🔭 astrophysics

Neutral hydrogen gas in supernova-driven galactic outflows

High-resolution GPU-accelerated simulations demonstrate that supernova-driven galactic outflows can reproduce the observed dominance of neutral mass by capturing essential mixing-layer condensation processes, a feature missed in lower-resolution studies that previously led to incorrect conclusions about the necessity of alternative driving mechanisms.

Original authors: Chong-Chong He, Mark Krumholz, Benjamin Wibking, Aditi Vijayan

Published 2026-09-29

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Chong-Chong He, Mark Krumholz, Benjamin Wibking, Aditi Vijayan

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Galaxies are not static islands of stars; they are living systems that breathe, exhaling vast amounts of gas into the space between them. This outflow, driven by the violent explosions of dying stars, acts as a cosmic regulator, stripping galaxies of the raw material needed to form new stars and shaping their future evolution. For decades, astronomers have known that these galactic winds carry not just hot, ionized gas, but also a significant amount of cool, neutral hydrogen—the same kind of gas that makes up the clouds where stars are born. Observations of our own Milky Way and other galaxies show that this neutral material can survive the journey for millions of years, traveling thousands of light-years away from the galactic disk before fading away. Yet, when scientists tried to recreate these winds in computer simulations, they hit a wall. The models consistently failed to produce enough neutral gas. In the digital universe, the cool clouds seemed to evaporate almost instantly, leaving behind only hot, ionized plasma. This discrepancy led many to speculate that supernovae alone were too weak to drive these winds, suggesting that invisible forces like cosmic rays or radiation pressure must be doing the heavy lifting.

A team of researchers has now challenged this conclusion, not by changing the physics of their models, but by sharpening the lens through which they view them. Using a powerful, specialized computer code designed to run on graphics processors, they simulated a supernova-driven galactic wind with a level of detail far exceeding previous attempts. The key to their success was resolution. In the world of computer simulations, "resolution" is akin to the pixel count on a screen; a low-resolution image is blurry and misses fine details, while a high-resolution image reveals the intricate structure hidden within. The researchers ran their simulations at three different levels of detail, effectively zooming in on the galactic wind. At the lowest resolution, comparable to earlier studies, the wind behaved as expected: the neutral gas vanished quickly, and the outflow was dominated by hot, ionized material. However, as they increased the resolution, the story changed dramatically. At their finest setting, where the simulation could distinguish structures as small as one parsec (about 3.26 light-years), the wind was dominated by neutral gas, carrying nearly 70 percent of the total mass. This result matched real-world observations perfectly, suggesting that the missing neutral gas in previous models was not a failure of physics, but a failure of detail.

The reason for this dramatic shift lies in how the wind interacts with the gas it carries. In the high-resolution simulation, the neutral gas does not simply ride along as a solid object; it is constantly being reshaped. The hot, fast-moving wind crashes into the cooler, denser clouds, creating a turbulent boundary layer where the two phases mix. In this mixing zone, the hot gas cools down rapidly and condenses, adding fresh mass to the neutral clouds. Simultaneously, the pressure of the hot wind pushes against the clouds, transferring momentum and accelerating them outward. The researchers found that this process of condensation and momentum transfer is so efficient that it allows the neutral clouds to survive the journey for tens of millions of years, growing larger as they travel further from the galaxy. Crucially, this survival mechanism only works if the simulation is fine-grained enough to resolve the thin, turbulent layers where the cooling and mixing happen. In the lower-resolution runs, these critical layers were too thin to be seen by the computer grid, so the code could not capture the condensation process. Without this mass replenishment, the clouds were stripped away and destroyed, leading to the false conclusion that neutral gas could not survive in such winds.

The implications of this finding are profound for our understanding of how galaxies evolve. It suggests that supernovae are entirely capable of driving winds rich in neutral gas, removing the need to invoke more exotic or uncertain mechanisms to explain the observations. The study also highlights a fundamental lesson for the field: the way a galaxy's wind is composed is not fixed at the moment it leaves the disk. Instead, the wind evolves as it travels, with the neutral fraction being set en route by the continuous process of condensation. This means that the material ejected into the space surrounding a galaxy is very different from the material that was originally launched. By demonstrating that the previous failure to reproduce neutral winds was a numerical artifact rather than a physical one, the researchers have provided a clearer path forward. Their work indicates that to truly understand the life cycle of galaxies, simulations must be detailed enough to resolve the delicate, turbulent interfaces where the hot and cold phases of the universe meet and exchange their properties.

Technical Summary: Neutral Hydrogen Gas in Supernova-Driven Galactic Outflows

Problem Statement
Observations of galactic outflows, particularly within a few kiloparsecs of the galactic disc, indicate that the majority of the mass flux is carried by neutral and molecular gas rather than ionized material. However, theoretical simulations of supernova-driven winds have historically struggled to reproduce this observation. Previous stratified-box simulations, where winds are driven solely by supernovae, typically show that neutral gas is ionized or destroyed before reaching distances of $\gtrsim 1$ kpc, resulting in neutral mass loading factors ( $\eta_{\text{HI}}$ ) of $\lesssim 10^{-4}$ . This discrepancy has led to speculation that non-thermal drivers, such as radiation pressure or cosmic rays, are required to entrain neutral material. Furthermore, the survival of neutral clouds against cloud-crushing timescales in hot winds has been difficult to explain without invoking idealized "wind-tunnel" scenarios or additional physical mechanisms like magnetic fields or thermal conduction.

