H-alpha light puts galaxy feedback simulations to the test

Galaxies are not simply collections of stars forming from a fixed supply of gas. Energy from stars and from matter falling onto central black holes can heat gas, move it around or drive it out. These feedback processes influence how galaxies grow, but their complex effects are difficult to model. The study asks whether the resulting H-alpha light from galaxies can help distinguish between two widely used simulations, Illustris and IllustrisTNG, which use different feedback prescriptions.

H-alpha is a specific color of light emitted by hydrogen. In galaxies, it often comes from gas energized by young, massive stars, so its brightness can serve as a signpost of recent star formation. But it is not a perfect one-to-one indicator: gas outside star-forming regions can also emit H-alpha, black-hole activity can contribute, and dust can dim the light before it escapes. The researchers therefore modeled several emission sources and how dust and gas affect the signal.

They analyzed outputs from the Illustris and IllustrisTNG computer simulations, including TNG100 and, at one epoch, the larger TNG300 volume. For each simulated galaxy, they estimated H-alpha emission from unresolved star-forming H II regions, collisional excitation and recombination in diffuse gas, and—where included—excitation driven by radiation from an active galactic nucleus (AGN). They also calculated dust attenuation using simulated gas positions and metallicities, accounting for viewing direction and for dust destruction in hot gas. For AGN light, they modeled how hydrogen absorbs Lyman-series photons as they travel through a galaxy.

The team compared the simulated H-alpha luminosity functions with published measurements around redshifts 1 and 1.5. A luminosity function describes how numerous galaxies are at different brightnesses. They adjusted one main parameter, linking simulated star-formation rates to H-alpha output from H II regions, to fit the observations. The observed data were not corrected for dust, so the comparison used simulated luminosities after modeled dust attenuation.

In the models, H II regions supplied most of the H-alpha light. For galaxies brighter than 10^41 erg per second, diffuse-gas processes contributed about 5–15% on average. Adding these processes made little difference to the overall fit. AGN-powered emission had a modest effect on the bright end of the luminosity function. In Illustris, it became the dominant modeled source for the brightest galaxies, above roughly 10^43 erg per second—but those bins contain few simulated objects, making that result subject to sampling noise.

The central comparison favored IllustrisTNG: its predictions provided a reasonable fit to the observed luminosity function for the emission models tested. The higher-feedback Illustris predictions differed significantly from the data at both redshifts, with the mismatch driven especially by the faint end, where the observational uncertainties used in the fit were smaller. This is evidence about these simulations and this observable, not a direct measurement of feedback strength in real galaxies.

That distinction matters because another probe points in a different direction. Stacked kinetic Sunyaev–Zel’dovich (kSZ) measurements—signals that can reveal the distribution of free electrons around groups of galaxies—have been reported to fit the higher-feedback Illustris simulation better. H-alpha mostly traces cooler gas inside galaxies, whereas kSZ is sensitive to ionized gas along the line of sight, including more extended gas. The two measurements may therefore expose shortcomings in different gas phases; the paper suggests that combining them could help test galaxy-formation models.

Several assumptions limit how firm the comparison can be. Dust properties at these redshifts are uncertain, and the adopted dust prescription was calibrated using nearby galaxies. The inferred fit parameter changes when the assumed temperature threshold for dust destruction changes. The authors also note that their treatment of Lyman-photon transmission may underestimate how much AGN-driven H-alpha is produced. Sparse bright-end bins and finite simulation volumes add uncertainty, while the fit does not include uncertainty in the simulated luminosity function itself.

The source is an arXiv preprint, presented in the supplied paper text as a draft; it should not be treated as peer-reviewed confirmation. Its conclusion is consequently best read as a model comparison that motivates further checks. More precise faint-galaxy H-alpha measurements and joint comparisons with kSZ data could help clarify which ingredients of feedback and gas physics need revision.