H2 chemistry in galaxy simulations: an improved supernova feedback model

TL;DR

This paper introduces an improved supernova feedback model for galaxy simulations that accurately reproduces molecular hydrogen formation and outflows, offering a more physically motivated alternative to traditional methods.

Contribution

It presents a new supernova feedback model integrated with the GIZMO code and KROME, which better captures the physics of bubble expansion and molecular hydrogen formation.

Findings

Mechanical feedback reproduces H2 column densities without altering gas thermodynamics.

Delayed-cooling suppresses star formation but artificially changes gas state.

Model remains consistent across different simulation resolutions.

Abstract

In this study, we present and validate a variation of recently-developed physically motivated sub-grid prescriptions for supernova feedback that account for the unresolved energy-conserving phase of the bubble expansion. Our model builds upon the implementation publicly available in the mesh-less hydrodynamic code GIZMO, and is coupled with the chemistry library KROME. Here, we test it against different setups, to address how it affects the formation/dissociation of molecular hydrogen (H$_2$). First, we explore very idealised conditions, to show that it can accurately reproduce the terminal momentum of the blast-wave independent of resolution. Then, we apply it to a suite of numerical simulations of an isolated Milky Way-like galaxy and compare it with a similar run employing the delayed-cooling sub-grid prescription. We find that the delayed-cooling model, by pressurising ad-hoc the gas, is more effective in suppressing star formation. However, to get this effect, it must maintain the gas warm/hot at densities where it is expected to cool efficiently, artificially changing the thermo-chemical state of the gas, and reducing the H$_2$ abundance even in dense gas. Mechanical feedback, on the other hand, is able to reproduce the H$_2$ column densities without altering the gas thermodynamics, and, at the same time, drives more powerful outflows. However, being less effective in suppressing star formation, it over-predicts the Kennicutt-Schmidt relation by a factor of about 2.5. Finally, we show that the model is consistent at different resolution levels, with only mild differences.