Draft version May 6, 2024

Typeset using LATEX twocolumn style in AASTeX631

SAUNAS I: Searching for Low Surface Brightness X-ray Emission with Chandra/ACIS

Alejandro S. Borlaff ,1, 2 Pamela M. Marcum,1 Mehmet Alpaslan,3 Pasquale Temi ,1 Nushkia Chamba ,1
Drew S. Chojnowski ,1 Michael N. Fanelli,1 Anton M. Koekemoer ,4 Seppo Laine ,5
Enrique Lopez-Rodriguez ,6 and Aneta Siemiginowska 7
1 NASA Ames Research Center, Moffett Field, CA 94035, USA
2 Bay Area Environmental Research Institute, Moffett Field, California 94035, USA
arXiv:2405.01625v1 [astro-ph.GA] 2 May 2024

3 ConstructConnect, Rookwood Exchange, 3825 Edwards Rd 800, Cincinnati, OH 45209

4 Space Telescope Science Institute, 3700 San Martin Dr., Baltimore, MD 21218, USA
5 IPAC, Mail Code 314-6, Caltech, 1200 E. California Blvd., Pasadena, CA 91125, USA
6 Kavli Institute for Particle Astrophysics & Cosmology (KIPAC), Stanford University, Stanford, CA 94305, USA
7 Harvard Smithsonian Center for Astrophysics, 60 Garden St, Cambridge, MA 02138, USA

ABSTRACT
We present SAUNAS (Selective Amplification of Ultra Noisy Astronomical Signal), a pipeline designed
for detecting diffuse X-ray emission in the data obtained with the Advanced CCD Imaging Spectrometer
(ACIS) of the Chandra X-ray Observatory. SAUNAS queries the available observations in the Chandra
archive, performs photometric calibration, PSF (point spread function) modeling and deconvolution,
point-source removal, adaptive smoothing, and background correction. This pipeline builds on existing
and well-tested software including CIAO, vorbin, and LIRA. We characterize the performance of SAUNAS
through several quality performance tests, and demonstrate the broad applications and capabilities of
SAUNAS using two galaxies already known to show X-ray emitting structures. SAUNAS successfully
detects the 30 kpc X-ray super-wind of NGC 3079 using Chandra/ACIS datasets, matching the spatial
distribution detected with more sensitive XMM-Newton observations. The analysis performed by
SAUNAS reveals an extended low surface brightness source in the field of UGC 5101 in the 0.3–1.0 keV and
1.0–2.0 keV bands. This source is potentially a background galaxy cluster or a hot gas plume associated
with UGC 5101. SAUNAS demonstrates its ability to recover previously undetected structures in archival
data, expanding exploration into the low surface brightness X-ray universe with Chandra/ACIS.

Keywords: X-ray astronomy (1810), Astronomical methods (1043), X-ray photometry (1820), X-ray
observatories (1819), X-ray telescopes (1825), Circumgalactic medium (1879)

1. INTRODUCTION gle12 , necessitating an elaborate deconvolution scheme

The Advanced CCD Imaging Spectrometer on the and hampering the ability to exploit the full capabili-
Chandra X-ray Observatory (Weisskopf et al. 2000, ties of the archive. Consequently, Chandra observations
hearafter Chandra/ACIS) provides an effective balance are under-explored to date in studies advancing the low
between angular resolution and sensitivity for the study X-ray surface brightness (SB) domain.
of diffuse galactic hot gas emission, with its field of view Future studies of low X-ray SB emission (∼10−8 to
(FOV) up to 16.9 × 16.9 arcmin2 and 0.492 arcsec of 10−11 s−1 cm−2 arcsec−2 and beyond) enabled by data
spatial resolution. Stacking multiple observations made processed to enhance detection of low-count regions
over Chandra’s 25+ year mission is one of the keys to could advance progress in several currently open ques-
obtaining the deepest observations of the universe in X- tions relevant to galaxy evolution, including the origins
ray. However, in most cases, the position of the target
on the detector changes within observations, introducing 1 Understanding the Chandra PSF https://cxc.cfa.harvard.edu/
serious challenges to acquiring a meaningful combined ciao/PSFs/psf central.html
2 Chandra/CIAO PSF presentation from 233rd AAS meeting:
image. The PSF (point spread function) broadens and
becomes more ellipse-shaped with increasing off-axis an- https://cxc.harvard.edu/ciao/workshop/nov14/02-Jerius.pdf
2

of diffuse soft X-ray emission in galaxies and feedback km s−1 Mpc−1 , see Spergel et al. 2007). All magnitudes
involvement (Kelly et al. 2021; Henley et al. 2010). are in the AB system (Oke 1971) unless otherwise noted.
Lambda Cold Dark Matter (LCDM) cosmology pre-
dicts filaments of diffuse gas from the cosmic web to 2. METHODOLOGY
accrete during their infall onto proto-galactic dark mat- 2.1. Observational challenges
ter (DM) halos (White & Rees 1978; White & Frenk
From an observational perspective, measuring diffuse
1991; Benson & Devereux 2010) where gas is heated to
X-ray halo properties in galaxies involves at least four
approximately the halo virial temperature (T > 106 K).
technical challenges:
This plasma, further shaped by energy injection from
active galactic nuclei (AGN, Diehl & Statler 2008), su- 1. Detection: The outskirts of X-ray halos are ex-
pernovae (SN) and stellar winds (Hopkins et al. 2012), tremely faint (≲ 10−8 – 10−11 s−1 cm−2 arcsec−2 ).
is detected as diffuse soft X-ray band emission around Separating the faint emission associated with
galaxies (Mulchaey 2000; O’Sullivan et al. 2001; Sato sources from that of the X-ray background (An-
et al. 2000; Aguerri et al. 2017). The origins and evo- derson & Bregman 2011) within such low count
lution of hot gas halos are important open questions in regimes is an extraordinarily challenging task.
astrophysics, as halos are both the aftermath and ac- Statistical methods that assume a normal (Gaus-
tive players of gas feedback processes, which modulate sian) distribution may not produce accurate re-
the star formation efficiency in galaxies (Rees & Ostriker sults.
1977; Silk 1977; Binney 1977; White & Rees 1978; White
& Frenk 1991). The largely-unexplored realm of extreme 2. Deblending: AGNs and XRBs are typically unre-
diffuse gas emission, likely associated with large depar- solved point sources that may contribute to the
tures from equilibrium (Strickland et al. 2004), is likely same X-ray bands where the hot gas halos are ex-
to preserve a unique historical record of these events. pected to emit (from ∼ 0.3 − 0.5 to 1.2 − 2 keV).
Such emission is also likely to be disregarded in stud- While in principle the detection of hot gas halos
ies using standard pipelines that are not optimized for in nearby galaxies may not require very high spa-
preservation of statistically significant but low SB de- tial resolution observations or spectral capabilities,
tections. the separation of such emission from that of point
This project is the first in a series that will study the sources does require them. High spatial resolution
hot gas halos around galaxies using X-ray observations observations reduce systematic contamination in
from the Chandra X-ray observatory. The first step is low surface brightness regimes.
to test the pipeline to reduce the Chandra/ACIS data
products, named SAUNAS (Selective Amplification of Ul- 3. Point spread function (PSF) contamination: The
tra Noisy Astronomical Signal). This paper describes distribution of diffuse emission is easily confused
the SAUNAS pipeline processing of data from the Chan- with the scattered, extended emission of the unre-
dra Data Archive3 and benchmarks it to previous works. solved bright cores that contaminate the outskirts
In particular, we focus on the comparison of results be- of the target through the extended wings of the
tween our analyses and those from other investigations PSF of the detector (Sandin 2014, 2015). Most
for two well-detected X-ray sources characterized in the studies do not correct for this type of scattering
literature: NGC 3079 and UGC 5101. The latter has effect, although a few works, such as Anderson
complex and extended X-ray emission, previously unex- et al. (2013), have explored the combined stacked
plored and only revealed by the current work. hot gas halo emission of 2165 galaxies observed
This paper is organized as follows. The SAUNAS with ROSAT (0.5–2.0 keV), convolving the com-
pipeline is described in Sec. 2. The selection of pub- bined surface brightness profiles by the PSF model
lished results for SAUNAS performance comparison is dis- to take into account the dispersion of light.
cussed in Sec. 3.1. The benchmark analysis is presented
4. Reproducibility & Accessibility: The methodolo-
in Secs. 3.2, and 3.3. The discussion and conclusions are
gies for calibration, detection, and characteriza-
presented in Secs. 4 and 5, respectively. We assume a
tion of X-ray emission have substantial differences
concordance cosmology (ΩM = 0.3, ΩΛ = 0.7, H0 = 70
between studies. Due to the Poissonian nature
of the X-ray emission, most studies employ differ-
3 Chandra Data Archive: https://cxc.harvard.edu/cda/ ent types of adaptive smoothing in their analysis.
These software methods tend to be custom-made
and infrequently made publicly available. Like-
 3

