BUGS2026: Building Galaxies from Scratch II

Recent JWST observations of high-redshift galaxies signal deviations of the stellar initial mass function (IMF) from the form inferred in the Solar neighbourhood. We present large-scale cosmological hydrodynamical simulations of the new COLIBRE galaxy formation model that has been adapted to allow non-universal IMFs, self-consistently adjusting metal yields and feedback energetics. Here we focus on results from simulations in a cosmological volume of (100 cMpc)^3 with a gas and dark matter particle resolution of ~ 10^6 Msolar. We vary the high mass slope of the IMF and choose to make the IMF increasingly top heavy (flatter slope) for stellar populations that form from dense gas. Our model yields an elevated far-UV luminosity in young stellar populations, resulting in high redshift (9 < z < 16) UV luminosity functions that are in stronger agreement with recent data from JWST than current galaxy formation models. One of the advantages of the COLIBRE model is its live dust model, enabling us to track the impact that a variable IMF has on dust formation and evolution in galaxies. The increase in the number of core-collapse Supernovae events from the top heavy stellar populations results in higher dust masses per galaxy in our variable IMF simulations than the original COLIBRE model, meaning that a significant fraction of the extra UV photons produced by top heavy stellar populations get extinguished. We compute dust attenuated UV luminosities using the radiative transfer code SKIRT and find that the most luminous galaxies are dimmed via dust by up to 5 magnitudes at z=9, or up to 3 magnitudes at z=12. In fact, accounting for dust attenuation has negligible impact on the shape of the UV luminosity function for 9 < z <16 but limits the attenuated magnitudes to M_UV > -20 (M_UV > -19 at z=15).

Simulating the Milky Way at individual-star resolution is key to uncovering its formation and evolution, as observations now reveal detailed dynamical and chemical structures. We present a star-by-star Milky Way simulation with 4.4 trillion particles on 9,216 nodes of Aurora, achieving 10,000× higher resolution than the current state of the art. Our code processes 125 billion particles per second, about three times faster than existing exascale GPU-based N-body/SPH simulations. The gravitational kernel sustains 934 PFLOP/s. The implementation is performance-portable, achieving near-ideal scaling on multi-vendor systems such as JSC JUPITER and ALCF Aurora. We present the feasibility of extreme-scale, high-fidelity galaxy simulations across heterogeneous architectures.

The growth of supermassive black holes (SMBHs) and the feedback they drive are central to galaxy evolution across cosmic time, yet the physical processes involved span an enormous dynamic range, from sub-parsec accretion and jet launching near the event horizon to the large-scale thermodynamics of the intracluster medium (ICM) and beyond. Capturing this self-consistently within a single simulation framework remains a formidable numerical challenge. I will present results from a suite of simulations performed with the moving-mesh code AREPO that combine super-Lagrangian refinement (SLR) with novel sub-grid accretion disc models, including spin evolution, to self-consistently drive jets and winds across an unprecedented range of scales.

On the smallest scales, I will present simulations of binary SMBHs embedded in gas-rich circumbinary discs. Thanks to SLR, these simulations resolve the gas streams and minidiscs that form around individual black holes, allowing us to track spin evolution and predict spin alignment timescales with implications for gravitational wave observations and post-merger recoil velocities.

At intermediate scales, I will present simulations of jet feedback in low-mass and dwarf galaxy systems, where jets launched via spin-based sub-grid models interact with the interstellar medium, driving energetic outflows whose properties we can directly characterise. I will discuss feedback efficiency and outflow properties in these environments, highlighting how jet power, morphology, and ISM coupling depend sensitively on local conditions, with important implications for SMBH self-regulation in low-mass systems.

On the largest scales, I will present high-resolution jet feedback simulations within cosmologically evolved cluster environments. These reveal how jets and their inflated lobes interact with a dynamic ICM shaped by cluster weather and magnetic fields, and how jet-driven turbulence can promote gas cooling under specific physical conditions, providing new insight into the self-regulating nature of AGN feedback. Mock X-ray and radio observables generated from these simulations offer direct tests for current and upcoming facilities.

I will conclude by reviewing current observational constraints on AGN feedback efficiency, highlighting where consensus has emerged and where significant discrepancies remain, before outlining key open questions and future directions toward a coherent, multi-scale framework for SMBH growth and feedback.

I will report on our recent effort to build from scratch galaxies at comic dawn including predictive models of star formation and supermassive black holes.

Massive black holes (MBHs) play a crucial role in galaxy formation simulations, significantly impacting surrounding gas and reducing star formation through feedback effects. While cosmological galaxy formation models are effective at quenching massive galaxies and suppressing cooling flows, the detailed mechanisms of active galactic nucleus (AGN) feedback remain uncertain. This talk outlines new progress in linking AGN feedback - via jets and winds - to observed multi-wavelength signatures. Using cosmological and isolated cluster simulations with jet feedback, I will show how AGN jets influence cluster thermodynamics and turbulence as measured using X-ray data from XRISM, and discuss modelling radio emissions with a Fokker-Planck solver for non-thermal electrons to compare spectral aging and multifrequency observations, in particular low-frequency LOFAR data. Additionally, I’ll share initial findings on connecting AGN activity to warm ionized and neutral gas phases traced by H-alpha, [OIII] emission and NaID absorption, and compare to recent JWST NIRSpec results.

Star formation in galaxies is highly inefficient. This is due to feedback processes, which span multiple gas phases and spatial scales. Large-volume cosmological simulations struggle to capture the multiphase nature of the interstellar medium. Warm and cold clouds remain spatially unresolved, which leads to numerical overcooling if left unaddressed. To prevent numerical fragmentation and excessive star formation, current large-scale simulations usually employ effective equation of state prescriptions. However, while stabilising, this approach oversimplifies the interstellar medium. It is locked in a single state, which prevents important dynamical effects, such as the launching of galactic winds. We present a sub-resolution, multi-fluid approach for modelling the interstellar medium. This method not only stabilises the gas, but also models the interactions between the different gas phases, stellar feedback and turbulence. It enables the self-consistent launching of hot, pressure-driven winds from star forming regions. We will discuss the advantages compared to effective equation of state approaches and will give an outlook on applying our model in future large-volume cosmological simulations.

I will talk about NewCluster project, a new high-resolution cluster simulation designed to serve as the massive halo counterpart of the modern cosmological galaxy evolution simulations. The zoom-in simulation targets a volume of 4.1σ overdensity region, which is expected to evolve into a galaxy cluster with a virial mass of 5×10^14Msun, comparable to that of the Virgo Cluster. The novelties of NewCluster are found in its resolutions. Its stellar mass resolution of 2×10^4Msun is effective for tracing the early assembly of massive galaxies as well as the formation of dwarf galaxies. The spatial resolution of 68 parsecs in the best-resolved regions in the adaptive-mesh-refinement approach is powerful to study the detailed kinematic structure of galaxies. The time interval between snapshots is also exceptionally short-15 Myr-which is ideal for monitoring changes in the physical properties of galaxies, particularly during their orbital motion within a larger halo. The simulation has up-to-date feedback schemes for supernovae and active galactic nuclei. The chemical evolution is calculated for ten elements, along with dust calculation that includes the formation, size change, and destruction. To overcome the limitations of the Eulerian approach used for gas dynamics in this study, we employ Monte Carlo-based tracer particles in NewCluster, enabling a wide range of scientific investigations. The simulation has reached z = 0.45, capturing the terminal stage of a major merger between two main clusters. We will present general properties of the NewCluster simulation, including scaling relations of galaxy properties such as the mass–size relation and the stellar-to-black hole mass relation. We will also highlight recent scientific results obtained using NewCluster.

Over the last decade, cosmological simulations have become an invaluable tool for understanding how galaxies form and evolve. Yet, despite their successes, different flagship models often disagree on key observables, revealing that the achievements of the current generation rest on uncertain foundations. The DREAMS simulation efforts have a primary objective of improving these foundations through systematic variations in dark matter physics, cosmology, and astrophysics. Coupled with machine learning, these variations allow us to better characterize simulation behavior, quantify key uncertainties, and more efficiently design future runs. This talk will first highlight how our new suite of ~800 zoom-in simulations with the IllustrisTNG model can successfully reproduce the full-box scaling relations of TNG50 at a fraction of the computational cost and with the addition of physics variations. Next, we extend this even further with a separate suite of 1000 simulations with continuously varied mass resolution that allows for effective Machine Learning-driven ``super-resolution'' mapping from low resolution to high resolution outputs. We can therefore, for the first time, characterize the resolution dependence of a galaxy formation model at a granular level. Finally, we demonstrate a novel method for ``baryonification'' of dark-matter-only volumes based on our zoom-in simulation suites, allowing us to effectively generate group catalogs for large box hydrodynamic simulations without needing to run them. Together, these approaches provide orders-of-magnitude speed-ups and a scalable framework for incorporating model uncertainty into next-generation galaxy simulations.

Recent observational evidence suggests that baryonic feedback, how energy from active galactic nuclei (AGN) and supernovae redistributes gas within and beyond halos, may be more extreme than commonly assumed in current cosmological hydrodynamical simulations. Whether such extreme feedback is physically plausible, which astrophysical mechanisms could drive feedback on these scales, and how simulation models must be updated to capture these processes remain open questions. In this talk, I will outline a path forward that combines new, publicly available multi-wavelength survey data with next-generation hydrodynamical simulations to transform our understanding of baryonic feedback. I will first present a roadmap for comparing simulations to kinetic Sunyaev–Zel’dovich (kSZ) measurements, showing that most state-of-the-art simulations under-predict gas ejection from haloes, underscoring the need for stronger or qualitatively different feedback models. I will then introduce a new suite of hydrodynamical simulations exploring AGN feedback, providing insight into the mechanisms required to reconcile recent observational evidence of large-scale gas redistribution, whilst maintaining consistency with key galaxy, group and cluster properties.

Ultra-faint dwarf (UFD) galaxies are the faintest galaxies known in the universe. As the most dark matter–dominated systems, they serve as ideal laboratories for probing cosmology on the smallest scales, with the potential to provide new constraints on the nature of dark matter. However, their small size also makes them highly sensitive to baryonic feedback processes, as well as to their implementation in numerical simulations.

The GEAR galaxy formation model has successfully reproduced key properties of classical dwarf galaxies as well as UFDs, including their luminosity, star formation histories (SFHs), metallicity and abundance ratios. However, studying the faintest galaxies requires pushing spatial resolution to new limits, which in turn calls for an adaptation of the physical model.

