X-ray stellar feedback in low-metallicity starbursts

HMXB, ULX and star formation

Extensive astronomical surveys, like those conducted with the Chandra X-ray Observatory, detect hundreds of thousands of unidentified cosmic sources. Machine learning (ML) methods offer an efficient, probabilistic approach to classifying them, which can be useful for making discoveries and conducting deeper studies. In earlier work, we applied the LightGBM (ML model) to classify 277,069 Chandra point sources into eight categories: active galactic nuclei (AGNs), X-ray emitting stars, young stellar objects (YSO), high-mass X-ray binaries, low-mass X-ray binaries, ultraluminous X-ray sources, cataclysmic variables, and pulsars. In this work, we present the classification table of 54,770 robustly classified sources (over 3 σ confidence), including 14,066 sources at >4 σ significance. To ensure classification reliability and gain a deeper insight, we investigate the multiwavelength feature relationships learned by the LightGBM model, focusing on AGNs, stars, and YSOs. We employ explainable artificial intelligence (XAI) techniques, specifically, Shapley Additive Explanations, to quantify the contribution of individual features and their interactions to the predicted classification probabilities. Among other things, we find infrared-optical and X-ray decision boundaries for separating AGN/stars, and infrared-X-ray boundaries for YSOs. These results are crucial for estimating object classes even with limited multiwavelength data. This study represents one of the earliest applications of XAI to large-scale astronomical data sets, demonstrating ML models’ potential for uncovering physically meaningful patterns in data in addition to classification. Finally, our publicly available, extensive, and interactive catalog will be helpful to explore the contributions of features and their combinations in greater detail in the future.

This presentation reviews Chandra's major contribution to the understanding of nearby galaxies. After a brief summary on significant advances in characterizing various types of discrete x-ray sources, the presentation focuses on the global hot gas in and around galaxies, especially normal ones like our own. The hot gas is a product of stellar and active galactic nuclear feedback--the least understood part in theories of galaxy formation and evolution. Chandra observations have led to the first characterization of the spatial, thermal, chemical, and kinetic properties of the gas in our galaxy. The gas is concentrated around the galactic bulge and disk on scales of a few kiloparsec. The column density of chemically enriched hot gas on larger scales is at least an order magnitude smaller, indicating that it may not account for the bulk of the missing baryon matter predicted for the galactic halo according to the standard cosmology. Similar results have also been obtained for other nearby galaxies. The x-ray emission from hot gas is well correlated with the star formation rate and stellar mass, indicating that the heating is primarily due to the stellar feedback. However, the observed x-ray luminosity of the gas is typically less than a few percent of the feedback energy. Thus the bulk of the feedback (including injected heavy elements) is likely lost in galaxy-wide outflows. The results are compared with simulations of the feedback to infer its dynamics and interplay with the circumgalactic medium, hence the evolution of galaxies.

We investigate the manifestation of the spiral structure in the distribution of high-mass X-ray binaries (HMXBs) over the host galaxy. We construct the simple kinematic model. It shows that the HMXBs should be displaced relative to the spiral structure observed in such traditional star formation rate indicators as the Halpha and FIR emissions because of their finite lifetimes. Using Chandra observations of M51, we have studied the distribution of X-ray sources relative to the spiral arms of this galaxy observed in Halpha. Based on K-band data and background source number counts, we have separated the contributions from high-mass and low-mass X-ray binaries and active galactic nuclei. In agreement with model predictions, the distribution of HMXBs is wider than that of bright HII regions concentrated in the region of ongoing star formation. However, the statistical significance of this result is low, as is the significance of the concentration of the total population of X-ray sources to the spiral arms. We also predict the distribution of HMXBs in our Galaxy in Galactic longitude. The distribution depends on the mean HMXB age and can differ significantly from the distributions of such young objects as ultracompact HII regions.