wise, the final data products (final science frames) 1. All available Chandra/ACIS observations con-
are seldom offered to the community. taining the user-supplied sky coordinates
The SAUNAS methodology presented in the current pa- are identified using find chandra obsid.
per attempts to address most of these points by 1) cor- The datasets and their best available
recting the PSF in the images, 2) separating the emis- calibration files are automatically down-
sion of point sources from that of diffuse extended ones, loaded using (download chandra obsid and
and 3) providing a quantitative metric to determine if download obsid caldb).
a detection is real or not. These two points imple-
mented in SAUNAS are the major difference with other 2. The raw observations are reprocessed using
existing codes for detection of extended X-ray emission, chandra repro (v4.16). To avoid over-subtraction
such as vtpdetect (Ebeling & Wiedenmann 1993) or of both the source and background counts neces-
EXSdetect (Liu et al. 2013), as they do not attempt to sary for the statistical analysis, the particle back-
deconvolve the observations using dedicated PSF models ground cleaning subprocess is set (check vf pha)
or to separate diffuse emission from point sources. to “no”. See the main CIAO manual8 for more in-
formation on this step.
2.2. SAUNAS pipeline
SAUNAS generates two main products: a) PSF- 3. All the available ACIS datasets are merged into a
deconvolved X-ray adaptively smoothed surface bright- single events file (merge obs). This product serves
ness maps and b) signal-to-noise ratio (SNR) detection as the phase 1 (first pass) observation file and is
maps. The X-ray adaptively smoothed surface bright- used to identify emission regions and to determine
ness maps provide the flux and luminosity of the hot gas the source spectra needed for PSF construction.
X-ray halos, while the SNR detection maps provide the
probability that the flux associated with each region on 4. The phase 1 merged observation file is used to de-
those maps is statistically higher than the local X-ray fine the angular extent of detected emission suffi-
background noise. cient for basic spectral characterization. The spec-
SAUNAS creates these products in four major steps (see tral information is used in the step following this
Fig. 1): 1) pre-processing of the archival Chandra X- one. The VorBin (Cappellari & Copin 2003) li-
ray observations using the Chandra Interactive Anal- brary generates a map of Voronoi bins, from which
ysis of Observations 4 software, (Fruscione et al. 2006, and a surface brightness profile is constructed.
CIAO, hereafter, see Sec. 2.2.1), 2) statistical resampling The preliminary detection radius (Rlim,0 ), defined
of the X-ray detection events by bootstrapping5 , 3) PSF as the radial limit having a surface brightness
deconvolution of the event maps using the Bayesian equal to 10% of the surface brightness at the cen-
Markov Chain Monte Carlo (MCMC) LIRA tool (Low- tral coordinates, is computed. If Rlim,0 is unde-
counts Image Reconstruction and Analysis, Donath et al. fined due to a low central surface brightness, the
(2022b); see Sec. 2.2.3)6 , and 4) adaptive smoothing us- presence of detectable emission is unlikely. For
ing VorBin (see Sec. 2.2.4)7 . SAUNAS requires a few user- such cases, Rlim,0 is arbitrarily set to 1/4 of the
input parameters, including the location of the target FOV defined by the user. The events inside this
(α, δ), field-of-view (FOV), and energy band. The main detection radius are used to construct a spectrum
steps of the pipeline are described in the following sub- employed in the next step to define the deconvolu-
sections. tion kernel (e.g., PSF) appropriate for this target.
The choice of a 10% limit is an optimal compro-
2.2.1. CIAO pre-processing
mise based on the analysis of Chandra/ACIS ob-
First, the data is pre-processed using CIAO in the fol- servations: including as much emission as possible
lowing way: from the source enhances the spectra used to gen-
erate the PSF. However, including a region too
4
large reduces computational efficiency. Note that
CIAO: Chandra Interactive Analysis of Observations https://
cxc.cfa.harvard.edu/ciao/ the spectrum derived in this step serves the sole
5 Bootstrapping: https://link.springer.com/chapter/10.1007/ purpose of informing PSF construction and is not
978-1-4612-4380-9 41 intended for physical characterization of the gas.
6 LIRA: Low-counts Image Reconstruction and Analysis - https:

//pypi.org/project/pylira/
7 VorBin: Adaptive Voronoi Binning of Two Dimensional Data 8 ACIS VFAINT Background Cleaning: https://cxc.harvard.
- https://pypi.org/project/vorbin/ edu/ciao/why/aciscleanvf.html
4

Figure 1. SAUNAS pipeline flowchart. From left to right: SAUNAS pre-calibrates the Chandra observations by first using Chandra
X-ray Center (CXC)/CIAO, which generates the event files, extended source masks, and the PSFs. The events in each individual
visit are first resampled via bootstrapping and then deconvolved using LIRA. Voronoi binning is applied to each deconvolved
observation, and merged into a single flux map after sky background correction.
5. CIAO’s task simulate psf, in combination with and 3.8±0.2 × 10−12 erg cm−2 deg−2 s−1 for the 2–
the spectral information provided by the previ- 8 keV band. This flux is equivalent to ∼ 0.03–0.003
ous step, is used to generate a PSF representa- photons arcsec−2 (1–2 keV) and ∼ 0.01–0.001 pho-
tive of each observing visit to the target. The tons arcsec−2 for a typical t = 104 –105 s exposure (She
PSF modeling is dependent on the spectra of both et al. 2017). As a consequence, the signal from spu-
the source and the background region, as well as rious groups of a few counts can dominate the shape
the target position within the detector (off-axis an- of the Voronoi bins used for adaptive smoothing in each
gle). The latter is unique to each visit. The pre- simulation if appropriate statistical methods are not im-
liminary detection radius defines both the circular plemented.
(R < Rlim,0 ) and annular (Rlim,0 < R < 2 Rlim,0 ) To enhance the robustness of the adaptively smoothed
apertures used to measure the source and back- mosaics and to reduce contamination from non-
ground spectra, respectively (specextract). The significant signal in the background, the X-ray events
aspectblur9 is set to 0.25, and the number of it- are re-sampled via replacement (bootstrapping) as an ad-
erations to 1000 per dataset. ditional (and user-optional) step before deconvolution.
Bootstrapping is especially well-suited for inferring the
6. Finally, the individual event files and PSFs cor- uncertainties associated with an estimand – such as the
responding to each visit are cropped to a cutout, median surface brightness in cases for which the Gaus-
with the preferred energy range selected. sian standard deviation regime does not apply or para-
metric solutions are too complicated or otherwise un-
The outputs from the pre-processing procedure known. Bootstrapping effectively reduces the leverage
with CIAO described above, are: 1) the detected that single events or very low count sources may have
event maps (named obsid Elow-upper flt evt.fits, in the background of the final mosaics by accounting for
where low and upper refer to the energy range the photon-noise uncertainties in the PSF deconvolution
limits and obsid is the observation ID identifica- and Voronoi binning steps through a non-parametric ap-
tion in the Chandra archive), 2) the exposure time proach, allowing for a better assessment of the uncer-
maps (obsid Elow-upper flt expmap.fits), 3) the tainties in the final simulations.
flux maps (obsid Elow-upper flt flux.fits), and 4) In our application, bootstrapping generates N ∼ 100
the PSF (obsid Elow-upper flt psf.fits). This set (hereafter, Nboot ) new X-ray event samples from the
of intermediate files is used in the remaining steps of the observed sample, preserving size (flux) and permitting
SAUNAS pipeline to generate the final maps. events to be repeated. While the number of bootstrap-
ping simulations is set to 100 by default as a compromise
2.2.2. X-ray event resampling: bootstrapping between precision and computational resources, Nboot
The X-ray sky background is a very low count regime. can be defined by the user in SAUNAS. Each resampled
Bartalucci et al. (2014) obtained a flux of 10.2+0.5
−0.4 ×
list of events is translated into an image, which is fed
−13 −2 −2 −1 into the next step, PSF deconvolution (Sec. 2.2.3).
10 erg cm deg s for the 1–2 keV band
2.2.3. LIRA PSF deconvolution
9 Aspectblur in CIAO: https://cxc.cfa.harvard.edu/ciao/why/ The LIRA (Connors et al. 2011; Donath et al. 2022b)
aspectblur.html package deconvolves the emission from sources in X-ray
 5

Simulated source PSF

PSF convolved source Simulated observation

LIRA deconvolved Final smoothed mosaic

Figure 2. SAUNAS analysis test on a synthetic dataset. Top left: Underlying distribution of the simulated test source. Top right:
Point spread function (PSF) of the simulated observation. Central left: Simulated underlying distribution of the test source
convolved by the PSF. Central right: Simulated observed events based on the PSF-convolved distribution. Bottom left: LIRA
PSF deconvolved average posterior image. Bottom right: Adaptively smoothed final mosaic. Dashed contours represent the 3σ
and dotted contours the 2σ detection level of X-ray emission. The equivalent exposure time for this test is τexp = 5 × 106 s cm2 .
6

data. Through the use of LIRA, SAUNAS removes the age are used (hereafter, Nstable ), which typically is equal
contamination from active galactic nuclei (AGNs) and to ∼100. To save computational resources, Nstable is
X-ray binary stars (XRBs) which can be significantly adapted based on the number of bootstrapping simula-
extended and easily confused with a diffuse halo if the tions so that the deconvolved dataset consists of a max-
PSF is not accurately corrected. LIRA uses a Bayesian imum of N = Nboot × Nstable = 1000 deconvolved
framework to obtain the best-fit PSF convolved model images (posterior samples).
to the observations, allowing the user to evaluate the
probability that a detection is statistically significant.
LIRA was designed to provide robust statistics in the 2.2.4. Adaptive Voronoi smoothing
low-count Poissonian regimes representative of faint ex-
The deconvolved datacubes, hereafter referred to as
tended halos, the primary science focus of our project.
”Bootstrapping-LIRA” realizations, serve as a proxy
As detailed in Sec. 2.2.1, the PSF models are generated
of the probability density distribution of the true X-
specifically for each target, taking into account their lo-
ray emission on a pixel-per-pixel basis, at the Chan-
cation in the detector and their spectral energy distribu-
dra/ACIS spatial resolution (a minimum of 0.492” px−1 ,
tions, on a per-visit basis. SAUNAS deconvolves data from
depending on the binning set by the user). To facili-
individual visits, using these PSF models as input into
tate the detection of extended, low surface brightness
LIRA. Discrete hard-band emission is produced primarily
structures such as hot gas halos – with apparent sizes
by point sources, including AGNs (Fabbiano et al. 1989;
substantially larger than the spatial resolution limit for
Fabbiano 2019), young stellar objects, and mass transfer
the galaxies – the use of spatial binning enhances the
onto the compact stellar object within XRB pairs (Wang
detection of regions with very low signal-to-noise ratio.
2012). Because these point sources contaminate the soft
Voronoi binning (VorBin, Cappellari & Copin 2003)
band emission, they are excised from the data. They
is applied to each of the N posterior samples in the de-
are identified using the Chandra Source Catalog (Evans
convolved datacube. This process generates N Voronoi
et al. 2010), and then removed from the event file by
tesselation maps, each one differing from the other be-
deleting events that lay within the cataloged positional
cause they were calculated from the Bootstrapping-LIRA
uncertainty ellipse of the source.
realizations. This dataset is a Voronoi map datacube
The Python implementation of LIRA is used to decon-
representing the probability density distribution of the
volve the X-ray event files, thus minimizing the effects of
surface brightness of the target.
the off-axis dependency associated with Chandra’s PSF,
A consequence of this binning approach is the loss of
such that data from different visits can be combined in
spatial resolution in the faintest regions of the image
a later stage. LIRA accepts five input arrays: a) counts
(halos, background) compared to the brightest regions
(number of events), b) flux (in s−1 cm−2 px−1 ), c) ex-
(i.e., the galactic cores). This loss is caused by the fact
posure ( s cm2 ), d) PSF, e) a first approximation to the
that the Voronoi technique varies the bin size in order
background (counts). The first four inputs are gener-
to achieve a fixed signal-to-noise ratio in the resulting
ated by the CIAO pipeline (Sec. 2.2.1), while the initial
map. As we are primarily interested in mapping the
baseline background is set to one. The number of LIRA
large scale halo structures, this loss in spatial resolution
simulations is set to 1000 (n iter max), in addition to
does not significantly impact our science goals.
100 initial burn-in simulations (num burn in). To speed
A surface brightness map is created by calculating the
up the process10 , SAUNAS splits the LIRA simulations
median across one of the axes of the Voronoi datacube.
in parallel processing blocks (defined by the number of
To prevent background emission from contaminating the
bootstrapping simulations), to be combined after the de-
final image, the scalar background level is determined
convolution process has finished. While 1000 LIRA sim-
individually for each realization of the Bootstrapping-
ulations are run on each of the N∼100 bootstrapping-
LIRA datacube. All sources, both resolved and unre-
resampled images described in Section 2.2.2, only the
solved, must be meticulously masked prior to measuring
last LIRA realizations (those produced after the decon-
the background level, to prevent systematically over-
volution process has stabilized) for each resampled im-
subtracting the background in the final mosaics. The
source masking and background correction process are
10 Even in parallel processing mode, PSF deconvolution takes the conducted iteratively:
largest fraction of time of the SAUNAS pipeline. As a reference, in
an Apple M1 Max 2021 laptop (32 Gb of RAM, 10 cores), the
computation of a 1024 × 1024 mosaic typically takes two hours,
with ∼90% of the time spent in deconvolution. 1. After the LIRA deconvolution process and before
the Voronoi binning is performed, point sources
 7

