CIGALE Fitting Code for Galaxy SED Analysis

• CIGALE Fitting Code is a modular tool that models galaxy spectral energy distributions using comprehensive physical modules such as star formation, stellar populations, and dust attenuation.
• It employs both grid-based methods and MCMC sampling to robustly infer parameters like star formation rate, stellar mass, and AGN contribution through Bayesian statistics.
• Its flexible design supports diverse applications from large surveys to detailed AGN studies, enabling practical extraction of multiwavelength properties.

CIGALE Fitting Code

CIGALE (Code Investigating GALaxy Emission) is a modular, physically motivated model-fitting tool designed for the interpretation of galaxy spectral energy distributions (SEDs) from the far-ultraviolet (UV) to the radio domain. Its primary purpose is the inference of physical parameters—such as star formation rate (SFR), stellar mass ($M_*$), dust properties, and bolometric outputs—by modeling the multiwavelength photometric and spectroscopic data of galaxies and, more recently, active galactic nuclei (AGN). CIGALE achieves this by constructing composite SEDs through combinations of star formation and enrichment histories, stellar population synthesis, nebular emission, dust attenuation, dust and AGN re-emission, and implements flexible Bayesian-like statistics or, in its CIGALEMC variant, Markov Chain Monte Carlo (MCMC) sampling for extracting parameter constraints.

1. Architecture, Model Components, and Workflow

CIGALE employs a modular architecture, where each physical process—star formation, stellar population synthesis (e.g., BC03/CB07, Maraston2005), nebular emission, dust attenuation, dust and AGN IR reprocessing, and non-thermal radio emission—is encapsulated as a separate code module (Boquien et al., 2018). The construction of a galaxy’s SED proceeds according to a sequential chain:

• Star Formation History (SFH): Users may select from delayed, exponentially declining, periodic, or arbitrarily imported SFH profiles. The “delayed” SFH has the form:

${\rm SFR}(t) \propto t\,\exp(-t/\tau)$

and can be further modified to include bursts or truncations (Hunt et al., 2018).

• Stellar Populations: Composite stellar population (CSP) SEDs are constructed by convolving the SFH with stellar templates (e.g., Bruzual & Charlot 2003) and an initial mass function (IMF).
• Nebular Emission: The predicted Lyman photon production from the young stellar component is used to determine nebular line and continuum emission, using updated CLOUDY-based template libraries.
• Attenuation/Reddening: Attenuation is applied, with options for two-component prescriptions (birth cloud + ISM; e.g., Charlot & Fall 2000), modified Calzetti laws, or customized slopes (including a 2175 Å bump and slope parameter $\delta$).
• Dust Emission: Absorbed UV–to–optical energy is re-injected into the SED as IR emission, implemented through energy balance, and modeled with templates such as Dale et al. or Draine & Li (DL07/DL2014).
• AGN Component: AGN torus templates (e.g., Fritz et al., later SKIRTOR) are available to model thermal and scattered AGN emission, with fitting to IR SED shapes and inclusion of polar dust components (X-CIGALE).
• Radio Emission: Non-thermal synchrotron is included, typically parameterized by a power-law linked to star formation and AGN radio loudness (Yang et al., 2022).

Each model SED is calculated and convolved with observed photometric filter response curves:

$$F_{\text{model},\alpha} = \frac{\int T_\alpha(\lambda)\,F_\lambda(\theta)\,d\lambda}{\int T_\alpha(\lambda)\,d\lambda}$$

where $T_\alpha(\lambda)$ is the filter transmission function and $F_\lambda(\theta)$ the model spectrum for parameter vector $\theta$ (Roehlly et al., 2011).

2. Statistical Inference: Grid-Based, Bayesian, and MCMC Sampling

The original CIGALE design employs an exhaustive grid-based approach, constructing an $N$-dimensional parameter cube and calculating all plausible SEDs across the permitted parameter values. For each observation, likelihoods are assigned according to

$$\chi^2 = \sum_i \frac{(F_{\text{obs},i} - F_{\text{model},i})^2}{\sigma_i^2}$$

and the posterior

$$P(\theta|D) \propto \exp(-\chi^2/2) \cdot p(\theta)$$

where $p(\theta)$ is the user-supplied prior (Roehlly et al., 2011, Boquien et al., 2018). Marginalized probability distributions for each parameter are constructed as likelihood-weighted sums.