Context. Feedback processes and the galactic shear regulate star formation. Aims. We investigate the effects of differential galactic rotation and stellar feedback on the interstellar medium (ISM) and on the star formation rate (SFR). Methods. A numerical shearing box is used to perform three-dimensional simulations of a 1 kpc stratified cubic box of turbulent and self-gravitating interstellar medium (in a rotating frame) with supernovae and H II feedback. We vary the value of the velocity gradient induced by the shear and the initial value of the galactic magnetic field. Finally, the different star formation rates and the properties of the structures associated with this set of simulations are computed. Results. We first confirm that the feedback has a strong limiting effect on star formation. The galactic shear has also a great influence: the higher the shear, the lower the SFR. Taking the value of the velocity gradient in the solar neighbourhood, the SFR is too high compared to the observed Kennicutt law, by a factor approximately three to six. This discrepancy can be solved by arguing that the relevant value of the shear is not the one in the solar neighbourhood, and that in reality the star formation efficiency within clusters is not 100%. Taking into account the fact that star-forming clouds generally lie in spiral arms where the shear can be substantially higher (as probed by galaxy-scale simulations), the SFR is now close to the observed one. Different numerical recipes have been tested for the sink particles, giving a numerical incertitude of a factor of about two on the SFR. Finally, we have also estimated the velocity dispersions in our dense clouds and found that they lie below the observed Larson law by a factor of about two. Conclusions. In our simulations, magnetic field, shear, H II regions, and supernovae all contribute significantly to reduce the SFR. In this numerical setup with feedback from supernovae and H II regions and a relevant value of galactic shear, the SFRs are compatible with those observed, with a numerical incertitude factor of about two.

Recent studies have started to cast doubt on the assumption that most stars are formed in clusters. Observational studies of field stars and star cluster systems in nearby galaxies can lead to better constraints on the fraction of stars forming in clusters. We aim to constrain the amount of star formation happening in long-lived clusters for four galaxies through the homogeneous study of field stars and star clusters. Using HST/ACS-WFPC2 images of the galaxies NGC45, NGC1313, NGC5236 and NGC7793, we estimate star formation histories by means of the synthetic CMD method. Masses and ages of star clusters are estimated using simple stellar population model fitting. Comparing observed and modeled luminosity functions we estimate cluster formation rates. By randomly sampling the stellar IMF, we construct artificial star clusters and quantify how stochastic effects influence cluster detection, integrated colors and age estimates. Star formation rates appear to be constant over the past 10-100 Myr. The number of clusters identified per galaxy varies, with few massive (>10^5Msun) and few old (>1Gyr) clusters. The galaxies NGC5236 and NGC1313 show high star and cluster formation rates compare to NGC7793 and NGC45. Stochastic sampling of the SIMF has a strong impact on estimation of ages, colors and completeness for clusters with masses <10^4Msun, while for high masses the effect is less pronounced. Stochasticity also makes size measurements highly uncertain at young ages, making it difficult to distinguish between clusters and stars based on sizes. The ratio of star formation happening in clusters compared to the global star formation ($\Gamma$) appears to vary for different galaxies. We find no obvious relation between $\Gamma$ and the star formation rate density within the range probed here. The $\Gamma$ values do, however, appear to correlate with the specific U-band luminosity (T_L (U)).