from the Chandra Source Catalog (CSC 2.0;11 2.3. Quality tests
likely X-ray binaries, SNe, AGNs) that lay in This section presents the results from a series of qual-
the image footprint are removed from the associ- ity tests designed to evaluate specific aspects of the out-
ated event file. Point source removal prevents the put mosaics generated with SAUNAS:
associated emission from impacting the adaptive
Voronoi maps and resulting in diffuse contamina- 1. Identify the fraction of false positives and false
tion that could be confused with a gas halo com- negative detections (Sec. 2.3.1).
ponent.
2. Estimate the flux conservation of the deconvolu-
2. A secondary mask is generated using CIAO’s rou- tion / Voronoi binning process (Sec. 2.3.2)
tine vtpdetect12 . This mask identifies the re-
gions with detectable extended X-ray emission 3. Quantify the quality of SAUNAS performance as
that are removed from the maps before measur- compared to that of other methods (arestore,
ing the background level. A mask is generated for Sec. 2.3.3).
each CCD of each visit through independent anal-
ysis. The masks are then combined into a single 2.3.1. False positive / False negative ratio
master extended source mask. For quality assessment, SAUNAS is tested using two dif-
ferent models varying the exposure time to reduce the
3. If a source was detected in the preliminary surface
photon flux and the detectability conditions:
brightness profile generated a part of the CIAO pre-
processing step (see Sec. 2.2.1, step 4), then those 1. A model of an idealized edge-on galaxy with two
pixels with R < Rlim,0 are also masked before the lobes emerging from a jet (double jet model).
background assessment.
2. A shell-like structure with a central bright source
4. After removing all the masked pixels using the (cavity model).
masks from the three previous steps, the first ap-
proximation of the background level (B0 ) is made The models are created as combinations
by measuring the median value of the unmasked of Gaussian 2D probability distributions
sigma clipped (σ = 3) pixels. The background (astropy.convolution.Gaussian2DKernel) with dif-
value is then subtracted from the voronoi binned ferent ellipticities and rotations as described in Table 1.
maps. Following PSF convolution, a synthetic observed events
map is generated using a random Poisson distribution
Once the individual observations have been back- (numpy.random.poisson).
ground corrected, all the flux maps are combined us- The double jet model includes the emission from
ing mean-weighting by the respective exposure times. three sources: the galactic disk, a bright core, and the
Finally, a refined background value (B1 ) is calculated lobes. The range of surface brightnesses is ∼ 10−6 –10−8
using the combined observations by repeating the pro- s−1 cm−2 arcsec−2 , excluding the considerably brighter
cess described above. The noise level is then estimated (three to five orders of magnitude brighter) peak surface
from the background distribution as the ratio between brightness of the core. Its morphology mimics the pre-
the median background level and the lower limit of the dominant structure observed in double jet radio galaxies
1σ error bar (equivalent to the 15.8% percentile). The fi- such as Centaurus A (Hardcastle et al. 2007).
nal background-subtracted, PSF-corrected, and Voronoi The other test simulation, a cavity model, contains a
binned surface brightness maps are derived by using a hollow shell with a central bright source. This model
median of the background-corrected bootstrapping-LIRA provides an important pipeline test for the reconstruc-
realizations. The final mosaics and the noise level are tion of cavities found in the intergalactic medium. The
used to generate three different frames to be stored in detection of cavity rims seen in projection against the
the final products: 1) an average adaptive X-ray surface diffuse emission from the hot intracluster and/or inter-
brightness map, 2) a noise level map, and 3) an SNR galactic medium is challenging. These large bubbles po-
map. tentially provide a useful record of interactions between
AGNs and the intergalactic medium, in which the ex-
11 Chandra Source Catalog Release 2.0: https://cxc.harvard. pansion of the associated radio lobes excavate the sur-
edu/csc2/ rounding medium (Pandge et al. 2021). Our test model
12 CIAO/vtpdetect: https://cxc.cfa.harvard.edu/ciao/ahelp/ is designed to be particularly challenging: an X-ray cav-
vtpdetect.html ity with a dominant central source representing an AGN
8

Figure 3. Fraction of false positives and false negatives in the SAUNAS detection maps derived from two truth models as a
function of the equivalent exposure time (cm2 s). Blue symbols and lines represent the fraction of false negatives, while red
represents the fraction of false positive detections in the mock maps. Cross symbols correspond to the double jet model and
filled circles represent the cavity model (see Table 1). Vertical dashed lines indicate the median equivalent exposure times for
the analyzed real observations in their respective bands.
(Blanton et al. 2001; Dunn et al. 2010). The surface the fraction of pixels that were incorrectly identified as
brightness background level of both models is fixed at false negatives (FN) and false positives (FP).
5×10−9 s−1 cm−2 arcsec−2 , and the equivalent exposure Fig. 2 demonstrates the deconvolution and smoothing
time is assumed to be flat an varying from τexp = 108 to process for a mock galaxy with τexp = 5 × 106 s, having
τexp = 104 s cm2 . For reference, τexp = 5 × 105 s cm2 , both diffuse X-ray emission and an extended PSF. The
equals ∼10 ks at 0.3 − 1.0 keV band13 . position angle selected for the model galaxy (Table 1)
is selected specifically to offer a nontrivial test for the
PSF deconvolution method. By using a position angle
The synthetic data are generated using the real of 45◦ , the resulting convolved image displays two elon-
PSF associated with the Chandra/ACIS datasets of gated features with apparently similar intensity (central
NGC 3656 (Arp 155, PID:10610775, Fabiano, G., Smith left panel in Fig. 2, PSF convolved source): one real, and
et al. 2012). This PSF, which displays the character- one created by the PSF. If the PSF elongated feature is
istic ellipsoid pattern of off-axis ACIS observations, is removed in the final images, we can conclude that the
selected as a worst-case scenario, given its extreme el- image reconstruction was successful.
lipticity due to its off-axis position in the detector array. After Poisson sampling (see Simulated observation
The readout streak14 is visible as a spike departing from panel in Fig. 2), the resulting events map is equiva-
the center of the PSF at a position angle of -70◦ approx- lent to the processed CIAO event files. The events map
imately (North = 0◦ , positive counter-clockwise). shows broad emission for the core of the galaxy model
The simulated observed events are passed to the in which the disk is indistinguishable. The two lobes are
SAUNAS pipeline for processing, followed by a compar- still present, but considerably blended with the emission
ison between the detected (3σ) maps and truth models. from the inner regions. The events are then processed
The quantitative quality test includes identification of using SAUNAS (LIRA deconvolution, Bootstrapping, and
vorbin steps).
13
The results from the PSF deconvolution (LIRA decon-
Chandra variation of effective area with energy https://cxc.
cfa.harvard.edu/proposer/POG/html/INTRO.html volved panel in Fig. 2) show a removal of most of the
14 Chandra/ACIS PSF: https://cxc.cfa.harvard.edu/ciao/PSFs/ PSF emission, recovering the signal from the disk of the
psf central.html galaxy and removing the PSF spike emission. However,
 9

Mock model Component Size µ q PA

(1) (2) (3) (4) (5) (6)

(σx , σy , pixels) [s−1 cm−2 arcsec−2 ] [◦ ]

Double jet Core 1,1 1.2×10−4 1 0
Disk 15,3 ∼ 10−6 –10−8 0.2 135
Lobes 7,7 ∼ 10−6 –10−8 1 0
Jet 25,2 5×10−7 0.08 45
Background – 5×10−9 – –
Cavity Core 1,1 1.2×10−4 1 0
Shell [30–45] ∼ 10−7 1 0
Background – 5×10−9 – –