In this talk, I will present the upgraded GEAR model, which now incorporates mechanical feedback, non-equilibrium cooling and explicit metal diffusion. Crucially, we introduced a novel star formation prescription using sink particles to resolve individual massive stars and their pre-supernova feedback.

Within this framework, I performed high-resolution hydrodynamic cosmological zoom-in simulations of UFDs and dwarf galaxies. My results demonstrate how this high-resolution treatment of individual stellar feedback events shapes the early assembly of UFDs and their influence on the chemical evolution. I will finally highlight that although macroscopic scaling relations remain robust, the stochastic nature of individual UFDs makes them highly sensitive to the choice of feedback and cooling physics.

Star formation is a fundamental process in galaxy evolution, yet many simulations continue to overproduce stars due to excessive gas cooling. To explore a physically motivated solution to this long-standing problem, we perform cosmological zoom-in radiation-hydrodynamics simulations of a dark matter halo reaching 10^10 Msun at z=6, using two distinct star formation models: a Schmidt-type model, where star formation efficiency is regulated by local gravo-thermo-turbulent conditions, and a sink-based model, where star formation is directly governed by gas inflows onto sink particles. We find that the sink-based model naturally produces highly bursty star formation, as young sink particles are embedded in strongly convergent flows and thus exhibit high accretion rates. This leads to coherent and strong stellar radiation that fully ionizes and disrupts star-forming clumps via photoionization heating before the first supernova explodes. As a result, star formation is rapidly quenched and clumps are efficiently dispersed, closely resembling the behavior seen in high-resolution GMC simulations. Due to the effective pre-processing by radiation feedback, supernovae explode in lower-density environments, injecting greater momentum into the ISM and driving stronger galactic outflows. In contrast, star formation within individual gas clumps in the Schmidt-type model is more inefficient and intermittent. Consequently, insufficient radiation feedback fails to heat star-forming clumps, as radiative cooling exceeds photoionization heating. As a result of the ineffective pre-processing, supernovae explode in dense environments, allowing clumps to survive for unrealistically long times (>100 Myr) and grow to large masses (>10^7.5 Msun). Compared to the Schmidt-type model, the sink-based model yields a total stellar mass lower by a factor of 3.2 and a Lyman continuum escape fraction higher by a factor of 10 at z=6. Moreover, by driving stronger metal-enriched outflows and suppressing excess star formation at the galaxy center, the bursty sink-based model exhibits better agreement with JWST observations in terms of gas-phase metallicity and galaxy size at a given stellar mass. Our results suggest bursty star formation is a key mechanism for enhancing feedback strength and alleviating the overcooling problem in galaxy formation simulations.

Understanding the self-regulation of star formation through stellar feedback and galactic outflows is a central challenge in galaxy evolution. I present Quokka, a GPU-accelerated adaptive mesh refinement (AMR) radiation-magnetohydrodynamics code that I co-developed, and its application to simulations of star formation and feedback-driven outflows in galactic patches. Quokka integrates MHD, self-gravity, spectrum-resolved radiation transfer, radiative cooling, star formation, and stellar feedback within a massively parallel framework that scales efficiently on modern GPU supercomputers. I will present results from tall-box simulations of galactic patches with star formation, where feedback from supernovae, radiation pressure, and radiative heating drives multiphase galactic outflows. I will discuss phase-resolved outflow properties, metal loading, their correlation with the star formation rate, and the properties of the interstellar radiation field, and what these imply for the baryon cycle in star-forming galaxies.

Simulating chemical evolution requires modeling how metals are produced and distributed by stellar populations. In cosmological simulations, this is typically done using the mean properties of stellar populations, obtained by integrating a continuous initial mass function (IMF). This approximation breaks down at high resolutions, where the discrete nature of star formation becomes important, and in the early Universe and metal-poor environments, where a small number of metal-free Population III stars can dominate the enrichment due to their expected high masses and top-heavy IMF.

In this talk, I present a framework for chemical enrichment that explicitly accounts for the stochastic nature of stellar populations. Our formulation retains the standard stellar particle description while resolving IMF discreteness effects by tracking the lifetimes and nucleosynthetic yields of individual stars. This approach enables a self-consistent treatment of enrichment from the first stars and captures the fluctuations in feedback and chemical yields that are smoothed out in traditional models. Using a suite of hydrodynamical cosmological simulations, we show that stochastic sampling and finite stellar lifetimes introduce enhanced scatter and systematically different abundance patterns in metal-poor regimes. These differences are important for interpreting the chemical signatures of the first stars and constraining the star formation histories of low-mass galaxies.

Finally, I present the public release of this framework, designed for straightforward integration into existing simulation codes and enabling more physically grounded chemical modeling in high-resolution cosmological simulations.

Star-forming clouds cannot be resolved in large-volume cosmological simulations of galaxy formation. The SFR, stellar feedback, and the gas phases of the ISM are thus determined by subgrid models. While current state-of-the-art models succeed in reproducing the basic properties of galaxies, they are missing key physics and generally cannot predict observables such as molecular abundances, emission lines, or turbulent velocity dispersions (without post-processing). We are developing a new equilibrium model that is similar in spirit to Springel & Hernquist 2003, but that explicitly includes a physical warm-cold phase distinction, low-temperature cooling rates, FUV heating, and turbulence. The model is guided by spatially resolved observations of the ISM that are now available thanks to surveys such as PHANGS or EDGE. I will describe the current status of our efforts and contrast the model with alternative approaches to ISM modeling.

The COLIBRE project includes simulations of galaxy formation and evolution in large cosmological volumes with a side length of up to 400 cMpc. The masses of the individual resolution elements range from ~10^5 Msun (highest resolution simulations, "m5") to ~10^7 Msun (lowest resolution simulations, "m7"). COLIBRE is to date the largest simulation project that directly models a multi-phase interstellar medium (ISM), which means that gas in the simulation may become neutral and cool to very low temperatures (T<100 K). At COLIBRE resolution, the classical Jeans length of self-gravitating gas is formally unresolved in cold gas. This has often been cited as numerically problematic but I will show how numerical fragmentation is avoided and how a realistic multi-phase ISM can form, despite the limited resolution. Furthermore, in modelling the chemical processes in neutral gas, I found that the transition from neutral to molecular hydrogen sensitively depends on reactions with oxygen species. In primordial networks that only include hydrogen and helium species or, more generally, in chemical networks that do not include the destruction of molecular hydrogen by oxygen species, the HI-H2 transition is shifted to lower densities, resulting in an excess of H2.

Understanding the multi-phase interstellar medium (ISM) is essential for modeling star formation and the evolution of galaxies. The neutral atomic ISM is typically split into two stable phases with distinct temperatures and densities based on the equilibrium between heating and cooling rates, and co-exist in pressure equilibrium. In the large-scale cosmological simulations of COLIBRE (Schaye et al., 2026), radiative and chemical processes are evaluated using the HYBRID-CHIMES model (Ploeckinger et al., 2025), where non-equilibrium calculations are combined with quasi-equilibrium cooling rates to model the ISM. We present the typical temperatures, densities, and pressures of the multi-phase ISM in COLIBRE galaxies and show how these properties depend on metallicity and mass of galaxies. We compare our results to expectations assuming thermal equilibrium and to available observations.

ΛCDM predicts that baryons and CDM emerge from recombination with distinct density fields, leading to compensated spatial variations in the local baryon-CDM ratio. Linear theory predicts that overdense regions will be baryon deficient relative to the cosmic mean, a feature that may be important for early galaxy formation. Yet most cosmological simulations to date still neglect this spatially varying baryon-CDM ratio when generating their initial conditions (ICs).

In this talk, I will present results from cosmological simulations using the FLAMINGO galaxy formation model that compare standard one-fluid initial conditions with physically consistent two-fluid initial conditions. I will show that the baryon deficiency imprinted at recombination survives non-linear collapse, and that its amplitude and time evolution are well described by a simple analytic model. In terms of galaxy formation, this deficiency leads to systematic reductions in halo gas supply, star formation rates, stellar masses, and black hole growth, with the largest effects at high redshift and negligible differences by the present day. As JWST and upcoming facilities increasingly probe the early Universe, incorporating two-fluid ICs into cosmological simulations will be essential for producing unbiased predictions of early galaxy formation and black hole growth.

Simulations of galaxy evolution must track the evolving abundances of various ions and molecules in the interstellar medium to robustly predict emission and absorption lines seen in galaxy observations and evaluate the radiative cooling of the gas. The chemical composition of the ISM evolves due to interactions of gas particles with each other as well as with photons, cosmic rays, and dust grains. Quantitatively, this evolution is described by a non-equilibrium chemical network consisting of coupled differential equations for the abundance of each species. Even for a simple molecule like CO, so many species are involved in its production that solving the full relevant chemical network is computationally impractical in most simulations of galaxy evolution. As an alternative approach, we develop machine learning methods to emulate non-equilibrium ISM chemical networks for use in hydrodynamic simulations. We generate a suite of training runs of a chemical network with over 600 ionic and molecular species, sampling values for the gas density, temperature, and cell size, as well as the incident radiation field strength. We train an autoencoder to project the full space of abundances down to a lower-dimensional latent space and emulate the evolution of the latent variables across arbitrary time steps using the relevant gas and radiation field parameters. In a simulation, this emulator can be implemented by 1) applying the encoder network to the current chemical abundances to get latent variables, 2) evolving these latent variables forward with the appropriate time step and local radiation field and gas properties, and 3) applying the decoder network to translate from latent variables back to the full set of (evolved) abundances. This emulator will enable galaxy simulations (which include a multiphase ISM) to directly model the density of important molecular and ionic tracers and produce mock emission and absorption line signatures that can be compared with observations.