We present high spatial resolution X-ray imaging data for the interacting galaxy pair NGC 7714/7715 (Arp 284) from the Chandra X-ray telescope. In addition to the unresolved starburst nucleus, a variable point source with LX ≈ 1040 ergs s-1 was detected 15 (270 pc) to the northwest of the nucleus, coincident with a blue, extremely optically luminous (MV ≈ -14.1) point source on Hubble Space Telescope images. Eleven other candidate pointlike ultraluminous X-ray sources (ULXs) were also detected in the vicinity of NGC 7714/7715, two of which exceed 1040 ergs s-1. Ten of these appear to be associated with interaction-induced features, but only two are associated with star formation regions. We also found diffuse emission with LX ≈ 3 × 1040 ergs s-1 extending 11'' (1.9 kpc) to the north of the nucleus. Its spectrum can be fitted with either a two-temperature MEKAL function (kT = 0.59 and 8 keV) or a 0.6 keV MEKAL function plus a power law (Γ = 1.8 ± 0.2). The hard component may be due to high-mass X-ray binaries (HMXBs) with possible contributions from inverse Compton radiation, while the soft component is likely from a superwind. Superbubble models imply an expansion age of ≈15 Myr, supporting previous assertions of an intermediate-age nuclear stellar population in addition to a 5 Myr starburst. We also detected extended X-ray emission associated with four extranuclear H II region complexes. The emission from these H II regions and the nuclear starburst could be due to either an enhanced population of HMXBs relative to Local Group galactic averages or to diffuse gas heated by winds from supernovae, if the X-ray production efficiency LX/Lmech is high (≈5%). To estimate LX/Lmech, we collected published data for well-studied H II regions and superbubbles in nearby galaxies. For H II regions with ages less than 3.5 Myr, the median LX/Lmech ≈ 0.02%, while for older star formation regions, LX/Lmech ≈ 0.2%–7%. Thus, it is possible that gas heating by supernovae may be sufficient to account for the observed X-rays from these H II regions. In galaxies much more distant than NGC 7714, for example, the Cartwheel galaxy, H II region complexes similar to those in NGC 7714 will be unresolved by Chandra and will mimic ULXs. No X-ray emission was detected from the Type Ib supernova SN 1999dn, with an upper limit of ≈2 × 1038 ergs s-1.

We present a comprehensive study of the total X-ray emission from the colliding galaxy pair NGC2207/IC2163, based on Chandra, Spitzer, and GALEX data. We detect 28 ultra-luminous X-ray sources (ULXs), 7 of which were not detected previously due to X-ray variability. Twelve sources show significant long-term variability, with no correlated spectral changes. Seven sources are transient candidates. One ULX coincides with an extremely blue star cluster (B-V = -0.7). We confirm that the global relation between the number and luminosity of ULXs and the integrated star formation rate (SFR) of the host galaxy also holds on local scales. We investigate the effects of dust extinction and/or age on the X-ray binary (XRB) population on sub-galactic scales. The distributions of Nx and Lx are peaked at L(IR)/L(NUV)~1, which may be associated with an age of ~10 Myr for the underlying stellar population. We find that ~1/3 of the XRBs are located in close proximity to young star complexes. The luminosity function of the X-ray binaries is consistent with that typical for high-mass X-ray binaries, and appears unaffected by variability. We disentangle and compare the X-ray diffuse spectrum with that of the bright XRBs. The hot interstellar medium dominates the diffuse X-ray emission at E<1 keV, has a temperature kT=0.28 (+0.05/-0.04) keV and intrinsic 0.5-2 keV luminosity of 7.9e+40 erg/s, a factor of ~2.3 higher than the average thermal luminosity produced per unit SFR in local star-forming galaxies. The total X-ray output of NGC2207/IC2163 is 1.5e+41 erg/s, and the corresponding total integrated SFR is 23.7 Msol/yr.