Table 1. Photometric and structural properties of the synthetic test models. Columns: 1) Name, 2) Component, 3) Size, 4)
Surface brightness, 5) Eccentricity, 6) Position angle.
a significant amount of noise is still visible, and the back- τexp = 5 · 105 s cm2 . The reason for this increase is the
ground level is difficult to estimate (lower left panel of lack of detection of the dimmer outer regions in con-
Fig. 2). trast with the brighter core (the lobes in the case of
After applying the bootstrapping and Voronoi bin- the double jet model, and the outer shell in the cavity
ning methods, the resulting final corrected mosaic (final model). Interestingly, the fraction of false positives does
smoothed mosaic panel, Fig. 2) clearly shows the signal not increase substantially even at extremely low equiva-
from the X-ray lobes, the disk, and the central bright lent exposure times, remaining stable at ∼ 10% down to
core over the background. The 2σ and 3σ contours τexp < 104 s cm2 . This result demonstrates that even in
show the detected features following the calibration pro- cases of extremely short exposure times, SAUNAS is not
cedures described in Sec. 2, demonstrating complete re- expected to generate false positive detections, which is a
moval of the PSF streak in the final mosaics (at a 99.7% critical requirement for our study.
of confidence level). The original shape and orientation 2.3.2. Flux conservation
of the disk is recovered, with the flux correctly decon-
volved into the bright core of the model galaxy. Due to In an ideal scenario, the total flux of the events pro-
its dim brightness, the jet that connects the lobes with cessed by SAUNAS should be equal to the total flux in
the main disk is notably distorted in the final mosaic, the pre-processed frames by CIAO. In practice, the base-
but still visible at a 2σ confidence level. For this test, line model assumptions during the deconvolution pro-
the fraction of pixels unrecovered by the pipeline that cess may affect the total flux in the resulting frames.
were part of the model sources (false negatives, FN) is LIRA assumes a flat background model that–combined
F N = 3.2%. On the other hand, the fraction of misiden- with the counts in the source–tries to fit all the events
tified pixels that were part of the background (false pos- in the image. However, deviations from this ideal sce-
itives, FP) is F P = 4.0%. The maps of false positives nario (non-uniform background, regions with different
and false negatives for this test are available in Appendix exposure time) generate differences between the input
B. and output flux. In order to understand the impact of
The test for the cavity model is repeated, sampling dif- flux conservation in LIRA deconvolved images, we must
ferent equivalent exposure times. The results are shown 1) analyze the relative difference of flux before and af-
in Appendix B. Fig. 3 presents a comparison of the false ter deconvolution, and 2) determine if the residuals of
positive and false negative fraction as a function of the the deconvolution process generate any systematic arti-
equivalent exposure time and model. For equivalent ex- ficial structure (i.e., photons may be preferentially lost
posure times higher than τexp = 106 s cm2 , the FP and around bright sources, generating holes in the image or
FN are lower than 5–10%. These fractions increase to- erasing the dim signal from halos).
wards shorter exposures as expected, showing a notable Total flux conservation is tested by measuring the ra-
increase to 20% of false negatives (true source emis- tio between the total flux in the input frames (those
sion that is unrecovered by SAUNAS) at approximately obtained at the end of the CIAO pre-processing, see
Sec. 2.2.1) divided by the total flux in the final, SAUNAS
10

context, the results of LIRA are compared with those

from CIAO/arestore15 .
The results are displayed in Fig. 10 (point source)
and Fig. 11 (circular extended source) and detailed
in Appendices A and B. To quantify the quality of
the different deconvolution methods, radial surface
brightness profiles of the truth (non convolved) model,
the convolved simulated observations, and the result-
ing deconvolved maps are constructed. The profiles
show that arestore tends to oversubtract the PSF,
generating regions of negative flux around the sim-
ulated source. In the point source case scenario,
arestore oversubtracts the background by more than
5 × 10−8 s−1 cm−2 px−1 , while LIRA recovers the back-
ground level with five times less residuals. The superior-
ity of LIRA over arestore to recover diffuse structures
Figure 4. Flux conservation in SAUNAS frames. The his- is even more obvious in the extended source scenario
togram represents the probability distribution of the ratio (Fig. 11): arestore shows a clear oversubtraction ring-
between the recovered flux after SAUNAS processing and the
like region around the source, dipping the background
total flux of the input, pre-processed frames.
level to 10−7.8 s−1 cm−2 px−1 as compared to the real
processed frames. We perform this test on real (truth model) level of 10−7 s−1 cm−2 px−1 . LIRA fits
(UGC 5101, see Sec. 3.3) and synthetic observations the background level significantly more faithfully, at a
(Sec. 2.3.1). The results are shown in Fig. 4. level of ∼ 10−7.2 s−1 cm−2 px−1 .
A total flux loss of ∼ 5% is detected in the SAUNAS We conclude that LIRA deconvolution results are bet-
processed frames when compared with the pre-processed ter suited for the detection of diffuse X-ray emission,
event maps by CIAO. The results are consistent in real such as extended hot gas halos, compared to other PSF
observations (recovered flux ratio of 95.0 ± 1.7%) and correction techniques, such as CIAO’s arestore. Despite
+2.7
in synthetic observations (95.4−2.4 %). Using different the model limitations described in Sec. 2.3.2, SAUNAS
simulations, we determined that this small flux loss is suppresses false positive extended emission detections
independent of the size of the FOV (in pixels), remain- without over-fitting the PSF, while recovering the true
ing stable at ∼ 5%. For the total area of the images morphologies of X-ray hot gas distributions. Thanks
analyzed, a 5% of lost flux is negligible and well within to the modularity of SAUNAS, future updates in the
the stochastic uncertainty of typical photometry (see the LIRA deconvolution software will be automatically im-
error bars in the profiles described in Fig. 5). We con- plemented in our pipeline, improving the quality of the
sider a flux conservation ratio lower than 100% (i.e., 90% processed frames.
– 99%) as erring on the side of caution from a statistical 3. APPLICATION TO REAL OBSERVATIONS
perspective: the bias of LIRA to lose flux implies that
3.1. Sample selection
SAUNAS will not generate false positive detections of hot
gas halos. We identified two astrophysical targets of interest for
testing the pipeline:
1. NGC 3079, a highly inclined barred spiral galaxy
with a prominent Fermi bubble (Hodges-Kluck
2.3.3. Quality PSF deconvolution test et al. 2020, the primary benchmarking target, see
Sec. 3.2).
While Sec. 2.3.2 reported on the conservation of to-
2. UGC 5101, an ultra-luminous IR galaxy that is un-
tal flux in the image as a whole, this section discusses
dergoing a galactic merger (Sanders et al. 1988;
whether SAUNAS introduces unwanted artificial struc-
Imanishi et al. 2001, the secondary benchmarking
tures (fake halos, or oversubtracted regions) in the pro-
target, see Sec. 3.3).
cessed maps. For this test, two additional types of test
sources are used: 1) a point source, and 2) a circular
extended source. Both of these sources have been previ- 15 arestore: https://cxc.cfa.harvard.edu/ciao/ahelp/arestore.
ously combined with a Chandra/ACIS PSF. To provide html
 11

The targets used to demonstrate SAUNAS capabili- XMM-Newton are available for this object, NGC 3079 is
ties were selected because they were known apriori to an ideal case for benchmarking the low surface bright-
have extended soft X-ray emission detected by telescopes ness recovery capabilities of the SAUNAS pipeline.
other than Chandra (NGC 3079), and the characteriza- To detect the X-ray galactic wind in NGC 3079, the
tion of the extended emission was well-documented with same bandpass (0.3–2.0 keV) as in Hodges-Kluck et al.
a detailed methodology that could be replicated in the (2020) is used. The available Chandra/ACIS obser-
published research. Insisting that the data come from a vations of NGC 3079 are detailed in Table 2. Each
different platform provides a truth model independent visit was reprocessed with independent PSF deconvo-
of systematic effects inherently associated with Chandra. lution, and then the visits were combined for Voronoi
Finally, these specific targets were selected in order to binning. Observations 19307 and 20947 were processed
test SAUNAS against simple and complex emission struc- but discarded due to the presence of very large-scale
tures associated with the different morphologies (a disk gradients and unusually high background levels in the
galaxy and an interacting system). detectors where the main emission from NGC 3079 is
located. After processing the remaining observations
3.2. NGC 3079 (2038 and 7851) with SAUNAS, extended emission ob-
served by Chandra is compared to the results from
Large-scale bipolar winds, Fermi and radio bubbles,
XMM-Newton. The PSFs of the 2038 and 7851 observa-
are examples of extended structures observed around
tions and their unprocessed events are available in Figs.
the center of the Milky Way in multi-wavelength ob-
16 and 18 in Appendices C and D respectively.
servations, including radio (MeerKAT, S-PASS), mi-
Following the results from Fig. 2 in Hodges-Kluck
crowave (WMAP), mid-infrared (MSX ), UV (XMM),
et al. (2020), four angular cone regions display diffuse
X-rays (Chandra, XMM-Newton, ROSAT ) and gamma
emission: north-east (θ = 40◦ ), south-east (θ = 110◦ ),
rays (Fermi -LAT) (Sofue 1995; Bland-Hawthorn & Co-
south-west (θ = −140◦ ), and north-west (θ = −60◦ ),
hen 2003; Su et al. 2010; Finkbeiner 2004; Carretti et al.
(θ is measured counter-clockwise, north corresponds to
2013; Heywood et al. 2019). While the presence of these
0◦ , see Fig. 5). Mimicking the methodology in the origi-
structures is well-known in our own galaxy, Li et al.
nal article, an amplitude of ±20◦ is set for all the cones
(2019) reported the first non-thermal hard X-ray detec-
around their central axis. Surface brightness profiles are
tion of a Fermi bubble in an external galaxy, NGC 3079
generated from the reprocessed Chandra observations,
(α = 150.491◦ , δ = +55.680◦ , D = 18.68 ± 1.32 Mpc,
providing a direct comparison with previous results.
11.04 arcsec kpc−1 Springob et al. 2005), using Chan-
The results show that the extended X-ray wind emis-
dra observations. Further works in X-ray and UV using
sion is detectable using Chandra observations, up to a
XMM-Newton and GALEX revealed a 30 kpc long X-
limit of Rlim ∼ 40 kpc from the center of NGC 3079
ray Galactic Wind Cone in NGC 3079 (up to 60 kpc in
(Rlim = 39.9+4.5
−5.1 kpc on average, extending up to Rlim =
FUV, Hodges-Kluck et al. 2020), potentially associated
37.9+4.1
−4.4 kpc in the North-East filament) at a confidence
with material that has been shocked by Type II super-
level of 95% (2σ). The filament in the south-west of
novae.
the galaxy is shortest at R ∼ 16–20 kpc. Interestingly,
The length of the X-ray wind cone of NGC 3079 (R ∼
the XMM observations reveal a slightly larger extent in
3 arcmin, 16.3 kpc) contrasts with that of the bubble
the X-ray emission on the west side (40 kpc) compared
found by Li et al. (2019) using Chandra observations
to the east side (30–35 kpc) according to Hodges-Kluck
(R ≲ 0.75 arcmin, 4.1 kpc). Hodges-Kluck et al. (2020)
et al. (2020).16 . The average limiting surface brightness
argued that the sensitivity of the longest Chandra obser-
(95% confidence level) is µ = 1.66+0.5
−0.5 × 10
−10 −1
s cm−2
vations in the soft X-ray band (E < 1 keV) is affected by −2
arcsec . Limiting surface brightness reaches its lowest
the molecular contaminant buildup on the detector win-
limit when combining all the filaments, suggesting that
dow, and as a consequence, these Chandra/ACIS obser-
the observations are limited by noise and not by sys-
vations were only used for point source identification on
tematic effects (if dominated by systematic gradients, a
NGC 3079 and subsequent masking for XMM-Newton.
lower SNR would result from combining all the regions).
Additionally, the available Chandra observations were
much shallower (124.2 ks, with only 26.6 ks of usable ex-
16
posure time due to contamination) than those of XMM Note that the authors do not specify the details of their
(300.6 ks). Despite Fig. 6 in Hodges-Kluck et al. (2020) methodology for measuring the radial limits in their X-ray obser-
vations, but rather infer the dimensions of the X-ray filaments by
showing signs of faint extended emission in the Chan- visual inspection of their Fig. 1b. In this work, we adopt a 95%
dra/ACIS datasets, the authors did not attempt to char- confidence level (p = 0.05) to claim statistical significance.
acterize it. Because ancillary X-ray observations from
12

