How well does MAGPHYS recover galaxy properties? A test using EAGLE simulated star-forming galaxies

Abstract

Spectral energy distribution (SED) models are widely used to infer the physical properties of galaxies from multi-wavelength photometry, but their accuracy is difficult to assess because the true properties of observed galaxies are generally unknown. We address this by fitting synthetic SEDs of ~31,000 star-forming galaxies drawn from the EAGLE cosmological simulations, post-processed with the SKIRT radiative transfer code, using the MAGPHYS SED modelling framework. This provides a controlled testbed with known intrinsic parameters, enabling a direct assessment of model accuracy and the origin of systematic biases. Under idealised conditions, fitting well-sampled ultraviolet-to-submillimetre SEDs at z=0.1, z=2, and z=5, MAGPHYS recovers stellar mass, star formation rate, specific star formation rate, dust mass, and dust luminosity to within <~0.14 dex, while mass-weighted stellar ages are not robustly constrained. We find that mismatches between the assumed star formation history (SFH) priors and the intrinsic SFHs of the simulated galaxies introduce systematic biases in stellar mass estimates, even when the fits provide good statistical agreement. To assess performance under realistic survey conditions, we construct a WAVES-like mock sample using optical and near-infrared photometry with realistic uncertainties. In this case, stellar masses and star formation rates remain well constrained (systematic offsets <~0.1 dex; scatters ~0.07 and ~0.15 dex, respectively), whereas dust properties degrade significantly without far-infrared data: dust luminosities show offsets of ~0.30 dex and scatters ~0.25 dex, and dust masses exhibit scatters ~0.3 dex. We conclude that MAGPHYS is a reliable tool for recovering key galaxy properties from broad-band photometry, but that SFH assumptions and limited wavelength coverage introduce significant uncertainties, particularly for dust and stellar ages.

Figures (12)

• Figure 1: Normalised distributions of nine parameters of the 10,288 galaxies at redshift $z=0.10$ (shown in blue), 18,206 galaxies at redshift $z=2.01$ (shown in green), and 2,434 galaxies at redshift $z=5.04$ (shown in red) of our idealised mock galaxy sample, and 35,051 galaxies at redshift $z \leq 0.27$ (shown in black) of our WAVES-like sample. The nine parameters displayed are: (a) stellar mass, (b) dust mass, (c) dust luminosity, (d) SFR, (e) sSFR, (f) average stellar age, (g) stellar metallicity, (h) dust temperature, and (i) half-mass radius. The vertical lines show the median of each distribution.
• Figure 2: Median synthetic SEDs of the three redshift samples (normalised in the $K$-band), showing the 16th and 84th percentile values as dotted lines, and the 50th percentile values as solid lines. The redshift $z=0.10$ galaxies are shown in blue, the redshift $z=2.01$ galaxies are shown in green, and the redshift $z=5.04$ galaxies are shown in red.
• Figure 3: The SEDs of three of the eagle galaxies when fit by magphys. The observed fluxes at each wavelength are shown as red dots, the unattenuated stellar emission are shown in blue, and the SEDs are shown in black, with the predicted fluxes at each wavelength shown as open, black circles. The galaxy ID is shown in the top left corner of each panel. The redshift of the galaxy and the $\chi^{2}$ value of each fit is shown in the top right corner of each panel. The residuals for each fit are shown below the SEDs, and share the same definition as the residuals shown in Fig \ref{['fig: Ave SEDs']}. The wavelengths shown are in the observer frame of reference.
• Figure 4: The median SED residuals for the three redshift samples, with redshift $z=0.10$ galaxies shown in panel a) in blue, redshift $z=2.10$ galaxies shown in panel b) in green, and redshift $z=5.04$ galaxies shown in panel c) in red. The error bars of all three panels show the 16th and 84th percentiles of the residuals, which share the same definition as the residuals shown in Fig \ref{['fig: SED fitting']}. The wavelengths shown are in the observer frame of reference.
• Figure 5: Comparisons between the true parameters of the eagle mock galaxies ($y$-axes), and the median-likelihood parameters recovered by magphys from fitting their skirt-processed SEDs ($x$-axes). The left-hand, middle, and right-hand columns show the comparison for the redshift $z=0.10$, $z=2.01$, and $z=5.04$ subsamples, respectively, for each parameter. The top row shows the comparisons of stellar masses, the middle row shows SFR, and the bottom row shows sSFR. All three parameters are coloured by the recovered magphys mass-weighted average stellar ages. The diagonal black line represents the one-to-one relation. Printed in the bottom right corners of the panels are the median offsets between the trends and the one-to-one relation, and the standard deviations of the trends. Printed in the top left corners of the panels are the median error bars of the recovered values, defined as the median difference between the 16th and 84th percentiles of each recovered value.