We present the results of two Chandra observations (separated by 1 yr) of the population of X-ray sources in the spiral galaxy M51 (NGC 5194 and NGC 5195). One hundred and thirteen X-ray sources have been detected in an 84 × 84 (20.4 × 20.4 kpc) region, and 84 and 12 of them project within the disks of NGC 5194 and NGC 5195, respectively. Nine and 28 sources have luminosities exceeding 1 × 1039 ergs s-1 (ultraluminous X-ray sources or ULXs) and 1 × 1038 ergs s-1 in the 0.5-8 keV band, respectively, assuming that they are associated with M51. The number of ULXs is much higher than found in most normal spiral and elliptical galaxies. Most of the X-ray sources and all seven of the ULXs in NGC 5194 are located in, or close to, a spiral arm, suggesting a connection with recent star formation. The cumulative luminosity function of the X-ray sources in NGC 5194 with L(0.5-8 keV) > 1038 ergs s-1 is well described by a power law, N[>L(0.5-8 keV)] ∝ L(0.5- 8 keV)-α with α = 0.91. The X-ray spectra of most of the detected sources are consistent with a power law with a photon index between 1 and 2, with a few sources showing harder or softer spectra. The spectra of most ULXs are consistent with both a power law and a multicolor disk blackbody (MCD) model, while a power-law model is preferable to an MCD model in two ULXs. One ULX (NGC 5194 source 69) shows drastic spectral steepening accompanied by a decline in luminosity by a factor of 3460 in the 2-10 keV band between the two observations. This source also exhibited a possible period of 2.1 hr in the 2000 observation. Another ULX (NGC 5194 source 26) shows strong emission lines from highly ionized species. The masses of the compact objects and mass accretion rates in ULXs and other X-ray sources are not well constrained by these observations. If we adopt an MCD interpretation, their MCD parameters imply that most of the X-ray sources are stellar mass (~5-10 M☉) black holes accreting near or above the Eddington rate, although other possibilities (intermediate-mass black holes and relativistically beamed emission) cannot be excluded. The power-law sources may instead represent Comptonized disk or nonthermal emission. Two ULXs have very soft spectra; MCD models require kT ≈ 0.1 keV. We discuss the possibility that this soft emission originates in an accretion disk around an intermediate-mass black hole. We also present a study of the nucleus of, and discrete sources (including two ULXs) in, the companion galaxy NGC 5195.

We present an X-ray study of the massive edge-on Sa galaxy, Sombrero (M 104; NGC 4594), based on XMM–Newton and Chandra observations. A list of 62 XMM–Newton and 175 Chandra discrete X-ray sources is provided, the majority of which are associated with the galaxy. Spectral analysis is carried out for relatively bright individual sources and for an accumulated source spectrum. At energies ≳2 keV, the source-subtracted X-ray emission is distributed similarly as the stellar K-band light and is primarily due to the residual emission from discrete sources. At lower energies, however, a substantial fraction of the source-subtracted emission arises from diffuse hot gas extending to ∼20 kpc from the galactic centre. The galactic disc shows little X-ray emission and instead shadows part of the X-ray radiation from the bulge. The observed diffuse X-ray emission from the galaxy shows a steep spectrum that can be characterized by an optically thin thermal plasma with temperatures of ∼0.6–0.7 keV, varying little with radius. The diffuse emission has a total luminosity of ∼3 × 1039 erg s−1 in the 0.2–2 keV energy range. This luminosity is significantly smaller than the prediction by current numerical simulations for galaxies as massive as Sombrero. However, such simulations do not include the effect of quiescent stellar feedback (e.g. ejecta from evolving stars and Type Ia supernovae) against the accretion from intergalactic medium. We argue that the stellar feedback likely plays an essential role in regulating the physical properties of hot gas. Indeed, the observed diffuse X-ray luminosity of Sombrero accounts for at most a few per cent of the expected mechanical energy input from Type Ia supernovae. The inferred gas mass and metal content are also substantially less than those expected from stellar ejecta. We speculate that a galactic bulge wind, powered primarily by Type Ia supernovae, has removed much of the ‘missing’ energy and metal-enriched gas from the region revealed by the X-ray observations.