The CIGALEMC extension substitutes the grid with an MCMC sampler (cosmomc/Metropolis-Hastings), transitioning points by

$$\beta(\theta_i, \theta_{i+1}) = \min\left(1, \frac{P(\theta_{i+1})\,q(\theta_{i+1},\theta_i)}{P(\theta_i)\,q(\theta_i,\theta_{i+1})}\right)$$

and ensures exploration according to the detailed balance condition

$$P(\theta_{i+1})\,T(\theta_{i+1},\theta_i) = P(\theta_i)\,T(\theta_i,\theta_{i+1}).$$

Marginalization then proceeds simply by binning the MCMC samples; the density of samples traces the posterior (Serra et al., 2011). The MCMC approach achieves approximate linear scaling with parameter number and allows robust exploration of non-Gaussian posteriors and parameter degeneracies.

Convergence diagnostics such as the Gelman–Rubin statistic ($|R-1| \leq 0.03$) are implemented to verify sufficient sampling (Serra et al., 2011).

3. Applications: Multiwavelength SED Fitting and Parameter Recovery

CIGALE’s versatility is demonstrated in diverse contexts:

• Galactic-Scale Surveys: Applied to large datasets (e.g., thousands of galaxies in KINGFISH (Hunt et al., 2018), HELP (Malek et al., 2018), Hawaii-HDF-N (Gao et al., 2018)), CIGALE fits photometry spanning GALEX UV, SDSS/optical, 2MASS/NIR, Spitzer and Herschel/MIR/FIR, and VLA radio. Output parameters include $M_*$, SFR, dust luminosity ($L_\mathrm{dust}$), $A_\mathrm{FUV}$, and AGN fraction.
• Cluster and Star-Forming Region Analysis: SEDs of HST-resolved star clusters are modeled as single-stellar populations (SSP), fitting for age, mass, and extinction. Power-law mass functions $dN/dM \propto M^\beta$ are recovered, typically with $\beta\approx-2$ (Turner et al., 2021).
• Physical Parameter Correlations: Using Bayesian posteriors (or PCA on results), scaling relations such as

$$\log(M_\mathrm{dust}) = 0.48\,\log(\mathrm{SFR}) + 0.71\,\log(M_*) + 0.23$$

are established (Hunt et al., 2018).

• Red/Blue Classification: D4000 index (4000 Å break) is reconstructed either directly from the SED or by fitting with CIGALE, providing clean discrimination between galaxy populations, especially when using narrow-band photometry (Renard et al., 2022).

Significant outcomes include robust identification of AGN (using AGNFRAC), statistical quantification of parameter uncertainties, and identification of peculiar or mismatched sources via independent $\chi^2$ criteria for stellar and dust/IR components (Malek et al., 2018).

4. AGN and X-ray Extensions: X-CIGALE and Beyond

X-CIGALE and subsequent versions improve AGN/host decomposition and extend the fitting to the X-ray and radio regimes (Yang et al., 2020, Yang et al., 2022):

• X-ray Module: Incorporates an AGN X-ray SED as a power-law with exponential cutoff, allowing constraints on AGN power by enforcing consistency with the observed relationship

$$\alpha_\mathrm{ox} = -0.3838\ \log\left(\frac{L_\nu(2500\,\text{Å})}{L_\nu(2\,\text{keV})}\right)$$

and including host X-ray contributions from HMXBs/LMXBs/ISM gas.

• Clumpy Torus and Polar Dust: The AGN IR reprocessing models are updated to state-of-the-art (SKIRTOR); polar dust extinction is modeled along the axis, reprocessed into IR using a grey-body, allowing refined fits for both Type 1 and 2 AGN (Yang et al., 2020, Mountrichas et al., 2020). The intrinsic X-ray anisotropy is implemented as $L_X(\theta) \propto 1+\cos\theta$, matching observed suppression in edge-on AGN (Yang et al., 2022).
• Radio Module: AGN radio emission is parameterized by radio loudness $$R_\mathrm{AGN} = L_\nu(\text{5\,GHz}) / L_\nu(2500\,\text{Å})$$ and power-law index, decoupling the AGN and star formation radio contributions (Yang et al., 2022).