a) b)

c)

Figure 5. Extended X-ray wind cones in NGC 3079, recovered in the Chandra/ACIS observations using SAUNAS. a) Broad-band
(Chandra: 0.3–2.0 keV) surface brightness profiles of the four filaments identified by Hodges-Kluck et al. (2020) using XMM-
Newton and GALEX observations. Top to bottom: All filaments, north-east, south-east, south-west, and north-west. Radial
detection limits are in the panels (95% confidence level). b) SAUNAS processed image showing 2σ contours (black, shown in white
in panel c)) with filament sectors in yellow. The radial detection limit indicated in panel a) for each of the four filaments is shown
as solid yellow sectors, while that of “all filaments” is shown as dashed yellow, following the methodology found in Hodges-Kluck
et al. (2020). The thick magenta circle in b) shows the maximum detection limit found with XMM-Newton, compatible with
our results. c) Comparison of the optical morphology (Pan-STARRS gri) of NGC 3079 with the extended X-ray emission.
 13

the 0.3–1.0 keV emission has a size of 24.0′′ × 14.2′′

(∼ 19.1 × 11.3 kpc, position angle of 90◦ ), and a total
Obs ID Instrument Exposure Mode Count Start date
X-ray luminosity of log LX = 41.6 erg s−1 .
time rate
Given these known robust detections, we employ
(1) (2) (3) (4) (5) (6) SAUNAS in the characterization of the low surface bright-
[ks] [s−1 ]
ness emission from UGC 5101. Three bandpasses are
used, to ensure a direct comparison to the analyses by
NGC 3079
Smith et al. (2019): soft (0.3–1.0 keV), medium (1.0–
2038 ACIS-S 26.58 FAINT 10.27 2001-03-07
2.0 keV), and hard (2.0–8.0 keV). The flux conserva-
7851 ACIS-S 5.00 FAINT 14.88 2006-12-27
19307 ACIS-S 53.16 FAINT 6.14 2018-01-30
tion ratio after PSF deconvolution in this exposure is
20947 ACIS-S 44.48 FAINT 6.10 2018-02-01 96.0 ± 0.02% in the three bands. The processed X-ray
UGC 5101 emission maps are presented in Fig. 6, in comparison
2033 ACIS-S 49.32 FAINT 9.53 2001-05-31 with the optical/NIR observations from HST, as well as
ancillary radio observations for reference. The PSFs and
unprocessed events of the UGC 5101 observations in the
three bands analyzed are available in Figs. 17 and 19 in
Table 2. Chandra/ACIS observations available within Appendices C and D, respectively.
10 arcmins of NGC 3079 and UGC 5101, retrieved from the
The results are summarized in Fig. 6. The analy-
Chandra Data Archive, as of February 2024 . Columns: 1)
Observation ID, 2) Chandra instrument, 3) total exposure sis of the Chandra/ACIS observations with SAUNAS re-
time per observation, 4) observation mode, 5) average count veal that even after PSF deconvolution, the soft X-
rate, 6) exposure start date. ray emission of UGC 5101 still shows extended emis-
sion around its core. The 0.3–1.0 and 1.0–2.0 keV
bands present X-ray emission with an elongated mor-
phology, with a characteristic bright plume-like struc-
3.3. UGC 5101 ture in the core, oriented in the north-south direction
(µsoft = 1–2×10−8 s−1 cm−2 arcsec−2 ), very similar
UGC 5101 (z = 0.039, D = 161.8 Mpc,
to the results of Smith et al. (2018). In contrast,
0.784 kpc arcsec−1 , Rothberg & Joseph 2006) is an
the hard band only shows a bright core in the cen-
irregular galaxy that is undergoing a potential major
ter, compatible with an unresolved source. In the soft
merger. This object has previously been identified as
band, the diffuse X-ray emission is detectable down
a Seyfert 1.5 (Sanders et al. 1988), a LINER (low-
to levels of µsoft = 1.23+1.02
−0.66 ×10
−9 −1
s cm−2 arcsec−2
ionization nuclear emission-line region) galaxy (Veilleux
(2σ), compared to the medium band level of µmedium =
et al. 1995), and a Seyfert 2 galaxy (Yuan et al. 2010).
1.25+1.38
−0.73 ×10
−9 −1
s cm−2 arcsec−2 (2σ). Both soft and
UGC 5101 has a very extended optical tidal tail (∼
medium band emissions are centered over the main core
40 kpc) to the west from the nucleus, with a second
of UGC 5101, showing the same orientation as observed
semicircular tidal tail that surrounds the bright core of
by Smith et al. (2019). The soft band emission extends
the galaxy with a radius of 17 kpc (Surace et al. 2000).
up to 25 arcsec (20 kpc) to the north and 17 arcsec
Radio, (Lonsdale et al. 2003), IR (Genzel et al. 1998;
(13.5 kpc) to the south (3σ).
Soifer et al. 2000; Armus et al. 2007; Imanishi et al.
The spatial distribution of X-ray emission around
2001), and X-ray observations with Chandra and XMM-
UGC 5101 is generally comparable to that detected in
Newton (Ptak et al. 2003; González-Martı́n et al. 2009)
previous works (Smith et al. 2019). However, at ap-
suggest the presence of a heavily dust-obscured AGN in
proximately 40–60 arcsec radius to the north-east (α,
the nucleus of this galaxy.
δ = 143.980◦ , +61.363◦ ), the SAUNAS map reveals a
The total exposure time and other information rele-
diffuse bridge connecting with UGC 5101, at a ∼ 2σ
vant to the Chandra/ACIS observations of UGC 5101
level (µsoft ∼ 6.2 × 10−10 s−1 cm−2 arcsec−2 in the
are provided in Table 2. The diffuse X-ray emission of
soft band). For clarity, we will refer to this extended
UGC 5101 has been previously analyzed in the litera-
emission as X1. Fig. 7 displays surface brightness pro-
ture. Huo et al. (2004) found evidence for an inner hot
file analysis results and associated comparisons with
gas halo of 8.7 kpc (10.4′′ ) and an outer halo of 14.3 kpc
X1. The central surface brightness of X1 is µsoft =
(17.0′′ ). Grimes et al. (2005) found that 95% of the
4.2+1.5
−1.3 × 10
−9 −1
s cm−2 arcsec−2 in the soft band and
0.3–1.0 keV emission is enclosed in the inner 8.75 kpc
µmedium = 1.54+0.76
−0.66 × 10
−9 −1
s cm−2 arcsec−2 in the
galactocentric radius (10.5′′ ). Smith et al. (2018, 2019)
medium band. The emission of X1 is detectable at a
analyzed the Chandra/ACIS observations, finding that
14

HST/ACS Chandra/ACIS

X1

UGC 5101

F435W
F435W + F814W
F814W 0.3-1.0 keV

GMRT 150 MHz Chandra/ACIS

1.0-2.0 keV

Chandra/ACIS

2.0-8.0 keV

Figure 6. Diffuse X-ray emission of UGC 5101 as detected with SAUNAS/Chandra in the 0.3–1.0 keV band (top), 1.0–2.0 keV
band (central), and 2.0–8.0 keV band (bottom). Left: HST/ACS color image (red: F814W, green: F435W + F814W, blue:
F435W ). Upper Right: SAUNAS map of the diffuse X-ray emission, corrected for PSF effect, point-sources, and background.
Solid contours represent 3σ detections and dotted contours the 2σ detection level of X-ray emission, represented in white (left
panel) and black (right panel) for contrast. Solid red contours show GMRT 150 MHz data. White dashed ellipse represents the
previous detection limits reported by Smith et al. (2019) of UGC 5101 in the same band.
 15