Methodology
The authors present a series of high-resolution simulations of a supernova-driven outflow using quokka, a GPU-accelerated, block-structured adaptive-mesh radiation-hydrodynamics code. The simulations model a stratified patch of a galactic disc with periodic horizontal boundaries and "diode" boundaries in the vertical ( $z$ ) direction to allow outflow.

Resolution Study: The core of the methodology is a resolution convergence study. The authors perform simulations at three grid spacings: $\Delta x \approx 1$ pc (flagship), $\Delta x \approx 2$ pc, and $\Delta x \approx 4$ pc. The highest resolution ( $\Delta x \approx 1$ pc) is achieved using a uniform grid of $1024^2 \times 8192$ cells over a domain of $\approx 1 \times 1 \times 8$ kpc.
Physics: The simulations include hydrodynamics, self-gravity, optically thin radiative cooling, photoelectric heating, and self-consistent star formation and feedback (FUV radiation and supernovae). The hydrodynamics scheme employs piecewise-parabolic reconstruction (PPM) with the HLLC Riemann solver, offering higher effective resolution for advection compared to the piecewise-linear methods common in prior work.
Analysis: The gas is decomposed into three phases based on electron abundance and temperature: neutral ( $x_e < 0.05$ ), warm ionized medium (WIM), and hot ( $T > 10^6$ K). The authors analyze the momentum budget, Lagrangian acceleration, and structural properties of neutral clouds (mass, effective radius, density) to understand the mechanisms driving and sustaining the neutral phase.

Key Results

Resolution Dependence of Neutral Mass: The simulations demonstrate a qualitative shift in outcomes based on resolution.
- At $\Delta x = 1$ pc, the outflow is dominated by neutral gas. At heights of $1-4$ kpc, neutral gas carries approximately 69% of the mass and ~40% of the mass flux. The neutral mass loading factor ( $\eta_{\text{HI}}$ ) remains stable at $\approx 0.3$ and does not decay with height.
- At $\Delta x = 2$ pc, the neutral mass loading drops to $\approx 0.17$ , and the spatial distribution of neutral gas changes qualitatively.
- At $\Delta x = 4$ pc, neutral gas is virtually absent in the wind ( $\eta_{\text{HI}} \lesssim 10^{-3}$ ), reproducing the results of previous lower-resolution studies.
- Crucially, the total mass loading factor varies only by tens of percent across resolutions, indicating that the suppression of neutral gas at lower resolutions is not due to a failure to launch the wind, but a failure to sustain the neutral phase.
Mechanism of Acceleration and Survival:
- Acceleration: The neutral gas is not ballistic; it accelerates upwards at $\sim 5$ km s $^{-1}$ Myr $^{-1}$ up to $\approx 2.5$ kpc. This acceleration is driven by momentum transfer from the hot phase via two channels: ram pressure/drag and condensation. The study finds that condensation becomes increasingly dominant at higher altitudes. Neither mechanism alone is sufficient to overcome gravity; their sum explains the observed acceleration.
- Survival: The survival of neutral clouds is attributed to rapid radiative cooling in the mixing layers between the hot wind and the cold clouds. The authors calculate that the mixing layer cooling time ( $t_{\text{cool,mix}}$ ) is roughly an order of magnitude shorter than the cloud-crushing time ( $t_{\text{cc}}$ ), satisfying the survival criterion proposed by Gronke and Oh (2018). This allows clouds to re-supply mass lost to stripping.
Resolution Requirement: The characteristic size of the cooling layer required for survival is calculated to be $\ell_{\text{cool}} \approx 3-6$ pc.
- At $\Delta x = 1$ pc, these layers are resolved by $\approx 3-6$ cells.
- At $\Delta x = 4$ pc, the mixing layer is reduced to a single cell, and typical clouds are at most two cells in size.
- The authors conclude that previous simulations failed to produce neutral-dominated winds because they could not resolve the mixing layers necessary for both momentum transfer and mass re-supply via condensation.

Significance and Claims
The paper claims that the steep vertical decline of neutral gas reported in previous supernova-driven outflow simulations is, at least in part, a numerical artifact resulting from insufficient resolution. By resolving the mixing layers, supernova feedback alone is shown to be capable of driving winds dominated by neutral material to distances of several kiloparsecs, in good agreement with observations of the Milky Way's nuclear wind.

The authors assert that:

Radiative cooling of mixing layers is sufficient to explain the survival of neutral clouds, without requiring magnetic fields or thermal conduction (though these may further enhance the effect).
Cosmic rays and radiation pressure are not strictly necessary to explain the presence of neutral gas in outflows, potentially reducing the need to tune uncertain parameters in models of those drivers.
Subgrid wind prescriptions in cosmological simulations, which are often calibrated on lower-resolution stratified-box runs, may be inheriting a numerical loss of the neutral phase. This implies that current models may underestimate the neutral mass flux delivered to the circumgalactic medium.
The neutral fraction of a wind is not fixed at launch but evolves en route due to condensation, meaning the composition at the halo boundary differs from the launch composition.

The authors note that their results are conservative (unmagnetized, no explicit thermal conduction) and that full numerical convergence has not yet been established, suggesting the reported neutral mass loading is a lower bound. However, the qualitative transition from $\eta_{\text{HI}} \sim 0$ to $\eta_{\text{HI}} \sim 0.3$ with increased resolution is a robust finding.

More like this