Investigation of cosmic ray transport and dynamical impacts in time-dependent numerical simulations of the multiphase ISM constitutes an exciting new avenue in the study of galaxies. CRs, in principle, may significantly affect galactic structure and evolution, but their impacts are highly contingent on CR-gas coupling. This coupling is, in turn, sensitive to both the adopted theoretical model and to environmental conditions, motivating the development of simulation suites that incorporate physically motivated, self-consistent CR treatments and cover diverse galactic environments. In pursuit of this goal, I will present new results from the TIGRESS++ suite of CRMHD shearing-box simulations of the multiphase ISM. By combining advanced treatments of star formation and ISM thermochemistry, realistic SN and FUV feedback, and environmentally sensitive live cosmic ray transport, we are better able than ever to probe the interplay between CRs and the galactic ISM on pc to kpc scales. Further, our simulation suite spans a wide range of galactic conditions, allowing accurate characterization of the behavior and impacts of CRs from dwarfs to starbursts. Amongst our results, we find that CRs have a fundamental impact on the nature, loadings, and phase structure of galactic outflows, with major implications for both galaxies and the circumgalactic medium. While CRs have minimal impact on average star formation rates when environment-dependent scattering is incorporated in simulations, they do appear to play a role in controlling the “burstiness” of inflows and outflows. We additionally draw conclusions about the nature of GeV CR transport, finding that diffusion dominates in the cooler gas of the galactic disk, while transport in the extraplanar regions and outflows is instead dominated by “dynamical” transport by streaming along magnetic fields and advection by gas motion.

The gaseous reservoirs of the interstellar and circumgalactic media (ISM and CGM) ultimately fuel star formation and thereby dictate how galaxies evolve. Modelling their highly turbulent internal dynamics is challenging, as they are not affected by a single physical process but rather by the interplay of feedback processes, such as supernova explosions and subsequently accelerated cosmic rays (CRs), with thermochemical processes and gravity. Our efforts to numerically model the plethora of microphysical processes and astrophysical feedback channels culminated in the development of the CRISP (Cosmic Ray and InterStellar Physics) framework. It contains a state-of-the-art transport model for the CRs, a detailed thermochemical network to solve for the thermodynamical state of the ISM and CGM, as well as descriptions for radiative and supernova feedback. We employ this model in high-resolution (up to 100 Msol) idealized simulations of Milky Way-mass galaxies to investigate the difference in wind driving between purely supernova-driven and CR-driven galactic winds. We will demonstrate that both exist at the same time but operate on different scales: while (clustered) supernovae are effective at driving a fountain flow with cold material out of the ISM close to the disk, only the additional forcing provided by CRs can accelerate cold gas up to O(10 kpc). Furthermore, the hot phase in vertically expanding supernova-fuelled plumes behaves differently in a CR-laden background medium due to altered buoyant properties. We will comment on how this decisively impacts mass and energy transport through the inner CGM.

Feedback processes play a central role in shaping the structure and evolution of galaxies. Baryonic matter cycles through stars, which return energy to the interstellar medium (ISM) via supernova explosions, thereby driving multiphase galactic winds. Cosmic rays (CRs), accelerated in supernova remnants, represent a key component of this feedback. While CRs can significantly contribute to wind driving, their overall impact is highly sensitive to the underlying CR transport model. In this work, I present high-resolution “tallbox” simulations of a galactic disk patch performed with the moving-mesh magnetohydrodynamics code Arepo, incorporating a range of CR transport prescriptions as well as the CRISP non-equilibrium thermochemistry model. We investigate how CR feedback influences both star formation and the properties of multiphase outflows. Our results show that CR-driven winds can efficiently sustain large-scale outflows, whereas purely thermal winds dissipate most of their energy within ~3 kpc above the disk midplane. We further demonstrate that the steady-state structure of the wind is strongly dependent on the chosen CR transport model. In particular, a model including CR advection, streaming, diffusion, and nonlinear Landau damping produces especially strong feedback. When ion–neutral damping is additionally included, CRs partially decouple from the cold ISM, leading to a reduced impact on the star formation rate while still maintaining efficient feedback on galactic scales. We address the apparent paradox that, despite infrequent CR–wave interactions, the CR population does not fully decouple from the plasma but instead propagates at velocities well below the speed of light. This behavior has important consequences for galaxy evolution, including modifications to the formation of star clusters, which exhibit lower velocity dispersions and are therefore more strongly bound by self-gravity.

Cosmic ray protons (CRs), with their substantial energy density, play a key role in galaxy evolution, and shaping the interstellar medium (ISM). Accelerated in supernova (SN) shocks, CRs propagate along magnetic field lines, distributing energy and momentum throughout the galaxy. This process ionizes and heats the gas, drives large galactic outflows by creating pressure gradients, and regulates star formation. Because CRs cannot be directly observed except in our local environment, diffuse gamma-ray emission provides a crucial observational tracer of their distribution, encoding information about their transport, energetics, and interaction with the ISM. High-resolution observations are, however, limited to the Milky Way, whose observed emission is shaped by our local environment within the Local Bubble.

To investigate the impact of CR feedback, we perform high-resolution magnetohydrodynamical simulations of Milky-Way-like galaxies, in which we follow individual massive stars and include self-consistent stellar feedback such as SNe and CRs, dynamically coupled to the MHD equations. We model the multi-phase interstellar medium using a non-equilibrium chemical network that includes hydrogen and carbon species, allowing us to take into account the relevant cooling and heating processes. Additionally, we back up our simulations with post-processed, multi-wavelength gamma-ray emission from CR protons, enabling us to analyze luminosities, spectra, full-sky emission maps, and angular power spectra for many observer positions, including those located inside Local Bubble-like environments.

We first present the effects of thermal and CR feedback on the galactic structure and gas dynamics across different ISM phases. We find that CRs convert fountain flows into sustained galactic outflows - absent in the MHD-counterparts - driving mass loss from the entire disk and magnetizing the CGM. We analyze these outflows in terms of energy loading and their impact on vertical structure. Finally, we show that the simulations naturally reproduce key observational properties of the Milky Way, including gamma-ray luminosities and spectral slopes. We also demonstrate that the local environment plays a significant role in shaping the simulated gamma-ray sky, highlighting the importance of understanding the observer’s local surroundings when using gamma-rays to trace Galactic CR physics.

The circumgalactic medium (CGM) is the nexus of galaxy formation. Fresh gas that fuels star formation must pass through the CGM to reach the interstellar medium, while material accelerated by feedback and expelled from galaxies builds up a gaseous reservoir in the CGM. Observations reveal vast amounts of cold gas in the CGM, yet the processes that shape its dynamics remain uncertain. Condensation powered by thermal instability can generate cold gas clouds in situ, but can be offset by feedback processes. Cosmic rays (CRs), a key component of non-thermal galactic feedback, impact cold gas clouds by heating and exerting pressure, which can slow the inevitable condensation process.

We present the results of a dedicated simulation suite designed to shed light on the role of CRs in the formation and evolution of cold gas clouds in the CGM. We show that the impact of CRs on thermal instability depends sensitively on the details of CR transport. In particular, fast CR transport enables rapid escape of CRs from condensing regions, thereby weakening their stabilizing effect and allowing clouds to collapse more efficiently. In contrast, inefficient transport leads to strong CR pressure support that suppresses cold gas formation. Furthermore, we emphasize the importance of numerical resolution: insufficiently resolved simulations systematically exaggerate CR-mediated effects on the multiphase structure of the CGM.

To directly connect these findings to observations, we generate synthetic absorption-line spectra using a Voronoi-based ray-tracing approach. The imprint of CR transport on observables is striking: efficient CR transport produces systematically stronger and broader absorption in low-ionization tracers, driven by both an increased cold gas fraction and enhanced kinematic complexity, and manifests in higher equivalent widths and covering fractions.

Magnetic fields permeate the Universe across a wide range of scales and strengths. On galactic scales, they are amplified by turbulent dynamo processes and eventually saturate at dynamically important levels. In realistic environments, this evolution involves both direct and inverse turbulent cascades, driven by energy injection from supernovae on small scales and extending up to galactic scales.

These processes can be investigated using numerical simulations. In this work, we study the growth of magnetic fields in an idealized isolated Milky Way–like galaxy using the SWIFT astrophysical simulation code with smoothed particle magnetohydrodynamics (SPH-MHD). We assess the reliability of this approach and identify the main numerical and physical challenges that arise in modeling galactic dynamos.

The mismatch between the number of massive galaxies (and their properties) observed by JWST at high-z and predictions from current theoretical and numerical models highlights a shortcoming in our understanding of star formation physics at this early epoch. It implies that baryonic physics models might be missing crucial physical dependencies or that star formation physics might change across cosmic time. These results highlight the need for improved physics models in numerical simulations, in order to provide a better framework for the interpretation of current and future observations.

I will present first results from the HIGHLANDS suite of cosmological zoom-in simulations, precursor to the HIMALAYA cosmological volume(s). The fiducial HIGHLANDS physics include a star formation model that depends on the cloud-scale turbulent properties of the gas, on-the-fly treatment for the Lyman-Werner radiation emitted by massive young stars and a model for population III stars. Non-equilibrium chemistry for hydrogen and helium, new metal yield tables, metal line cooling for the interstellar medium down to 10K, as well as early and supernova stellar feedback are also included. The HIGHLANDS suite comprises galaxies across a wide range of halo masses from ~1e9 to 1e12 Msun evolved to z=5. Furthermore, each galaxy is simulated multiple times with variations in the key physics, e.g. switching off local Lyman-Werner radiation, making this an ideal testbed for the impact of these physics on the formation of the earliest galaxies.

I will discuss the complex interplay between Lyman-Werner radiation, star formation and stellar feedback, and their impact on galaxy properties at z=5, as well as their evolution across cosmic time. I will show that the total stellar mass and star formation rates at z=5 differ significantly between runs with and without local Lyman-Werner radiation, a direct consequence of Lyman-Werner radiation inhibiting further popIII star formation following the formation of the first stars. This results in different star formation histories and dramatically different gas fraction evolution and highlights the importance of capturing this effect on the thermo-chemical evolution of the gas at runtime. Furthermore, the natal properties of stars (e.g. birth density, metallicity) are significantly affected as well. These results present an important step towards a better understanding of the impact these small-scale processes have on galaxy formation and evolution in the early universe.

Interpreting multi-wavelength observations of high-redshift galaxies from JWST and ALMA requires numerical simulations that self-consistently couple ISM chemistry, stellar feedback, and radiative transfer across a wide dynamic range, along with reliable predictions of observable properties. We post-process the MEGATRON suite of cosmological zoom-in radiation hydrodynamics simulations using the radiative transfer code SKIRT. the MEGATRON simulations incorporate non-equilibrium thermochemistry and on-the-fly radiative transfer in Milky Way–mass halos. We perform self-consistent dust continuum and non-LTE line radiative transfer calculations for [OI], [OII], [OIII], [NII], and [CII] emission, accounting for both stellar and dust radiation. Our sample spans redshifts z = 4–8.5 and includes five sub-grid physics models that vary in star formation efficiency, supernova feedback strength, hypernova feedback, and the initial mass function.