3σ confidence level with a comparable angular area to pipelines for X-ray observations, such as CADET, based
UGC 5101, but with a maximum surface brightness 20– on machine-learning algorithms, allow for the identifica-
30 times lower than the main object (see Fig. 7). tion of specific source morphologies, such as X-ray cav-
Fig. 27 in Smith et al. (2019) shows a hint of ities (Plšek et al. 2024). The potential combination of
what might be emission jutting to the North-East of SAUNAS for generating low surface brightness detection
UGC 5101 where we see X1, but at a considerably lower maps with existing morphological identification and seg-
detectability. The X1 feature has not been discussed mentation software will be explored in the future.
previously in the literature as part of the UGC 5101 sys- Another limitation of the SAUNAS pipeline is the pre-
tem, but rather as a potential higher-z galaxy cluster cision of the PSF. The generation of the Chandra/ACIS
(Clerc et al. 2012; Koulouridis et al. 2021) in need of PSFs depends on multiple factors, including, but not
spectroscopic confirmation. limited to, the position of the source on the detector,
Observations of the Giant Metrewave Radio Telescope the SED of the source, or the specific parameters fed
(GMRT) 150 MHz all-sky radio survey17 (Intema et al. into the MARX simulation software (like the aspect blur).
2017, see bottom left panel in Fig. 6) confirm the de- For example, LIRA deconvolution software only accepts
tection of an adjacent source centered over the recov- one PSF for the whole image, and as a consequence,
ered X-ray emission, with a surface brightness of µ = the shape of sources at high distances from the center
10−4 Jy arcsec−2 . The GMRT flux maps are shown as of the image might be inaccurate. This phenomena can
contours in Fig. 6, revealing a peak of radio emission over cause residuals if observations present bright sources at
the center of X1 in addition to UGC 5101. GALEX UV high angular distances from the center of the source,
observations provide a near-ultraviolet (NUV) flux of since the deconvolution will be based on the PSF at
5.14±0.15×10−6 Jy (Seibert et al. 2012) but only upper the center of the observation, but not at the location of
limits in the far-ultraviolet (FUV) band (9.8 · 10−6 Jy). the secondary contaminating source. As an attempt to
Recent JWST observations (GO 1717, PI: Vivian U., quantify this effect, we estimate in Fig. 8 the variation
MIRI) of UGC 5101 were inspected for this work, but of the PSF size (R90% , radius that contains 90% of the
they suffer from extreme saturation of the bright core flux of a point source) vs. angular separation to the
of the galaxy, and the outer X-ray emitting region lies source using CIAO psfsize srcs18 , based on the Chan-
outside the footprint, so they were discarded for this dra/ACIS observations on UGC 5101. The results show
study. While investigating the nature of this extended that the PSF increases a factor of ×2 in ∼2 arcmin (×10
X-ray emission is beyond the scope of this paper focused in ∼10 arcmin). In our science cases, no bright object
on the presentation of the SAUNAS pipeline, we briefly was observed in the environment of the main sources
discuss the main hypotheses (hot gas plume or high-z (NGC 3079, UGC 5101), so the main contributors to the
galaxy cluster) in Sec. 4. scattered light are the sources for which the PSF was
calculated. However, observers must be wary of strong
4. DISCUSSION residual PSF wings from nearby sources at ∼2 arcmin
and longer distances. While a complete analysis of the
4.1. Limitations uncertainties of the PSF in Chandra is out of the scope
We have demonstrated the SAUNAS methodology to be of the current paper, we refer to the Appendix in Ma
successful in recovering dim, extended surface bright- et al. (2023) for a review in the field.
ness X-ray features under low signal-to-noise ratio con-
ditions through performance tests using both synthetic 4.2. NGC 3079
(Section 2.3) and real (Section 3) X-ray datasets. The analysis of the Chandra/ACIS observations in
There are, however, several limitations of SAUNAS in the field of NGC 3079 revealed signs of X-ray wind
its current form that will be addressed in future versions out to galactocentric distances R ∼ 30 kpc, compat-
of the pipeline. Among them, SAUNAS does not attempt ible with previous observations using XMM–Newton
to provide a quantitative separation between extended (Hodges-Kluck et al. 2020). While XMM–Newton is
sources, such as a segmentation map. Deblending of able to trace the extended X-ray emission out to larger
extended X-ray sources is one of the main objectives distances (∼40 kpc) in some directions, some considera-
of a complementary code, EXSdetect (Liu et al. 2013), tions must be made in order to compare XMM–Newton
using a friend-of-friends algorithm. Other specialized results with the benchmark study provided here:

17 TIFR GMRT Sky Survey (TGSS) Archive: https://vo.astron. 18 CIAO/psfsize srcs: https://cxc.cfa.harvard.edu/ciao/ahelp/
nl/tgssadr/q fits/imgs/form psfsize srcs.html
16

Figure 7. Surface brightness profiles of the diffuse X-ray emission of UGC 5101 and the extended diffuse north-east source (X1)
detected with SAUNAS/Chandra in the 0.3–1.0 and 1.0–2.0 keV bands. Radially averaged surface brightness profile (blue upward
triangles: 0.3–1.0 keV band, purple downward triangles: 1.0–2.0 keV band). Shaded areas represent the 1σ and 2σ error bars.
Solid blue and dashed purple vertical lines represent the 2σ detection limits for the 0.3–1.0 keV and the 1.0–2.0 keV bands.
Blue and purple stars show the average surface brightness of the north-east extended emission X1, represented at the measured
galactocentric distance from UGC 5101.
the usable time in Chandra/ACIS (26.6 ks) obser-
vations.
2. XMM-Newton has a larger effective area
(4650 cm2 at 1 keV) than Chandra (555 cm2 ),
at the expense of a lower spatial resolution19
(XMM-Newton/FWHM = 6 arcsec vs. Chan-
dra/FWHM = 0.2 arcsec). While the aperture is
smaller, proper masking of point sources improves
detectability of dim structures by reducing the
background noise.
3. The analysis of the X-ray emission by Hodges-
Kluck et al. (2020) is based on the inspection of
the quadrant stacked images with a certain sig-
nal and radial threshold (see their Fig. 4, central
panel). The methodology they use to calculate the
limiting radius of the diffuse X-ray emission is not
clearly stated in their analysis, making a direct
and accurate comparison of results difficult.
Figure 8. Variation of the Chandra/ACIS PSF size as a
function of the angular separation to the center of the FOV. Despite the differences of the detection methods, we
Vertical axis: Radius enclosing 90% of the flux from the PSF conclude that SAUNAS is able to recover extended, low
at 1.0 keV, based on the observations of UGC 5101. Hori- surface brightness X-ray emission using Chandra/ACIS
zontal axis: Angular separation to the center of the source, X-ray observations of NGC 3079, in excellent agreement
approximately the center optical axis. The horizontal dotted with the deeper exposure taken by XMM–Newton.
lines mark the PSF sizes that correspond to ×2, ×5, ×10,
and ×20 the PSF size at its center (×1). 4.3. UGC 5101

1. XMM–Newton observations of NGC 3079 combine

an ∼11 times longer exposure time (300.6 ks) than 19 https://xmm-tools.cosmos.esa.int/external/xmm user
support/documentation/uhb/xmmcomp.html
 17

higher-z. In fact, the center of the Chandra/ACIS X-

ray emission overlaps remarkably well with that of a
background galaxy. Fig. 9 shows the HST/ACS imag-
ing (bands) centered over X1, with the soft-band X-ray
emission contours overlapped for reference. The peak
of X-ray emission is coincident with the position of a
background galaxy (WISE J093555.43+612148.0). Un-
fortunately, WISE J093555.43+612148.0 does not have
spectroscopic or photometric redshifts available.
While resolving the nature of X1 is beyond the scope
of this paper, we conclude that the test performed
with the Chandra/ACIS observations of UGC 5101 us-
ing SAUNAS demonstrates the pipeline’s capabilities
in successfully producing adaptively smoothed, PSF-
deconvolved X-ray images in different bands. The image
reduction process presented here allows for a better cali-
bration of the background to recover details at both high
resolution and surface brightness (inner core structure of
Figure 9. Hubble Space Telescope ACS imaging over the the merging galaxy) as well as extended ultra-low sur-
diffuse extended emission X1 found adjacent to UGC 5101. face brightness regions, such as the previously unknown
Pseudo-RGB color combination: Blue: F435W, Green: extended emission around UGC 5101.
F435W + F814W. Red: F814W. Grey contours represent
the (3, 5, 7, and 10) σ detection levels obtained in the 0.3– 5. CONCLUSIONS
1.0 keV band from Chandra/ACIS observations, processed
with SAUNAS. Notice the merger shell structure of UGC 5101 In this paper we have presented SAUNAS: a pipeline
at the bottom right corner. to detect extended, low surface brightness structures
on Chandra X-ray Observations. SAUNAS automatically
Section 3.3 described evidence for extended low queries the Chandra Archive, reduces the observations
surface brightness emission (X1, µsoft = 4.2+1.5 −1.3 × through the CIAO pipeline, generates PSF models and
10−9 s−1 cm−2 arcsec−2 , 0.3–1.0 keV) located in the deconvolves the images, identifing and masking point
north-east of the UGC 5101 merging galaxy. X1 has sources, and generating adaptative smoothed surface
been previously detected in X-ray by Smith et al. (2019) brightness and detection SNR maps for the sources in
but its emission was not discussed nor treated as part of the final mosaics. We have demonstrated through tests
UGC 5101’s outskirts. Other works (Clerc et al. 2012; on simulated data and comparisons to published results
Koulouridis et al. 2021) tentatively classified X1 as a that the SAUNAS pipeline distinguishes itself from other
potential background galaxy cluster, but this feature re- existing X-ray pipelines by meeting the following main
mained unconfirmed as spectroscopic observations are objectives:
unavailable. X1 is detected also in GMRT 150 Mhz ob-
1. Generate X-ray detection maps for extended
servations as a secondary source adjacent to UGC 5101,
sources in a consistent, statistically reproducible
confirming the existence of a feature at this location.
way.
Two main hypotheses regarding the nature of X1 are:
2. Provide a modular framework for reduction of
1. X1 is part of the extended X-ray emitting envelope Chandra/ACIS observations focusing on the de-
of UGC 5101. tection of faint extended sources, simplifying the
access to X-ray archival observations for multi-
2. X1 is a background source, potentially the ex- wavelength studies.
tended envelope of a higher-z object, such as a
massive early-type galaxy or a cluster. Our approach to meeting these objectives is to assess
the statistical probability that signal in low-count areas
Although the X-ray emission in the soft and medium is real. This strategy can both produce detections of
bands of X1 is adjacent to that of UGC 5101, and both previously-overlooked diffuse emission as well as mini-
objects have a dominant emission in the soft band com- mize false positive detections of extended hot gas emis-
pared to the medium and hard (see Figs. 6 and 7, sion. In Sec. 3, we compare SAUNAS-processed archival
the emission could still be part of a hot gas halo at Chandra/ACIS data to published results. This section
18