Understanding how the baryonic physics affects the formation and evolution of galaxies is one of the most critical questions in modern astronomy. Significant progress in understanding stellar feedback and modeling them explicitly in simulations have made it possible to reproduce a wide range of observed galaxy properties. However, there are still various pieces of missing physics and uncertainties in galaxies of different mass range. In this thesis, I will explore these missing pieces in baryonic physics on top of the Feedback in Realistic Environments (FIRE) stellar feedback in the cosmological hydrodynamic zoom-in simulations (FIRE-2 suite) and isolated galaxy simulations. These high-resolution simulations with FIRE physics capture multi-phase realistic interstellar medium (ISM) with gas down to 10K, and star formations in dense clumps in giant molecular clouds. They are, therefore, an ideal tool for investigating the missing pieces in baryonic physics. In the first part of the thesis, Chapter 2, I will focus on the discrete effects of stellar feedback like individual supernovae, hypernovae, and initial mass function (IMF) sampling in dwarfs (109-1010 M⊙). These discrete processes of stellar feedback can have maximum effects on the small galaxies without being averaged out. I will show that the discretization of supernovae (SNe) is absolutely necessary, while the effects from IMF sampling and hypernovae (HNe) is not apparent, due to the strong clustering nature of star formation. In the second part of the thesis, Chapter 3-4, I will focus on fluid microphysics, exploring their effects on galaxy properties and their interplay with stellar feedback in sub-L* galaxies. I will demonstrate that, once the stellar feedback is explicitly implemented as FIRE stellar feedback model, fluid microphysics such as magnetic fields, conduction, and viscosity only have minor effects on the galaxy properties like star formation rate (SFR), phase structure, or outflows. Stellar feedback also strongly alters the amplifications and morphology of the magnetic fields, resulting in much more randomly-oriented field lines. However, despite the stellar the amplification of magnetic fields in ISM gas is primarily dominated by flux-freezing compression. In the final part of my thesis, I focus on the massive cluster ellipticals of 1012-1014 M⊙, where the physical mechanisms that regulate the observation-inferred flows are highly uncertain -- the classic cooling flow problem. I showed that solutions in the literature not associated with an active galactic nucleus (AGN), including stellar the cosmic ray from stellar magnetic fields, conduction, and morphological quenching, cannot possibly quench the galaxies, mostly because of the insufficient energy and the limited size of the affected region. After ruling out the non-AGN feedback solutions to the flow problem, I will go into the most accessible, and perhaps promising solution: feedback, exploring the generic classes of AGN feedback models proposed in the literature. I am going to show that enhancing turbulence and injecting cosmic ray are probably the most important aspects of AGN feedback in galaxy quenching. Since they provide non-thermal pressure support that stably suppresses the core density, they can stably reduce the flows without overheating the galactic cores.

We discuss the dependence of various properties of the star formation rate (SFR) and efficiency (SFE) in molecular clouds (MCs) on the maximum mass reached by the clouds, based on a previously-published model for MC and SFR evolution in which the clouds were assumed to be undergoing global collapse, and the SFR was controlled by ioniztion feedback. Because the model neglects various other processes, the results presented are upper limits. We find that clouds with $\Mmax \lesssim 10^4 \Msun$ end their lives with a mini-burst, at which the SFR reaches a peak of $\sim 10^4 ~\Msun \Myr^{-1}$, although its time average is only $\SFRavg \sim \hbox{ a few} \times 10^2 \Msun \Myr^{-1}$. The corresponding efficiencies are $\SFEmax \lesssim $60\%$ and $\SFEavg \lesssim $1\%$. For more massive clouds ($\Mmax \gtrsim 10^5 ~ \Msun$), the SFR first increases and then remains roughly constant for $\sim 10^7$ yr, because the clouds are influenced by the stellar feedback since earlier in their evolution. We find that $\SFRavg$ and $\SFEavg$ are well represented by the fits $\SFRavg \approx 100 (1+\Mmax/2 \times 10^5 ~ \Msun)^{2} ~ \Msun \Myr^{-1}$ and $\SFEavg \approx 0.024 (\Mmax/10^5 ~ \Msun)^{0.28}$, respectively. The massive model clouds follow the SFR-dense gas mass relation obtained by Gao \& Solomon for infrared galaxies, extrapolated down to MCs scales. Low-mass clouds fall above this relation, in agreement with recent observations. An integration of the model-predicted $\SFRavg$ over a Galactic GMC mass spectrum yields a realistic value for the Galactic SFR. Our results reinforce the suggestion that star-forming GMCs may be in global collapse, and still have low net SFRs and SFEs, due to the evaporation of most of the cloud material by the feedback from massive stars.