We find that the spatial distributions of FIR and optical line emission are strongly shaped by the adopted sub-grid physics. Models with stronger feedback produce more extended and irregular morphologies with reduced dust attenuation, while compact, disk-dominated systems with efficient star formation exhibit stronger obscuration. Among the tracers, [OI] lines show the tightest correlation with star formation rate, whereas [CII]158 μm correlates with the neutral hydrogen gas mass. The [OIII] emission depends on multiple factors, including ionization state, gas density, metallicity, and gas mass. The gas ionization playing a dominant role in driving high [OIII] luminosities at fixed star formation rate. These results provide physically motivated calibrations for FIR and optical star formation tracers and offer a robust theoretical framework for interpreting ALMA and JWST observations of the high-redshift galaxies.

Simulations of galaxy formation during the epoch of reionization must simultaneously model radiative transfer, stellar feedback, gas flows, and chemical enrichment across a wide dynamic range. This makes the problem difficult not only physically, but also numerically. In this talk, I will discuss what is needed to move toward a better model of galaxy formation in reionization simulations.

Rather than treating reionization history alone as the primary benchmark, I argue that improved models should also explain how feedback shapes the gas around early galaxies, how metals are produced and transported, and how these processes appear in observable tracers. Using examples from oxygen absorption, feedback-regulated enrichment, and emerging galaxy scaling relations, I will show how these diagnostics can be used to test whether simulations capture the relevant physics in a self-consistent way. I will also discuss numerical directions, including local timestepping, that may help make such models more computationally feasible.

Stellar feedback-driven outflows couple the evolution of individual stars to the flow of mass and energy on circum- and intergalactic scales, making them one of the most intrinsically multi-scale processes in galaxy formation. Capturing this coupling remains a central challenge for numerical models, particularly in connecting the discrete physics of stellar evolution to emergent galactic-scale winds.

We present Imladris (Smith 2026), a flexible framework for incorporating explicitly realised stellar populations in galaxy formation simulations across a wide range of resolutions, from star-by-star treatments to intermediate-resolution models with individually sampled stellar populations. Imladris includes a comprehensive treatment of stellar feedback including supernovae (core-collapse, pair-instability, and Type Ia), stellar winds (from massive and AGB stars), and radiation (photoionising, photodissociating, and photoelectric heating), all self-consistently linked to individually tracked stars, alongside detailed enrichment following 27 elements.

Using idealised star-by-star simulations of low-redshift dwarf galaxies, we show that the clustering of supernovae, regulated by early feedback, sets the structure and efficiency of emergent galactic winds. This clustering is highly sensitive to numerical resolution: at coarse resolution, where many supernova progenitors are grouped in a single particle, feedback becomes artificially over-clustered, leading to systematically more bursty and violently ejective outflows. This highlights the importance of resolving, or explicitly modelling, the discrete nature of massive star formation in order to capture realistic wind properties.

We then present first results from a new suite of ultra-high-resolution cosmological zoom-in simulations (Nelson et al.; Smith et al., in prep.), reaching a baryonic mass resolution of 3 Msun in galaxies with Mstar ~ 1e6 - 1e9 Msun at z = 5.5. These simulations reveal a qualitatively different outflow regime at high redshift: feedback drives highly time-variable, short-lived, and strongly ejective outflows, in contrast to the more steady, quasi-equilibrium winds seen in our analogous low-redshift systems. We interpret this transition as arising because, at high redshift, galaxy evolution proceeds on time-scales closer to the lifetimes of massive stars, preventing a steady, feedback-regulated state and instead driving persistent burstiness; capturing this behaviour fully self-consistently requires a star-by-star treatment of stellar feedback, as implemented in Imladris.

The circumgalactic medium (CGM) acts as both the reservoir of gas that fuels star formation and galaxy growth, and the interface where galaxies deposit the products of stellar evolution and feedback, making it a crucial site for constraining galaxy formation physics. The column densities and covering fractions of ions in the CGM provide powerful probes of gas flows and metal enrichment. Different ions (Mg II, O VI, C IV) trace distinct phases of the CGM, and their combination can place strong constraints on feedback processes.

In this talk, I will present results from a systematic study exploring how variations in galaxy formation parameters, particularly those governing stellar feedback and galactic outflows, impact CGM properties and observables. Using new cosmological zoom-in simulations of Milky Way-mass galaxies, I quantify how changes in feedback parameters within the Auriga galaxy formation model influence the transport and distribution of metals in the CGM and alter its multi-phase structure. We find that the azimuthal distribution of metals is particularly sensitive to the metal loading of galactic winds.

I will show how varying key feedback parameters (such as mass loading, metal loading, and injection velocity of galactic winds) affects the star formation histories of galaxies, the relative fractions of cool and hot gas, and the spatial distribution of metals in the CGM. I also quantify how these variations map onto observable diagnostics, such as the column densities of metal ions.

By linking controlled variations in feedback models to CGM structure and observables and comparing with observations, this work provides a systematic framework to rule out certain regions of the feedback parameter space in simulations of galaxy formation.

The circumgalactic medium (CGM) is commonly described as a multi-phase system, in which cool (~10^4 K) clouds and filaments are embedded in a quasi-static, volume-filling hot halo at the virial temperature. This picture underpins emerging subgrid and two-fluid models which will potentially be incorporated into the next generation of galaxy formation simulations, and which independently evolve the cool and hot phases and their interaction.

I will present idealized calculations and cosmological simulations showing that this standard picture breaks down in many systems, particularly at the <~L* mass scale. In this regime, the CGM mass is dominated by the cool phase, while the hot phase is confined to transient and localized shocks. The CGM is thus more accurately described as an approximately isothermal, supersonically turbulent flow. Applying two-fluid models calibrated to the canonical multi-phase paradigm would therefore lead to systematically incorrect results in this regime.

I will also demonstrate how resolution limitations in cosmological simulations can be overcome in supersonically turbulent CGM, using scaling relations derived from simulations of isothermal turbulence.

In recent decades, the study of galaxy formation and evolution has progressed significantly, with numerical simulations now able to reproduce many observed properties of galaxies. However, the underlying feedback processes that regulate galaxy growth remain elusive, leading simulations to rely on weakly constrained subgrid models. We therefore present ARCHITECTS, a suite of radiation hydrodynamics zoom-in cosmological simulations designed to isolate and discriminate the impact of different feedback prescriptions on the circumgalactic medium (CGM). We demonstrate that even simulations produce comparable stellar masses, their CGM properties can diverge fundamentally, and be driven by distinct feedback modes. By generating synthetic quasar sightlines and comparing them to observations, we show that HI, MgII, CIV, and OVI covering fractions act as highly sensitive discriminators between subgrid models. Finally, we discuss the critical limits and necessary precautions required when performing direct comparisons between simulated and observed CGM data.

Cosmological simulations now routinely reproduce many properties of dwarf galaxies, including star formation rates that place them on the stellar-to-halo mass relation. However, this can be achieved through two distinct feedback modes. In ejective feedback models, highly mass-loaded winds allow a large fraction of baryons to cool and accrete, only to be ejected into the circumgalactic medium (CGM). In contrast, preventive feedback models feature high specific-energy outflows that efficiently transport energy – but not mass – into the CGM and beyond, heating the gas and suppressing further infall. In this work, we investigate the role of these feedback modes in a sample of ~1e10-1e11 Msun dwarf halos using cosmological zoom-in simulations. These simulations use adaptive mesh refinement to capture high-specific-energy outflows, together with an implementation of discrete supernovae (SNe). We show that episodic, SNe-driven shock heating sustains the inner CGM at approximately the virial temperature. This process also increases the ratio tcool/tff > 10 in the outer CGM and intergalactic medium (IGM), placing the gas in a radiatively stable regime. Warm outflows (> 3e5 K) dominate the energy budget, and their high specific energy allows them to penetrate the CGM, escape the halo, and heat the IGM. In contrast, cool outflows (< 3e5 K) dominate the mass budget and are largely recycled back into the interstellar medium (ISM), where they fuel future star formation. We identify a sharp transition at ~5 Gyr (z ~ 1) that marks a shift in the balance between ejective and preventive feedback. At early times (< 5 Gyr), satellite accretion of cold, dense gas drives CGM cooling rates comparable to the SNe energy injection rate, maintaining tcool/tff < 10 in a radiatively unstable regime where gas is rapidly recycled and ejected. At later times (> 5 Gyr), the CGM cooling rate declines and the CGM baryon fraction decreases, leading to a transition toward a preventive feedback mode in which SNe feedback maintains tcool/tff > 10. Finally, we discuss how this feedback transition relates to the closure radius and compare our results to recent observational constraints, including OVI absorption.

Resonant emission lines such as Lyman-alpha and Mg II are among the brightest tracers to directly map the circumgalactic medium (CGM), probing its cool hydrogen and metal-enriched gas. Their resonant nature causes numerous scatterings that reshape spectral profiles and surface brightness distributions and complicate interpretation. We post-process the AGORA CosmoRun cosmological zoom-in simulations of a 10^12 Msun halo, run with multiple simulation codes sharing common initial conditions and calibrated physical models, using the GPU-accelerated THOR Monte Carlo radiative transfer code to produce synthetic IFU datacubes. THOR's speed and flexibility across different data structures enables systematic RT post-processing across all participating codes, allowing us to assess how differences in the simulated multiphase CGM propagate into observables. We find that predicted surface brightness profiles and spectral morphologies are sensitive to the underlying CGM thermodynamics, kinematics, and metal distribution. At the same time, significant code-to-code variation persists, reflecting genuine differences in how each code shapes the multiphase gas around the galaxy. Understanding these code and model variations is critical for robustly interpreting CGM emission from resonant lines in current and upcoming IFU surveys.

We present recent development results from the new Moving Mesh code on the block, ShadowSWIFT. The next generation of ShadowFax, now in SWIFT. This talk will discuss preliminary results, as well as development progress and highlight the challenges of creating advanced hydro simulators.

In the Thesan-Zoom project, we study the formation and evolution of galaxies at high redshifts with zoom-in simulations conducted using the AREPO-RT code. The simulations resolve different phases of the ISM and follow local feedback from young stars, as well as multi-frequency radiative transfer.