demonstrates that the proposed methodology succeeds

in recovering the extended emission detected in a selec- 1 The authors thank to the anonymous referee for the pro-
tion of local Universe targets. While the CIAO pipeline 2 vided input that helped to improve this publication sig-
provides a canonical and highly efficient procedure to re- 3 nificantly. The SAUNAS X-ray surface brightness and
duce the Chandra observations, the secondary analysis 4 signal-to-noise maps of NGC 3079 and UGC 5101 are
of the resulting event files is usually performed in an in- 5 publicly available in Zenodo in FITS format: DOI:
dependent way by the observers. Such a situation results 6 10.5281/zenodo.10892485. The list of Chandra datasets,
in two suboptimal consequences: 1) Most X-ray stud- 7 obtained by the Chandra X-ray Observatory, are con-
ies are focused on single objects, or very small samples 8 tained in DOI: 10.25574/cdc.225. A.B. acknowledges
(three or four objects), and 2) most studies develop their 9 the tireless support from Nicholas Lee, Tara Gokas,
own procedure to correct the PSF effects (if considered), 10 Kenny Glotfelty, Catherine Cranmer, and the rest of
to generate smoothed maps, and to determine the signif- 11 the CXC Helpdesk team. Without your dedication,
icance of emission over the background. Planned future 12 this project would have not been possible. This re-
work includes an analysis of the extended emission of 13 search has made use of data obtained from the Chan-
nearby galaxies using Chandra/ACIS archival data, and 14 dra Data Archive and the Chandra Source Catalog, and
releasing the tools to the astronomical community. In 15 software provided by the Chandra X-ray Center (CXC)
this first article, we made the processed maps available20 16 in the application packages CIAO (Fruscione et al. 2006)
for the community through the Zenodo open repository. 17 and Sherpa (Freeman et al. 2001). A.S. acknowledge
A benefit of the automated functionality provided by 18 support from NASA contract to the Chandra X-ray
this tool is its provision of straightforward access to 19 Center NAS8-03060. The work conducted at NASA
high-level archival Chandra products and facilitation of 20 Ames Research Center was funded through NASA’s
their use in multi-wavelength studies. In future works 21 NNH22ZDA001N Astrophysics Data and Analysis Pro-
of this series (Borlaff et al. in prep.) we will explore the 22 gram under Award 22-ADAP22-0118. This work was
X-ray emission of a sample of targets using the SAUNAS 23 authored by an employee of Caltech/IPAC under Con-
pipeline, focusing on the evolution of lenticular galax- 24 tract No. 80GSFC21R0032 with the National Aeronau-
ies based on Chandra/ACIS data in combination with 25 tics and Space Administration. This paper represents
Hubble and Spitzer observations. The serendipitous dis- 26 the views of the authors only and should not be inter-
covery presented in this work in one of the galaxies 27 preted as representing the views of ConstructConnect,
studied; UGC 5101, an on-going merger galaxy, demon- 28 Inc.
strate that the combination of multi-wavelength legacy
archives, such as those of Chandra, GMRT, and Hub-
ble, may already hold the information to disentangle the Facilities: Chandra, HST
impact of the different evolutionary drivers in galaxies.
Software: Matplotlib (Hunter 2007), CIAO,
astropy (Astropy Collaboration et al. 2018, 2013, 2022),
LIRA (Donath et al. 2022a)21 , VorBin (Cappellari &
Copin 2003)

APPENDIX

A. PSF DECONVOLUTION EFFICIENCY TEST

In this section, a set of synthetic observations generated with CIAO/MARX22 are used to evaluate the reliability of
the SAUNAS algorithm when applied to a simple point source. SAUNAS’s ability to accurately recover diffuse emission is
significantly governed by limitations imposed by LIRA, the associated deconvolution tool. SAUNAS could have instead
utilized the widely-used and proven arestore tool, which can restore emission structures down to scales comparable
to the Chandra/ACIS resolution (0.492′′ ). Here we benchmark these two PSF deconvolution methodologies using
simulated observations of an unresolved object constructed by convolving a point source with a highly off-axis PSF

20 The SAUNAS X-ray surface brightness maps of NGC 3079 and

UGC 5101 are publicly available in Zenodo: https://zenodo.org/
records/10892485.
21 pyLIRA: https://github.com/astrostat/pylira
22 Using MARX to Simulate an Existing Observation: https://
cxc.cfa.harvard.edu/ciao/threads/marx sim/
 19

from Chandra/ACIS, associated with the observations of 3C 264 (NGC 3862, α = 176.2709◦ , δ = +19.6063◦ , Obs. ID:
514). The simulated observations processed with SAUNAS (Sec. 2.2) are compared to the results produced by standard
application of arestore. Both methods use the same number of iterations (Niter = 1000). For each method, surface
brightness profiles are constructed from Voronoi binning of the deconvolved data and compared to that of the model
point source.
The results are shown in Fig. 10. The PSF convolved point source shows the characteristic elliptical shape of the
off-axis PSF from Chandra/ACIS. The surface brightness profiles obtained from the images show that CIAO/arestore
provides output images with more flux at their core than SAUNAS. However, CIAO/arestore’s deconvolved image
has a higher noise in the surroundings of the center (R = [0 − 10] px) than SAUNAS, including some clear signs
of oversubtraction (see the Voronoi bins at the bottom right image) around the center of the object. In addition,
CIAO/arestore leave a characteristic residual at larger distances (R = [10 − 20] px) that could easily be confused
with a shell of extended X-ray emission. In contrast, SAUNAS provides a deconvolved image with less central flux but a
smoother transition to the background level and without the presence of residual halos of emission or oversubtraction.
We conclude that CIAO/arestore concentrates more signal into a single point source at the expense of higher noise in
the resulting images when compared to the methodology utilized by SAUNAS described in Sec. 2.2.
B. SAUNAS EXTENDED TEST MODELS
Appendix A demonstrated that the combination of LIRA + Bootstrapping methods adopted in the SAUNAS pipeline
provides a more accurate representation of the real distribution of light compared to CIAO/arestore, including avoiding
arestore’s PSF over-subtraction. Given that the main aim of SAUNAS is the detection of extended sources, we extend
the analysis from Appendix A to SAUNAS processing of an extended source model.
Figure 11 shows the result from this analysis. A simulated source with a central surface brightness of
µ =10−3 s−1 px−1 and a background level of µ =10−7 s−1 px−1 is convolved with the same PSF used by the point
source tests described in Appendix A. The resulting event file of convolved data is then processed by SAUNAS and
deconvolved by a standard application of CIAO/arestore. A comparison of the associated surface brightness profiles
provides both quantitative and qualitative assessments of the different light reconstruction methods.
The top right panel of Fig. 11 shows that the methodology adopted in SAUNAS produces a result that is more closely
aligned with our science-driven requirements. Proper treatment of the fainter regions surrounding objects is a critical
factor for the detection of faint extended emission, such as hot gas X-ray halos around galaxies. While SAUNAS produces
a well-behaved profile that smoothly transitions to the background level at large radii, CIAO/arestore manufactures
an over-subtracted background region surrounding the object, similar to its treatment of point sources (Appendix A).
Figures 12 and 13 show the results of the false positive / false negative quality test described in Sec. 2.3.1 for the
double jet model. In Figs. 14 and 15 the equivalent results are shown for the cavity model. Each row represents
different equivalent exposure times, from τexp = 5 × 107 s cm2 to τexp = 5 × 104 s cm2 . We refer to the caption in the
figures for details.
C. NGC 3079 AND UGC 5101 POINT SPREAD FUNCTION
This section presents the PSFs generated for the NGC 3079 (see Sec. 3.2) and UGC 5101 (see Sec. 3.3), Chandra/ACIS
observations. The PSFs were generated using MARX as described in Sec. 2.2. The panels in Figs. 16 and Fig. 17 show
the different PSFs obtained for the three bands (0.3–1.0 keV, 1.0–2.0 kev, and 2.0–8.0 kev) in UGC 5101, and for the
two datasets analyzed in the broadband (0.3–2.0 keV) for NGC 3079.
D. NGC 3079 AND UGC 5101 EVENT MAPS
This section presents the event maps as observed by Chandra/ACIS and processed by CIAO for the NGC 3079 (see
Sec. 3.2) and UGC 5101 (see Sec. 3.3) observations. Note that the events in the panels represent the raw event counts
without any SAUNAS processing, and thus they include contamination by sky background, gradients generated by the
different equivalent exposure time across the field of view, and point source contamination. The panels in Fig. 18 show
the events obtained in the Chandra 2038 and 7851 visits to NGC 3079 in the 0.3–2.0 keV broadband, and Fig. 19 show
the events obtained for the three bands (0.3–1.0 keV, 1.0–2.0 kev, and 2.0–8.0 kev) in UGC 5101.