The Chandra Deep Field-South and North surveys (CDFs) provide unique windows into the cosmic history of X-ray emission from normal (nonactive) galaxies. Scaling relations of normal-galaxy X-ray luminosity (L X) with star formation rate (SFR) and stellar mass (M ⋆) have been used to show that the formation rates of low-mass and high-mass X-ray binaries (LMXBs and HMXBs, respectively) evolve with redshift across z ≈ 0–2 following L HMXB/SFR ∝ (1 + z) and L LMXB/M ⋆ ∝ (1 + z)2−3. However, these measurements alone do not directly reveal the physical mechanisms behind the redshift evolution of X-ray binaries (XRBs). We derive star formation histories for a sample of 344 normal galaxies in the CDFs, using spectral energy distribution (SED) fitting of FUV-to-FIR photometric data, and construct a self-consistent, age-dependent model of the X-ray emission from the galaxies. Our model quantifies how X-ray emission from hot gas and XRB populations vary as functions of host stellar-population age. We find that (1) the ratio L X/M ⋆ declines by a factor of ∼1000 from 0 to 10 Gyr and (2) the X-ray SED becomes harder with increasing age, consistent with a scenario in which the hot gas contribution to the X-ray SED declines quickly for ages above 10 Myr. When dividing our sample into subsets based on metallicity, we find some indication that L X/M ⋆ is elevated for low-metallicity galaxies, consistent with recent studies of X-ray scaling relations. However, additional statistical constraints are required to quantify both the age and metallicity dependence of X-ray emission from star-forming galaxies.

Feedback from massive stars plays an important role in the formation of star clusters. Whether a very massive star is born early or late in the cluster formation timeline has profound implications for the star cluster formation and assembly processes. We carry out a controlled experiment to characterize the effects of early-forming massive stars on star cluster formation. We use the star formation software suite Torch, combining self-gravitating magnetohydrodynamics, ray-tracing radiative transfer, N-body dynamics, and stellar feedback, to model four initially identical 104 M ⊙ giant molecular clouds with a Gaussian density profile peaking at 521.5 cm−3. Using the Torch software suite through the AMUSE framework, we modify three of the models, to ensure that the first star that forms is very massive (50, 70, and 100 M ⊙). Early-forming massive stars disrupt the natal gas structure, resulting in fast evacuation of the gas from the star-forming region. The star formation rate is suppressed, reducing the total mass of the stars formed. Our fiducial control model, without an early massive star, has a larger star formation rate and total efficiency by up to a factor of 3, and a higher average star formation efficiency per freefall time by up to a factor of 7. Early-forming massive stars promote the buildup of spatially separate and gravitationally unbound subclusters, while the control model forms a single massive cluster.

Sub-virial gravitational collapse is one mechanism by which star clusters may form. Here we investigate whether this mechanism can be inferred from observations of young clusters. To address this question, we have computed SPH simulations of the initial formation and evolution of a dynamically young star cluster through cold (sub-virial) collapse, starting with an ellipsoidal, turbulently seeded distribution of gas, and forming sink particles representing (proto)stars. While the initial density distributions of the clouds do not have large initial mass concentrations, gravitational focusing due to the global morphology leads to cluster formation. We use the resulting structures to extract observable morphological and kinematic signatures for the case of sub-virial collapse. We find that the signatures of the initial conditions can be erased rapidly as the gas and stars collapse, suggesting that kinematic observations need to be made either early in cluster formation and/or at larger scales, away from the growing cluster core. Our results emphasize that a dynamically young system is inherently evolving on short timescales, so that it can be highly misleading to use current-epoch conditions to study aspects such as star formation rates as a function of local density. Our simulations serve as a starting point for further studies of collapse including other factors such as magnetic fields and stellar feedback.