We investigate how feedback regulates star formation, distributes metals and creates channels for ionizing photons to escape into the IGM. We compare our simulations to high-redshift galaxies observed with JWST, and discuss the impacts of bursty star formation on properties like the nitrogen-to-oxygen ratio, recently highlighted by JWST detections of nitrogen-enhanced galaxies.

Using Monte-Carlo post-processing radiative transfer, we study what galaxies, stars and phases in the star formation/feedback cycle contribute most to the escape of ionizing photons, and how this depends on resolution and implemented physics. We make predictions for the mean (volume-averaged) escaping ionizing emissivity as a function of redshift, and discuss what galaxies are predicted to drive cosmic reionization. We show that this is quite sensitive to the implemented feedback physics.

Understanding the formation and evolution of the earliest galaxies, and their role in driving cosmic reionization, remains a central challenge in modern cosmology. The physical properties of high-redshift (z ≳ 6–15) galaxies including their star formation efficiencies, feedback processes, ionizing photon production, and escape fractions, play a crucial role in shaping the ultraviolet luminosity function (UV LF) and the reionization history of the Universe. In particular, feedback-regulated star formation is expected to strongly influence the abundance of faint galaxies, while potentially contributing to the observed excess of bright sources at z > 10. Accurately modeling the spectral energy distributions (SEDs) of these early systems is essential for interpreting current and upcoming observations. Recent observations with the James Webb Space Telescope (JWST), particularly from the GLIMPSE survey spanning 6 < z < 16, have pushed the observational frontier to unprecedented depths, revealing galaxies as faint as MUV≈−12. These systems represent the lowest-mass, least luminous galaxies yet observed and serve as key laboratories for probing star formation and feedback in the early Universe. However, interpreting these observations remains challenging, as the mapping between intrinsic physical properties and observed quantities is highly non-linear and model-dependent. In this context, mock observations provide a critical bridge between theoretical models and observational data by forward-modeling the complex processes that shape observable signals.

To this end, I have implemented a pipeline to post-process the SPHINX radiation-hydrodynamic simulations with CLOUDY, enabling accurate predictions of intrinsic galaxy emission, including both stellar and nebular components from the ISM and CGM. I then use the RASCAS radiative transfer code to generate mock JWST observations that account for radiative transfer through a dusty ISM, as well as resonant line effects. Building on this framework, I am preparing a public catalog as part of the SPHINX data release, with a focus on extending toward faint galaxy populations.These mock datasets will include stellar continua, nebular emission, key recombination and metal lines, and dust attenuation effects, enabling direct comparison with JWST GLIMPSE observations.In this talk, I will present the methodology and highlight the challenges in producing robust and accurate predictions for mock observations. I will further demonstrate how such simulation-based mocks can be used to connect physical galaxy properties to observables, and, as a key science application, to investigate the role of rotational support in very high-redshift galaxies.

Reionisation marks the transformation of the intergalactic medium (IGM) from a cold, neutral state to an ionised one. This process unfolds in two main stages: an early phase of H I and He I reionisation, driven primarily by stellar sources, followed by He II reionisation, powered mainly by accreting black holes. Capturing this evolution self-consistently in simulations requires simultaneously resolving the ionising sources with a realistic galaxy formation model, following radiation transport on the fly, and using a large enough volume to encompass the largest ionised regions.

In this talk, I will present the first results from the Lumina simulation, which follows the evolution of the IGM through H I, He I, and He II reionisation while self-consistently modelling stellar, black-hole, and X-ray sources. Lumina combines the IllustrisTNG galaxy formation model with GPU-accelerated radiation transport in a cosmological volume 500 cMpc on a side. I will highlight results on the topology and cosmic variance of both hydrogen and helium reionisation, as well as on the statistical properties of the underlying source populations, showcasing the unique power of combining large volume with high resolution.

The synergy between JWST and ALMA has transformed our ability to probe star formation, allowing us to simultaneously trace unobscured and dust-obscured star formation across cosmic time. Nevertheless, converting luminosities into star formation rates (SFRs) remains challenging owing to observational limitations or uncertainties in calibration methods.

Using state-of-the-art hydrodynamical simulations NewHorizon and NewCluster, coupled with radiative transfer modeling using SKIRT, we generate synthetic multi-wavelength observables of z > 2 galaxies. This framework allows us to assess how dust attenuation, viewing geometry, star formation histories, and metallicity variations impact commonly used SFR tracers, from both infrared indicators (continuum and emission lines) and nebular emission lines, particularly Hα.

Within this broader context, we place particular emphasis on Hα as a key SFR tracer in the JWST era. While it has served as a powerful probe of instantaneous star formation in the local and low redshift Universe, we assess its calibration at high redshift, finding it sensitive to attenuation prescriptions, dust geometry, and the assumed conversion between luminosity and SFR.

Modern simulations of galaxy formation span large cosmological volumes and contain up to millions of galaxies. In order to fairly test and compare the properties of simulated galaxies with observations, mock images have to be produced using radiative transfer methods. Accurate radiative transfer codes, such as the SKIRT Monte Carlo code, cannot feasibly be run on the largest cosmological simulations, such as the ones recently produced as part of the COLIBRE project. I will present PARTRIDGE, a fast radiative transfer method implemented in Python, for use in the ultraviolet to near-infrared range (0.2-2 micron). It treats dust attenuation in an approximate, line-of-sight fashion and does not model dust emission. It does not include scattering explicitly, but uses attenuation coefficients derived from analytical calculations that approximate the first-order effects of scattering. I will present the inner workings of PARTRIDGE and results of comparisons with SKIRT, which demonstrate that despite its simplicity, PARTRIDGE predicts galaxy luminosities up to 10% different from SKIRT, with the disagreements typically much smaller. Other resolved metrics, such as galaxy sizes or pixel brightness histograms, are also very similar. Importantly, PARTRIDGE is 3-4 orders of magnitude faster, and can flexibly produce multiple types of images for a singe galaxy in one call, including 1) different orientations, 2) different telescope modes (filter choices), 3) dust vs. no dust, 3) nebular emission vs. no emission, and so forth. These considerations demonstrate that PARTRIDGE can be used on all snapshots of large cosmological simulations at a fraction of the cost that was needed to produce the snapshots in the first place, all with a relatively small price in accuracy.

Modern hydrodynamical galaxy simulations resolve the multi-phase interstellar medium in increasing detail, but converting their output into synthetic emission-line maps comparable to IFU observations remains a chronic bottleneck. We present DiffuseIonizedGasMix, a new photoionisation module for the Monte Carlo radiative transfer code SKIRT (Paper I submitted to A&A). It characterises the local ionising radiation field by its ionisation parameter and four spectral-shape ratios across 1–6 Ryd, interpolates temperatures and opacities in pre-computed Cloudy grids, and solves the full multi-element ionisation balance in-line to predict hydrogen recombination and forbidden metal lines, all within SKIRT's existing dust-aware iteration cycle. The module is validated against Cloudy across 60 spherical-shell H II region models spanning a wide range of photon rates, densities, and stellar ages, and is benchmarked on a Milky-Way-analogue galaxy against the dedicated photoionisation code COLT. We will discuss the approach, the validation, and how the module's co-location with SKIRT's dust radiative transfer enables self-consistent forward modelling of emission lines, dust attenuation, and dust re-emission in a single pipeline, directly applicable to the output of cosmological galaxy simulations.

Mock observations from simulations have become essential for our community in order to (1) assess quantitatively how our ideas about galaxy formation are consistent with observations, and (2) interpret observational results and understand their possible biases.

I will present the new version of RASCAS (RAdiative SCAttering in Simulations). RASCAS is a publicly available and actively developed community code to construct mock observations from astrophysical simulations. It performs Monte Carlo radiative transfer on adaptive meshes with an octree structure, and features full MPI parallelisation with perfect scaling through flexible domain decomposition and adaptive load balancing. While initially developed for resonant line transfer (Lyman-alpha in particular) through a mix of hydrogen and dust, RASCAS can also propagate arbitrary SEDs from any collection of sources — star particles, gas cells, or user-defined emitters.

The new release introduces a fully redesigned, object-oriented architecture that enables (1) straightforward integration of new simulation formats, and (2) a flexible framework to define and combine radiation-matter interactions, without touching the core code. The interactions shipped with the public version of the code include resonant lines (possibly with fluorescent decay channels), photo-ionisation (of H, He, and He+), scattering and absorption by dust. This new design lowers the barrier to community-driven contributions and extensions, and preserves the HPC efficiency of the code. RASCAS v2 also incorporates lessons learned from a few years of various users generating mock spectra (incl. stellar continuum, emission lines, absorption lines), mock images, mock data cubes, and quasar absorption line spectra, and offers a user-friendly and efficient python workflow. I will present the main features and design principles of this new version, and illustrate its capabilities through selected science examples.

As next generation ground and space based telescopes come online we are expecting to discover a wealth of faint stellar systems where the overlap between what has dark matter (ie dwarf galaxies) and what doesn’t (ie globular clusters) becomes more apparent. Within the context of simulations, collisionless cosmological simulations have allowed the emergence of collisional systems such as globular clusters within that of dwarf galaxies. This presents a unique opportunity to resimulate these collisional systems within their hosts density field using NBODY6DF, which is capable of resolving the two body dynamics important to that of a star clusters long term evolution.

Using the naturally forming star clusters within the high resolution EDGE simulations of dwarf galaxies, I show how low mass star clusters tidally evolve within their hosts potential field and naturally populate the unclassified region which lies between ancient dark matter free globular clusters and that of dark matter rich dwarf galaxies. I will compare the results to recent observations, drawing similarities between that of evolved star clusters from the simulations and interesting objects such as Unions I/Ursa Major III in which the community is currently debating about

Simulating galaxy formation requires bridging an enormous range of spatial scales, connecting the large-scale cosmological environment down to the parsec-scale multi-phase interstellar medium (ISM). Star clusters play a central role in this process: they regulate the ISM, dominate the ionising photon budget, and may provide massive black hole seeds — yet they remain notoriously difficult to resolve in cosmological volumes. To overcome this limitation, we present a novel framework implemented in RAMSES-RT that embeds sink-particle-based star cluster formation within radiation-hydrodynamics simulations. This approach enables the formation and evolution of individual star clusters directly from the multi-phase ISM in a fully cosmological context.