These modules have been validated on SDSS, COSMOS, and AKARI-NEP samples, with tested performance in decomposing AGN and host contributions, even under X-ray non-detection constraints.

5. Performance, Computational Scaling, and Limitations

Performance and Efficiency

• The original grid-based implementation scales exponentially with the number of free parameters; millions to hundreds of millions of models may be necessary for high-dimensional searches (Roehlly et al., 2011, Boquien et al., 2018).
• The MCMC approach (CIGALEMC) scales roughly linearly, achieving convergence (to parameter posteriors) with one order of magnitude fewer model evaluations (e.g., 160,000 vs. 3.5 million for a six-parameter fit) (Serra et al., 2011).
• Python re-implementation (pcigale) and later versions leverage multiprocessing and block-wise computations, enabling fits across large surveys and large grids (Boquien et al., 2018, Roehlly et al., 2013). However, pure Python is less performant than Fortran, occasionally necessitating Numpy-based or compiled optimizations.
• “Block-by-block” variance calculations and memory sharing enable large grid operation even with limited RAM.

Limitations

• Poorly constrained parameters (e.g., SFH timescales, dust bump slope $\delta$) yield broad or skewed posteriors that may be hidden in fixed-grid marginalizations (Serra et al., 2011). The outcome is sensitive to choice of priors and parameter limits.
• Error estimates on derived quantities can be underestimated due to grid discretization or limited sampling (Renard et al., 2022).
• As in all MCMC/likelihood-based methods, convergence diagnostics and proper management of burn-in and chain thinning are critical to avoid biased inference.

Contamination and False Positives

• The AGN fraction parameter (AGNFRAC) captures AGN IR excess, but contamination can result if the IR SED is reproduced by star formation and dust alone, or if MIR photometry is of low SNR (Siudek et al., 10 Jun 2025). When incorporating high SNR WISE data, star-formation contamination in AGN-selected samples falls from 62% to 15%.

6. Robustness and Application to New Generation Surveys

CIGALE underpins template libraries for photometric redshift codes (e.g., TOPz), SED-based D4000 determination, and major survey analysis for datasets from GALEX, SDSS, 2MASS, Spitzer, Herschel, and VLA (Tempel et al., 31 Mar 2025, Renard et al., 2022, Kompaniiets, 2023):

• Photometric Redshifts: Synthetic SEDs from CIGALE provide the basis for physically informed template sets, redshifted for likelihood maximization and marginalized over parameter subsets, with flux/uncertainty corrections and luminosity-function-based priors (Tempel et al., 31 Mar 2025).
• SED-based Classification and Outlier Detection: The modular output of CIGALE allows scrutiny of AGN, star forming, composite, and peculiar galaxies by comparing IR/stellar fit quality, SED shapes, and parameter consistency, facilitating large-scale classification in environments where traditional diagnostics (e.g., BPT or X-ray) are unavailable or incomplete (Siudek et al., 10 Jun 2025).
• Scaling Relations: New SED-based scaling laws (e.g., $M_*/L_{\mathrm{W1}}$), principal component decomposition, and cross-correlation of parameters (e.g., SFR, $M_*$, $M_{\rm dust}$, $12 + \log(\rm O/H)$) reveal empirical trends and intrinsic parameter degeneracies (Hunt et al., 2018).

Future directions emphasize expansion to panchromatic coverage, direct integration of high-resolution spectroscopy and narrow-band photometry, and enhanced MCMC/posterior error treatments as survey sample sizes and wavelength coverage increase.

---

CIGALE, through its grid-based and MCMC implementations, modular design, and extensibility to X-ray and radio, provides a robust computational framework for physical parameter estimation of galaxies and AGN using multiwavelength SED data. Its integration with major surveys and role as the reference SED fitter for extragalactic astrophysics is supported by extensive validation and a broad community user base. The extension to CIGALEMC notably addresses the curse of dimensionality in Bayesian inference, improving both accuracy and computational feasibility (Serra et al., 2011, Boquien et al., 2018).