, Table C.19 · p. 28

Appendix C.4.2: dustatt_modified_starburst

• Table C.21 lists input parameters for the dustatt_modified_starburst module.Source: Table C.21 · p. 27
• Table C.22 lists output parameters for the dustatt_modified_starburst module.Source: Table C.22 · p. 27

Appendix C.5.1: dale2014

• Table C.23 presents the input parameters of the dale2014 module.Source: Table C.23
• Table C.24 presents the output parameters of the dale2014 module.Source: Table C.24 · p. 28

Appendix C.5.2: dl2007

• The passages identify four dl2007 module parameter tables: Tables C.25 and C.27 list input parameters, while Tables C.26 and C.28 list output parameters.Source: Table C.26 · p. 28, Table C.27 · p. 28, Table C.25 · p. 29

Appendix C.5.4: casey2012

• The Casey 2012, synchrotron, and fritz2006 appendices identify tables of module input and output parameters.Source: Table C.30 · p. 29, Table C.31 · p. 29, Table C.32 · p. 29

Appendix C.8.1: restframe_parameters

• Rest-frame outputs include the Calzetti UV slope, Balogh D4000 break, IRX from GALEX FUV and dust luminosity, emission-line wavelengths, luminosities, and colours.Source: Appendix C.8.1 · p. 30, Table C.36 · p. 30, [[restframe_parameters]] · p. 34

Appendix C.9.1: redshifting

• The redshifting module is documented with separate input- and output-parameter tables.Source: Table C.37 · p. 31, Table C.38

Appendix D: Usage of CIGALE executables

• The executable-usage table covers CIGALE model generation and fitting, pcigale-plots for SEDs, χ2 and PDFs, and pcigale-filters for filter management and plots.Source: Table D.1 · p. 31

Appendix E: Example of a pcigale.ini configuration file

• The example input file uses object IDs, redshifts, optional distances, fluxes, and uncertainties; an optional parameter file can replace configuration parameters and is intended for savefluxes.Source: Appendix E · p. 31
• The example configuration selects sfhdelayed, bc03, nebular, dustatt_modified_starburst, and dale2014; the module list also describes radio, rest-frame parameters, and mandatory redshifting with IGM.Source: Appendix E · pp. 31–32, # * dustatt_modified_CF00 (modified… · p. 32, # * casey2012 (Casey 2012 dust… · p. 32
• Example settings specify BC03 Chabrier populations, nebular emission, modified-starburst attenuation, rest-frame measurements, and PDF analysis of selected properties.Source: [[bc03]] · p. 33, [[nebular]] · p. 33, [[dustatt_modified_starburst]] · p. 33