In our model, each sink particle represents a forming star cluster whose growth is tied directly to the local gas inflow rate. We track sub-grid stellar populations with self-consistent ages and metallicities, allowing multi-wavelength radiative and supernova feedback to regulate the surrounding gas without relying on artificially imposed efficiency parameters. Applied to a zoom-in simulation of a dark matter halo reaching $10^{10}M_\odot$ at $z = 6$, we perform convergence tests across three spatial resolutions, reaching a finest cell size of 2.8 pc. Our simulations (Thompson et al., in prep.) reveal two key results:

- The Necessity of Radiative Processes: By comparing control runs with and without radiative feedback, we show that the initial cluster mass function and ionising escape fractions are highly sensitive to pre-supernova radiation. This underscores the critical need for explicit radiative transfer when modelling clustered star formation in early environments. - Hierarchical Assembly from Clusters to Galactic Nuclei: We track the tidal evolution, trajectories, and mergers of individual clusters, explicitly linking clustered star formation to the hierarchical assembly of nuclear star clusters and the potential delivery of intermediate-mass black-hole (IMBH) seeds to galactic centres.

By connecting resolved, parsec-scale star cluster formation and explicit radiative feedback directly to galaxy-scale observables, this framework provides a self-consistent, predictive picture of early galaxy evolution — from the birth of the first clusters to the build-up of galactic nuclei.

JWST provides an unprecedented window into cosmological and galaxy formation processes thanks to its observational view of the first galaxies in the Universe. Many observed systems display high stellar surface densities compared to typical conditions in lower redshift star clusters. I will present results from high resolution cosmological hydrodynamics simulations in AREPO that are designed to resolve a large statistical sample of early star clusters–the building blocks of JWST’s highest redshift galaxies. We investigate the dynamical properties and environment of star clusters at cosmic dawn from both a numerical and analytical perspective. Using a methodology that leverages structure finding algorithms to mimic observational techniques, we find that the simulations are consistent with observed star clusters with high surface density in a feedback-free case. This suggests that these high redshift clusters may have formed in an environment where feedback processes were weaker. I will discuss these results and their implications, including how JWST can now probe an era where star clusters preserve dynamical signatures of the process of hierarchical galaxy formation. I will also describe how we can couple these results with analytic modeling to understand the continued dynamical evolution of proto-globular clusters and nuclear star clusters.

Large-scale cosmological simulations have entered a powerfully predictive regime and provide quantitative insights into the formation and evolution of galaxies. However, they face two fundamental and related limitations: (i) low numerical resolution and (ii) strong simplifications for the physics of star formation, stellar feedback, the small-scale ISM, and supermassive black holes.

In this talk I will look towards the future of high-resolution galaxy formation simulations. I will present our efforts to simulate galaxies in the full cosmological context with a new model that extends Imladris -- multi-phase gas with non-equilibrium primordial chemistry, the formation and evolution of individual stars, individually resolved stellar winds and supernova, stellar radiation i.e. photoionization feedback and photoelectric heating -- into the cosmological context, including the formation, resolved dynamics, and thermal feedback from AGN. We are realizing this model in a series of cosmological galaxy simulations run at ~3 solar mass resolution (sub-pc spatial resolution) for the first billion years of the evolution of the Universe (to z=5.5). I will discuss early science results on the formation of very dense and compact stellar systems, connections to JWST, and comment on the challenges and future role of these single-star, resolved ISM type models for studying galaxy formation.

In the last decade, simulations of galaxy formation such as IllustrisTNG, EAGLE, and Simba have provided statistical predictions of galaxy observables across cosmic time. However, their crude assumptions about the state of the interstellar medium have inhibited a full understanding of the complex physics that shape galaxy evolution. Concurrently, JWST observations have enabled the inference of star-formation histories and kinematics of early galaxies (~10^8 Msun) at z>6. Investigating the baryonic processes affecting these early galaxies thus requires detailed prescriptions for star formation and stellar feedback in galaxy formation models, beyond those employed in large-scale cosmological simulations. To this end, we present the first results from a suite of cosmological zoom-in simulations (Nelson et al. in prep) employing the new multiphase ISM model "Imladris" (Smith 2026) motivated by high-resolution idealised simulations. Our model explicitly tracks the spatial clustering of star formation and feedback from individual massive stars by self-consistently tracking supernovae, stellar winds and radiation. We further include a variety of heating and cooling channels coupled with non-equilibrium chemistry, thereby tracking multiphase gas down to 10 K.

I study our pathfinder runs, where we simulate galaxies of 10^6-10^9 Msun in stellar mass (z=6) at a resolution of 3-24 Msun. The interstellar medium of these galaxies is dominated by the warm (<8000 K) phase in mass, albeit with strong temporal variations. Leveraging our sub-parsec-scale resolution, I examine the formation of gaseous disks in the early Universe and find that the rotational support of early low-mass galaxies is a transient feature closely linked to star formation. I further link the high stochasticity of the sizes and masses of gas and stellar components to observables of gas kinematics, such as v/sigma, typically derived from HII region emission lines. I thus provide physical insights into the assembly of galaxy disks at highly resolved spatial and temporal scales in the early Universe, lying within reach of cutting-edge observational campaigns.

We present a suite of cosmological zoom-in simulations of 11 isolated dwarf galaxies, targeting halos of ~10^9 Msun, using the radiation-magnetohydrodynamic code AREPO-RT. The objects are selected from the THESAN simulation and re-simulated with the zoom-in technique at a resolution of 17 Msun, employing the RIGEL galaxy formation model. RIGEL self-consistently models the effects of stellar feedback from individual massive stars in the interstellar medium of dwarf galaxies. Each simulated galaxy experiences its unique reionization history. The combination of this inhomogeneous reionization background with the halo growth history results in the observed diversity of these ultra-faint reionization relics in terms of stellar mass, size, and morphology. Remarkably, one galaxy exhibits a compact morphology, with more than 60% of its stellar mass contributed by a nuclear star cluster (NSC). We find that this NSC originates from an intense starburst at z~10, which formed 10^6 M☉ of stars within 10 Myr. During this burst, a 9×10^5 Msun cluster with a half-mass radius of 3.1 pc emerges, strikingly similar in mass, size, and redshift to the massive cluster observed in the "Cosmic Gems" at z~10. We discuss how external (reionization-driven) and local (stellar radiative) feedback, combined with halo growth history, shape the properties of these dwarf galaxies. We also examine the importance of local radiative feedback in this cosmological context and find that its absence significantly increases stellar mass and clustering while leaving the sizes of dwarf galaxies unchanged. Our results highlight the need for high-resolution, RMHD simulations to correctly capture the interplay between reionization, feedback, and star cluster formation in the early Universe.

Galaxy mergers can lead to some of the most extreme bursts of star-formation in the Universe, in addition to providing sites for star-formation outside of the main bodies of galaxies via gas bridges between merging galaxies. These processes have traditionally been modeled at either low-resolution and/or with simplified physical assumptions, hindering the trustworthiness of these simulations. Here I showcase early results based on the Imladris model, which includes (but is not limited to) gas heating and cooling down to ~10 K, non-equilibrium chemistry, and feedback — both pre- and supernova feedback — from individual stars (Smith 2026, submitted). I mainly focus on results from the upcoming suite of cosmological zoom-in simulations targeting a diversity of galaxies during the first billion years of cosmic time (Mhalo(z=5.5) ~ 10^8-11 Msun) with baryon resolution down to 3 Msun. I quantify the properties and conditions of star-formation to compare during mergers and fly-bys versus periods of relative isolation. These major mergers can lead to temporarily enhanced periods of star-formation, and even star-formation occurring in the connecting bridges of gas between the merging galaxies. I then speculate on potential observational signatures of these high-redshift mergers, such as via the early formation of a stellar halo and star clusters. Lastly, I connect these results with idealized wind-tunnel setups, also with the Imladris model, which focus on the ability of ram-pressure-stripped gas to remain cool and form stars. This is the first such setup to include non-equilibrium chemistry, molecular gas, and feedback from individual stars at ~4-20 Msun baryon resolution, providing a useful connection between galaxy and cloud-crushing scales.

Star clusters (SCs) are a crucial component of galaxy formation; they are the result of star formation in environments of high density and pressure, typically at high redshift where such conditions are more common. Cosmological simulations that incorporate the formation, evolution and disruption of these SCs offer a means to follow the co-evolution of SCs and their host galaxies, connecting the SCs forming at high z to their surviving descendant globular clusters (GCs). They also offer the means to address many outstanding questions regarding the prevalence of SC formation at high redshift, the connection between the assembly histories of galaxies and their GC populations, and the contribution of disrupted SC mass to galaxies, their nuclear star clusters, and their stellar haloes and streams. Unfortunately, simulations with sufficient volume to host representative galaxy populations lack the resolution to self-consistently model the formation of SCs, and so a common approach is to associate SC populations to stellar particles in these simulations through use of a subgrid model.

In my talk I will present our implementation of such a model into the COLIBRE simulations of galaxy formation. Our model inherits many features of the successful E-MOSAICS project, and includes several key advancements to the SC formation and disruption physics, plus a prescription for SC size evolution. COLIBRE's modelling of cold gas down to temperatures of 10 K yield a more realistic ISM structure than was possible in E-MOSAICS, with most of the mass in dense clumps with a low volume-filling fraction. The environments in which SCs form are therefore more realistic, and the stronger tidal forces present in a "clumpy" ISM are more effective at disrupting SCs, alleviating a key weakness of E-MOSAICS. I will present some key results from cosmological simulations incorporating this model, and discuss some key limitations of our purely subgrid approach to SC formation and evolution, and how we are working to overcome them.

The rapid emergence of massive black holes in the early Universe remains a major challenge for models of black hole formation and growth. Recent JWST observations, including compact AGN-hosting systems such as “little red dots,” place new constraints on the early co-evolution of black holes and galaxies, particularly at high redshift where accretion, feedback, and environmental effects remain poorly understood. In this talk, I will present results from the SEEDZ simulation suite, designed to investigate early black hole growth across a range of spatial resolutions and physical models. We begin with controlled isolated galaxy simulations to validate our subgrid prescriptions, demonstrating that sustained hyper-Eddington accretion can occur in dense natal environments where supernova feedback fails to efficiently suppress gas inflow. In some cases, supernova-driven shocks instead enhance local overdensities, further facilitating black hole growth. We then extend this analysis to cosmological zoom-in simulations of an overdense 1 Mpc region. While lower-resolution runs exhibit minimal growth, higher-resolution simulations produce rapid early black hole growth, with multiple seeds reaching ∼10⁴ M⊙ by z ≳ 21. These results highlight the critical role of resolution in capturing dense gas structures and accurately modeling early accretion physics. We build on the results of these advanced simulations on relatively lower-resolution simulations with refined physical models to better connect with JWST observations. These simulations form compact, AGN-hosting galaxies whose sizes, stellar masses (∼10⁷ M⊙), and black hole masses (≳10⁶ M⊙ by z ≈ 10) consistent with the properties of recently observed compact high-redshift AGN. The systems naturally exhibit elevated black hole-to-stellar mass ratios. Overall, our results emphasize that early black hole growth is strongly regulated by resolution, dense gas structure, and inefficient feedback in high-redshift environments. This work provides a physically consistent framework for interpreting JWST observations of compact AGN and offers predictions for future high-redshift black hole populations detectable with facilities such as LISA.

The entropy-profile problem has been a persistent issue in modern studies of simulated galaxy groups and clusters. The presence of an entropy plateau and higher entropy gas in the central regions of halos is not a feature seen in X-ray observations. Why such a characteristic attribute is present in simulations may have many culprits such as missing physics or low resolution. This mismatch is exacerbated at low masses, where galaxy groups show the largest deviations from observations. While recent advancements in AGN feedback modelling (e.g. Hǔsko et al. 2024) have shown promise in preserving cool cores, such models typically require resolutions that are not achievable in large cosmological volumes.

A key limitation here lies in the under-resolved environment surrounding black holes. In such conditions, kinetic jets may be unable to effectively tunnel through the surrounding gas, instead depositing their energy closer to the black hole and producing a more isothermal-like result (leading to the undesired increased entropy). We therefore propose a focused approach to test whether the entropy plateau can be mitigated through improved local resolution.

Recent developments in astrophysical codes have seen a surge in adaptive particle refinement methods, with both PHANTOM (Price et al. 2018) and GIZMO (Hopkins 2014) now including implementations of this approach. To this end, we introduce a novel particle splitting routine in the SWIFT SPH code that can locally enhance resolution on-the-fly. We test this method within the context of improving physics around black holes in simulations of isolated groups and clusters (previously studied in Nobels et al. (2022) and Hǔsko et al. (2023)).

We therefore analyse how black hole evolution and hot gas profiles compare between simulations with and without particle splitting and at various mass scales. We focus on whether local refinement can mitigate the issues brought about by resolution degradation, aiming to evaluate whether it can become a potential strategy for improving cosmological simulations.

Arkenstone BH (ArkBH) is an extension of the Arkenstone framework, originally developed to resolve high specific energy stellar feedback driven outflows in cosmological simulations. ArkBH extends this approach to black hole feedback by providing a framework for launching and resolving high specific energy black hole driven outflows, particularly relevant to jet mode feedback. We demonstrate that ArkBH successfully quenches galaxies in both isolated and cosmological zoom-in simulations by counteracting gas inflow from the circumgalactic medium into the interstellar medium. Importantly, the model also avoids local self regulation of black hole accretion and feedback, a common limitation of many existing subgrid prescriptions. This sets the stage for our current work implementing the model in larger cosmological boxes. This will enable robust statistical studies of the importance of preventive feedback in galaxy evolution and allows us to better constrain the effects of varying outflow velocities, energies, and mass loadings. Lastly, these cosmological runs will allow us to compare our results against key observational constraints, including the kinematic Sunyaev-Zel’dovich effect and intracluster medium velocity dispersions.

Modeling the formation and evolution of supermassive black holes (SMBH) remains one of the largest unresolved uncertainties in cosmological simulations, with different black hole (BH) seeding, accretion, dynamics and feedback prescriptions leading to widely varying predictions for BH demographics and coevolution with galaxies. Because these processes occur below the resolution limit of current simulations, they must be implemented through subgrid models whose assumptions can significantly impact emergent galaxy–BH scaling relations. Here, I use the recently developed BRAHMA cosmological simulations, built on the IllustrisTNG model, to investigate the impact of different BH seeding prescriptions on the M-σ relation. These simulations adopt heavy (~10^5 Msun) seeds and systematically varied BH seeding models motivated by direct-collapse BH formation scenarios. I show that these different seed models lead to different normalizations of the M-σ relation at higher redshifts (z > 2) across all σ, and at low redshift for systems with low σ (50 km/s < σ < 80 km/s). The most lenient seed model also shows negligible evolution in the M-σ relation across redshift, while more restrictive models have substantially lower normalization on the M-σ relation for high σ (~100 km/s) at high redshifts and evolve upward toward the local relation. Interestingly, current estimates for BH masses and stellar velocity dispersions of high-z JWST AGN suggest a high-z M-σ relation that is consistent with the local relation, similar to our lenient seed model, though current observational uncertainties remain large. More broadly, this work demonstrates that SMBH seeding is a key driver of variability in galaxy–BH scaling relations and highlights the need for improved physical modeling and observational constraints to inform next-generation cosmological simulations.

Dwarf galaxies are ideal laboratories to probe the interplay between galaxy formation and the growth of black holes (BHs) in the early Universe. Mounting observational evidence reveals the presence of BHs in low-mass galaxies across cosmic time, with JWST uncovering a likely population of overmassive BHs at 2≲z≲11. Simulations struggle to reproduce this high-redshift regime, motivating revisions to models of BH accretion and feedback from active galactic nuclei (AGN). To address this, we present high-resolution cosmological zoom-in simulations of a dwarf galaxy based on FABLE physics, introducing novel sink-based BH accretion models and relaxing the fiducial assumption of strong supernova feedback. BHs accrete more efficiently in the sink-based runs compared to the `traditional' Bondi-based counterparts, with AGN feedback leading to early, rapid quenching maintained by fast, hot and metal-enriched outflows. These outflows pollute the outer circumgalactic medium, yielding flat metallicity gradients down to z=0. We further assess the performance of two widely used virial estimators and find significant departures from the true dynamical mass, especially during the high-redshift dwarf assembly. Since our galaxy is dark-matter-dominated at all times and radii, BH growth, tied to the baryon cycle, shows no clear correlation with global dynamical properties. Efficient AGN feedback, produced by overmassive BHs relative to extrapolated local black hole mass–stellar mass relations, indicates that dormant BHs residing in local, quenched dwarfs might be the relics of some of the high-redshift JWST BHs.

Recent JWST observations indicate that massive quenched galaxies are more abundant than predicted by most existing galaxy formation simulations and semi-analytic models. Remarkably, the COLIBRE simulations of galaxy formation have succeeded in resolving this tension, although the precise reason for their improved agreement with the JWST data remains unclear. In this talk, I demonstrate that this agreement is largely due to super-Eddington growth of supermassive black holes (BHs) at high redshift. By running and analysing a series of (100 comoving Mpc)^3 simulations with the COLIBRE subgrid physics, varying the maximum allowed BH accretion rate in units of the Eddington rate, I show that only the fiducial COLIBRE model, which permits super-Eddington accretion events, is consistent with the latest JWST constraints at z > 6. Moreover, I demonstrate that in COLIBRE, about 50 per cent of BH mass growth at high redshift occurs in the super-Eddington regime, even though such episodes are extremely rare in time. This work highlights the important role of super-Eddington BH accretion in simulations of galaxy formation for reproducing the early emergence of massive quenched galaxies in the real Universe.

Over the last decade, there has been substantial progress in simulating supermassive black holes (SMBHs) in a cosmological context, using a range of subgrid models to capture the complex and uncertain physics of black hole growth. These include treatments such as Bondi-like accretion, repositioning, and feedback, often calibrated to reproduce key observational constraints. However, accurately capturing the galactic environment is also essential as gas supply and replenishment regulate the fueling of active galactic nuclei. This is particularly relevant toward higher redshifts, where galaxies are denser, clumpier, and more gas-rich. In this talk, we present first results from a new suite of cosmological zoom-in simulations of galaxies evolved to z ≈ 5.5 from Heidelberg: these incorporate a model for the multiphase interstellar medium down to 10 K, a single-star treatment of stellar evolution and feedback, as well as black holes seeded at ~1000 solar masses. We show that, when adopting subgrid Bondi-like prescriptions similar to those used in large-volume cosmological simulations, the growth of massive black holes at early times is limited. Using these simulations as a framework, we systematically test alternative seeding, positioning, and accretion formulas in order to constrain the physical conditions required for rapid SMBH growth. Building on this, we make predictions on the link between the larger galactic environment and the central regions in setting black hole growth rates. The goal is to explore a pathway to studying black hole–galaxy co-evolution in the first billion years and inform future accretion and feedback models for simulations.

Recent cosmological simulations have begun modelling the dynamics of supermassive black holes (SMBHs) using dynamical friction subgrid models, instead of BH repositioning. These simulations predict a substantial population of wandering black holes within massive galaxies, with some systems containing over a thousand such objects. We present results from a series of cosmological zoom-in simulations that use different halo mass thresholds for black hole seeding, and investigate how the population of black holes is affected. Specifically, we use the seeding criteria from the ASTRID, EAGLE and IllustrisTNG simulations.

Our simulations are run with a new version of the KETJU code, which combines regularised integration of sufficiently massive SMBHs with a dynamical friction subgrid model for lower-mass SMBHs. This allows us to track SMBH binaries all the way down to the separations where gravitational wave emission leads to coalescence. We find that the choice of seeding threshold has a negligible effect on central SMBHs, all of which consistently follow the observed SMBH mass- stellar mass relations. However, the seeding criteria have a profound impact on the wandering population of SMBHs. In the most massive galaxies, the number of wandering black holes varies by an order of magnitude depending on the seeding criteria.

The wandering SMBHs generally remain near their initial seed mass and exhibit low Eddington accretion rates. We also find notable differences in black hole accretion rate densities at high redshifts and in the occupation fractions of low-mass systems across all redshifts, both of which could potentially be used to observationally constrain seeding models. Finally, we demonstrate that the frequency of SMBH mergers changes significantly with the seeding prescription, highlighting the necessity of accurate seeding models in providing reliable merger-rate predictions for the upcoming LISA mission.

We present a new suite of SWIFT zoom simulations selected from a suite of large volume parent simulations, designed for simultaneously exploring the wide-field regime over an unprecedented dynamic range of halo masses. We illustrate how these zooms, combined with modern neural density estimation methods, can be used to build emulators that can predict one point and higher order statistics in Gigaparsec effective volumes, whilst simultaneously varying the underlying cosmological and astrophysical model parameters. We show early results for halo property and spatial distributions, as well as galaxy predictions from the SHARK semi-analytic mode, and demonstrate the application of the conditional emulators within a simulation based inference framework. Finally, we make predictions for upcoming wide field surveys on Euclid, Roman and Rubin, and present ongoing work utilising full hydrodynamic zooms.

We introduce a new suite of 1,024 cosmological hydrodynamical zoom-in simulations of Milky Way-mass halos from the DREAMS Project, designed to systematically disentangle theoretical uncertainties in galaxy formation physics from intrinsic halo-to-halo variance [arXiv:2512.00148]. By varying key astrophysical parameters governing supernova wind energy, wind speed, and AGN feedback efficiency within the IllustrisTNG model, alongside cosmological parameters, we explore a vast parameter space of galaxy formation scenarios.

To overcome the computational bottleneck of evaluating this high-dimensional space, we utilize a hierarchical generative machine learning framework, incorporating conditional normalizing flows and Variational Diffusion Models, to accurately emulate both central host properties and variable-length satellite populations [arXiv:2409.02980]. By introducing a novel observational weighting scheme constrained by the empirical stellar mass-halo mass relation, we reveal broad degeneracies in the fiducial feedback parameters, demonstrating that single-model tuning misses complex parameter interdependencies [arXiv:2602.03613].

Applying these pseudo-posterior constraints, we assess the impact of feedback variations versus accretion history. For central galaxies, we demonstrate that a Gaia-Sausage-Enceladus-like merger history is exceptionally rare (1.5% occurrence) and drives specific structural shifts, such as disk compaction, though immense halo-to-halo scatter persists. For satellites, we show that while sub-grid parameter variations strongly regulate stellar mass, intrinsic variance overwhelmingly dominates satellite population statistics [arXiv:2512.02095]. Finally, we highlight a persistent tension where the current sub-grid model fails to reproduce the extended half-light radii of massive satellites ($>10^9 M_\odot$) observed in the SAGA survey, underscoring the need for more bursty, multiphase interstellar medium prescriptions in future galaxy formation models.

To determine if this size-mass tension is an artifact of the TNG sub-grid model or a broader theoretical challenge, we will also present preliminary results from ongoing extensions to the DREAMS project framework. This includes three new suites of high-resolution dwarf galaxy simulations implemented across distinct hydrodynamical codes—FIRE3, RAMSES, and ChaNGa. Furthermore, we will introduce new mass-varied and resolution-varied simulation suites utilizing the IllustrisTNG and FIRE3 models. These suites extend the halo mass range to include $10^9$ to $10^{14} M_\odot$, allowing us to recreate uniform box results from a suite of zoom-in simulations, creating a new way to understand parameter and resolution variations across a full range of halo masses.

Low mass galaxies (10^6 < M_* /M⊙ < 10^9) serve as powerful laboratories for probing both dark matter physics and the feedback processes that regulate galaxy evolution. As the most dark matter-dominated systems known, they present a longstanding challenge to ΛCDM predictions: observations favor shallow central density cores, while dark matter–only simulations produce cuspy profiles. This “core–cusp” discrepancy may arise from either the microphysics of dark matter or the effects of baryonic feedback, but these scenarios can produce degenerate observational signatures, complicating efforts to distinguish between them.

At the same time, state-of-the-art galaxy formation simulations rely on subgrid prescriptions to model unresolved processes such as star formation and feedback, introducing tunable parameters that shape galaxy evolution in nonlinear ways. In this work, we present a systematic investigation of these uncertainties using a suite of 500 high-resolution hydrodynamical simulations designed to isolate the impact of varying feedback prescriptions. Our simulations achieve a spatial resolution of ~350 pc and span a controlled parameter space in key processes, including star formation efficiency and energy injection.

We analyze the resulting galaxies using a set of complementary diagnostics: the stellar mass–halo mass (SMHM) relation, cumulative star formation histories, simple star formation timescales, and radial dark matter density profiles. Together, these probes provide a cohesive framework for linking how stars form over time to changes in the inner structure of dark matter halos.

By exploring a broad and controlled parameter space, this work provides a more nuanced view of how baryonic processes shape low-mass galaxies and their halos. These results place new constraints on subgrid feedback models and highlight the importance of using multiple diagnostics when interpreting both simulations and observations.

For decades, it has been hypothesized that stellar feedback drives large-scale changes in the history, composition, and kinematics of galaxies from the scale of dwarfs up to disks like the Milky Way. Numerous pieces of observational evidence support this view: the rotation curves of dwarves, the relative deficit of baryons and metals in disc galaxies, and the metal-polluted CGM visible in UV absorption and X-ray emission. Constraining the details of how stellar feedback works on the scale of galaxies, however, has been an arduous process that is still ongoing. In this talk, I will present new results examining how we can design simulation studies to efficiently explore the uncertain parameter space of small-scale stellar physics using statistical and machine-learning approaches to more reliably compare our simulations to observations, quantify the uncertainty in those simulations, and to maximize the scientific return-on-investment for large simulation campaigns. I will focus on results applying these concepts to understand the sensitivity of dwarf galaxy kinematics, formation history, and morphology to differences in feedback and star formation physics.

Population III stars seeded the chemical enrichment of our Universe during the epoch of reionization, yet to this day their detection remains elusive. In this talk, I present new results from high-resolution (~pc) cosmological hydrodynamical simulations of an ultra-faint dwarf galaxy evolved to redshift z=0 with a diversity of stellar evolution and feedback models (with/without Pop.III and/or Pop.II radiative feedback). Leveraging these models allows us to connect the impact of each model onto present-day observables. Comparing between models, I will present results systematically quantifying how Pop.III feedback impacts the properties of the ultra-faint dwarf, from stellar mass to, notably, stellar abundances of metal-poor stars. Through careful tracking of metal ejecta, I will show which z=0 metal-poor stars are true descendants of Pop.III explosions, and where to find them, once inhomogeneous chemical enrichment from cosmological formation is taken into account. Finally, I will show preliminary results of a new numerical implementation combining star-by-star modelling for both Pop.III and Pop.II stars, enabling a more robust link for how individual high-redshift explosions set the abundances of individual low-mass stars at z=0.

Ultra-faint dwarf galaxies are predicted to be the records of the conditions of early, high-redshift universe. Here, we present the properties of a cosmological suite of simulated ultra faint dwarf galaxies, at an unprecedented resolution of $4M_\odot$, using the hydrodynamical LYRA galaxy formation model. The dwarfs in the suite range from $M_ \star \approx 100 M_\odot$ to $\approx 10^6 M_\odot $. We contrast the properties of these dwarfs in two different early Lyman-Werner backgrounds (LW) radiations: (i) negligible LW prior to $z=10$, (ii) a stronger, redshift-dependent LW prior to z=10. We find that the stellar mass-halo mass relation and occupation fraction in these two models show stark differences in halo mass range of $M_{halo} < 10^8 M_\odot$, shifting the occupation by more than an order of magnitude. This results in large differences in the in-situ vs ex-situ fraction of stars, a topic hotly debated these days given the observations of dwarf galaxies around the Milky Way. Finally we present the mass-metallicity relations (MZR) and mass-size relations of the simulated dwarfs. We find that the weaker LW model, results in lower metallicites at fixed stellar mass, though with a large scatter. Interestingly, the stronger LW naturally reproduces the observed flatenning of the MZR relation at low stellar masses.

Ultra-faint dwarf galaxies (UFDs) are the faintest objects formed from primordial density perturbations. As such, they provide a powerful laboratory for probing the nature of dark matter on very small scales. These systems probe physical regimes beyond those typically accessed by standard methods, such as the Lyman-alpha forest, and therefore offer a unique opportunity to place new constraints on the properties of dark matter.

Beyond variations in their size, UFDs also exhibit a diversity of morphologies, in particular in their ellipticity. Some systems, such as Boötes I and Hercules, have ellipticities exceeding 0.6. This elongation is often attributed to tidal stripping due to interactions with the Milky Way. However, using well-controlled numerical simulations of UFDs on orbits constrained by Gaia data, I will demonstrate that, within a ΛCDM framework, this claim is not well supported. On the contrary, I will show, using very high-resolution cosmological zoom-in simulations, that such ellipticities naturally emerge from the complex assembly history predicted by ΛCDM. However, this result comes with a caveat: this assembly history, together with continuous perturbations from subhaloes, imposes a minimum size that is larger than that observed for some UFDs. To mitigate the continuous perturbations from these mini-haloes, which lead to kinematic heating of UFDs, I will present simulations assuming warm dark matter (WDM). I will show that, although the final morphology of UFDs is strongly affected by WDM and the abundance of mini-haloes is suppressed, the observed dynamical heating is still present.

This unexpected result, confirmed by idealized simulations, suggests the presence of spurious numerical heating and raises concerns about the ability of standard gravity solvers to accurately capture the dynamics of the inner regions of dark matter haloes, and, consequently, the sizes of UFDs.

In this talk, I will briefly introduce a new self-interacting dark matter module developed for the cosmological multi-physics simulation code Arepo-2. Compared with the previous implementation, this module is more efficient and versatile, featuring innovative communication and neighbor-search schemes. It allows users to switch between different numerical treatments of SIDM and enables simulations of more complex SIDM physics, including velocity-dependent scattering, anisotropic scattering, dissipative interactions, multi-state SIDM, and many more. I will conclude the talk by outlining several important new simulation suites that can now be carried out efficiently with this module.

Understanding the interplay between dark matter physics and baryonic processes is critical for identifying the nature of dark matter. Self-interacting dark matter (SIDM), where dark matter particles can scatter with each other, is well motivated by theoretical models and has become an increasingly common module within galaxy formation codes. Baryonic effects are often degenerate with dark matter physics, so a full picture is possible only through simulations that span a range of galaxy formation models and underlying particle physics. I will present an updated implementation of SIDM in the Gizmo code that allows for a fully customizable velocity dependence in the interaction cross section as well as anisotropy in the scattering angle. I will show preliminary results using the FIRE galaxy formation model that demonstrate the reliability of the code and the differences from existing models with constant cross sections and isotropic scattering. This code will provide a better description of the effects of SIDM in merging and infalling halos, which is essential to properly model and constrain the SIDM parameter space. In addition, it will add to the existing range of alternative dark matter models implemented in Gizmo for consistent model comparisons within the same galaxy formation model. Building upon my previous simulations using Arepo, this implementation will also allow, for the first time, comparisons of the same SIDM models between drastically different galaxy formation codes.