REFERENCES
Aguerri, J. A. L., Agulli, I., Diaferio, A., & Dalla Vecchia, Anderson, M. E., & Bregman, J. N. 2011, ApJ, 737, 22,
C. 2017, MNRAS, 468, 364, doi: 10.1093/mnras/stx457 doi: 10.1088/0004-637X/737/1/22
20

PSF convolved point source

SAUNAS CIAO/arestore

Figure 10. Point-source PSF deconvolution test. Top left panel: Emission from a point-source object and a background level
of µ =10−7 s−1 px−1 convolved with a reference Chandra/ACIS PSF (NGC 3862, α = 176.2709◦ , δ = +19.6063◦ , Obs. ID: 514).
Top right panel: Surface brightness profiles of the convolved source (grey solid line), SAUNAS deconvolved image (blue dashed),
and CIAO/arestore deconvolved (orange dotted) images. The horizontal dotted line represents the sky background of the model.
Bottom left panel: SAUNAS deconvolved image. Bottom right panel: CIAO/arestore deconvolved image. Note that the convolved
image (events map) and the CIAO/arestore deconvolved image have been processed using Voronoi binning for visualization
purposes of the surface brightness. See the legend and colorbar in the figure.
Anderson, M. E., Bregman, J. N., & Dai, X. 2013, ApJ, Bartalucci, I., Mazzotta, P., Bourdin, H., & Vikhlinin, A.
762, 106, doi: 10.1088/0004-637X/762/2/106 2014, A&A, 566, A25, doi: 10.1051/0004-6361/201423443
Armus, L., Charmandaris, V., Bernard-Salas, J., et al. Benson, A. J., & Devereux, N. 2010, MNRAS, 402, 2321,
2007, ApJ, 656, 148, doi: 10.1086/510107 doi: 10.1111/j.1365-2966.2009.16089.x
Astropy Collaboration, Robitaille, T. P., Tollerud, E. J., Binney, J. 1977, ApJ, 215, 483, doi: 10.1086/155378
et al. 2013, A&A, 558, A33, Bland-Hawthorn, J., & Cohen, M. 2003, ApJ, 582, 246,
doi: 10.1051/0004-6361/201322068 doi: 10.1086/344573
Astropy Collaboration, Price-Whelan, A. M., Sipőcz, B. M., Blanton, E. L., Sarazin, C. L., McNamara, B. R., & Wise,
et al. 2018, AJ, 156, 123, doi: 10.3847/1538-3881/aabc4f M. W. 2001, ApJ, 558, L15, doi: 10.1086/323269
Astropy Collaboration, Price-Whelan, A. M., Lim, P. L., Cappellari, M., & Copin, Y. 2003, MNRAS, 342, 345,
et al. 2022, ApJ, 935, 167, doi: 10.3847/1538-4357/ac7c74 doi: 10.1046/j.1365-8711.2003.06541.x
 21

PSF convolved circular source

SAUNAS CIAO/arestore

Figure 11. Extended source PSF deconvolution test. Top left panel: Emission from a circular source object with a central
surface brightness of µ =10−3 s−1 px−1 and a background level of µ =10−7 s−1 px−1 convolved with a reference Chandra/ACIS
PSF (NGC 3862, α = 176.2709◦ , δ = +19.6063◦ , Obs. ID: 514). Top right panel: Surface brightness profiles of the ground-
truth (non-PSF convolved) test source (red solid line), the PSF-convolved source (grey dashed-dotted line), SAUNAS deconvolved
image (blue dashed), and CIAO/arestore deconvolved (orange dotted) images. The horizontal dotted line represents the sky
background of the model, and the vertical dotted line represents the radial limit of the circular test source (R = 15 pixels).
Bottom left panel: SAUNAS deconvolved image. Bottom right panel: CIAO/arestore deconvolved image. Note that the convolved
image (events map) and the CIAO/arestore deconvolved image have been processed using Voronoi binning for visualization
purposes of the surface brightness. See the legend and colorbar in the figure.
Carretti, E., Crocker, R. M., Staveley-Smith, L., et al. 2013, Diehl, S., & Statler, T. S. 2008, ApJ, 680, 897,
Nature, 493, 66, doi: 10.1038/nature11734 doi: 10.1086/587481
Clerc, N., Sadibekova, T., Pierre, M., et al. 2012, MNRAS, Donath, A., Siemiginowska, A., Kashyap, V., et al. 2022a,
423, 3561, doi: 10.1111/j.1365-2966.2012.21153.x
in Proceedings of the 21st Python in Science Conference,
Connors, A., Stein, N. M., van Dyk, D., Kashyap, V., &
98–104, doi: 10.25080/majora-212e5952-00f
Siemiginowska, A. 2011, in Astronomical Society of the
Pacific Conference Series, Vol. 442, Astronomical Data Donath, A., Siemiginowska, A., Kashyap, V., & Solipuram,
Analysis Software and Systems XX, ed. I. N. Evans, K. R. 2022b, in AAS/High Energy Astrophysics Division,
A. Accomazzi, D. J. Mink, & A. H. Rots, 463 Vol. 54, AAS/High Energy Astrophysics Division, 108.14
22 7 2
Events τexp = 10 s cm τexp = 5 × 106 s cm2 τexp = 106 s cm2

Final

False positive/negative

Figure 12. SAUNAS processing test using the double jet model as a function of the equivalent exposure time (see Sec. 2.3.1
and Table 1). Top row shows the simulated event images. Middle row shows the final recovered surface brightness maps after
processing with SAUNAS. Dashed contours represent the 3σ and dotted contours the 2σ detection level of X-ray emission. Bottom
row represents the false positive (red) and false negative (blue) detection maps for each simulation (see Sec. 2.3.1). From left
to right columns, the equivalent exposure times are τexp = 107 , 5 × 106 , and 106 s cm2 . See the labels in the panels. Colorbars
represent the number of events per pixel (event images) and the surface brightness flux (final mosaics).
Dunn, R. J. H., Allen, S. W., Taylor, G. B., et al. 2010, Freeman, P., Doe, S., & Siemiginowska, A. 2001, in Society
MNRAS, 404, 180, doi: 10.1111/j.1365-2966.2010.16314.x of Photo-Optical Instrumentation Engineers (SPIE)
Ebeling, H., & Wiedenmann, G. 1993, PhRvE, 47, 704, Conference Series, Vol. 4477, Astronomical Data
doi: 10.1103/PhysRevE.47.704 Analysis, ed. J.-L. Starck & F. D. Murtagh, 76–87,
doi: 10.1117/12.447161
Evans, I. N., Primini, F. A., Glotfelty, K. J., et al. 2010,
Fruscione, A., McDowell, J. C., Allen, G. E., et al. 2006, in
ApJ, 189, 37, doi: 10.1088/0067-0049/189/1/37
Society of Photo-Optical Instrumentation Engineers
Fabbiano, G. 2019, X-Rays from Galaxies, ed. B. Wilkes &
(SPIE) Conference Series, Vol. 6270, Observatory
W. Tucker, 7–1, doi: 10.1088/2514-3433/ab43dcch7
Operations: Strategies, Processes, and Systems, ed. D. R.
Fabbiano, G., Gioia, I. M., & Trinchieri, G. 1989, ApJ, 347, Silva & R. E. Doxsey, 62701V, doi: 10.1117/12.671760
127, doi: 10.1086/168103 Genzel, R., Lutz, D., Sturm, E., et al. 1998, ApJ, 498, 579,
Finkbeiner, D. P. 2004, ApJ, 614, 186, doi: 10.1086/423482 doi: 10.1086/305576
 5 2
23
Events τexp = 5 × 10 s cm τexp = 105 s cm2 τexp = 5 × 104 s cm2

Final

False positive/negative

Figure 13. (Continuation of Fig. 12) SAUNAS processing test using the double jet model as a function of the equivalent exposure
time (see Sec. 2.3.1 and Table 1). Top row shows the simulated event images. Middle row shows the final recovered surface
brightness maps after processing with SAUNAS. Dashed contours represent the 3σ and dotted contours the 2σ detection level of
X-ray emission. Bottom row represents the false positive (red) and false negative (blue) detection maps for each simulation (see
Sec. 2.3.1). From left to right columns, the equivalent exposure times are τexp = 5 · 105 , 105 , and 5 × 104 s cm2 . See the labels
in the panels. Colorbars represent the number of events per pixel (event images) and the surface brightness flux (final mosaics).
González-Martı́n, O., Masegosa, J., Márquez, I., Guainazzi, Hodges-Kluck, E. J., Yukita, M., Tanner, R., et al. 2020,
M., & Jiménez-Bailón, E. 2009, A&A, 506, 1107, ApJ, 903, 35, doi: 10.3847/1538-4357/abb884
doi: 10.1051/0004-6361/200912288 Hopkins, P. F., Quataert, E., & Murray, N. 2012, MNRAS,
Grimes, J. P., Heckman, T., Strickland, D., & Ptak, A. 421, 3488, doi: 10.1111/j.1365-2966.2012.20578.x
2005, ApJ, 628, 187, doi: 10.1086/430692 Hunter, J. D. 2007, Computing in Science & Engineering, 9,
90, doi: 10.1109/MCSE.2007.55
Hardcastle, M. J., Kraft, R. P., Sivakoff, G. R., et al. 2007,
Huo, Z. Y., Xia, X. Y., Xue, S. J., Mao, S., & Deng, Z. G.
ApJ, 670, L81, doi: 10.1086/524197
2004, ApJ, 611, 208, doi: 10.1086/422164
Henley, D. B., Shelton, R. L., Kwak, K., Joung, M. R., &
Imanishi, M., Dudley, C. C., & Maloney, P. R. 2001, ApJ,
Mac Low, M.-M. 2010, ApJ, 723, 935, 558, L93, doi: 10.1086/323636
doi: 10.1088/0004-637X/723/1/935 Intema, H. T., Jagannathan, P., Mooley, K. P., & Frail,
Heywood, I., Camilo, F., Cotton, W. D., et al. 2019, D. A. 2017, A&A, 598, A78,
Nature, 573, 235, doi: 10.1038/s41586-019-1532-5 doi: 10.1051/0004-6361/201628536
24 7 2
Events τexp = 10 s cm τexp = 5 × 106 s cm2 τexp = 106 s cm2

Final

False positive/negative

Figure 14. SAUNAS processing test using the cavity model as a function of the equivalent exposure time (see Sec. 2.3.1 and Table
1). Top row shows the simulated event images. Middle row shows the final recovered surface brightness maps after processing
with SAUNAS. Dashed contours represent the 3σ and dotted contours the 2σ detection level of X-ray emission. Bottom row
represents the false positive (red) and false negative (blue) detection maps for each simulation (see Sec. 2.3.1). From left to right
columns, the equivalent exposure times are τexp = 107 , 5 × 106 , and 106 s cm2 . See the labels in the panels. Colorbars represent
the number of events per pixel (event images) and the surface brightness flux (final mosaics).
Kelly, A. J., Jenkins, A., & Frenk, C. S. 2021, MNRAS, Mulchaey, J. S. 2000, ARA&A, 38, 289,
502, 2934, doi: 10.1093/mnras/stab255 doi: 10.1146/annurev.astro.38.1.289
Koulouridis, E., Clerc, N., Sadibekova, T., et al. 2021, Oke, J. B. 1971, ApJ, 170, 193, doi: 10.1086/151202
A&A, 652, A12, doi: 10.1051/0004-6361/202140566 O’Sullivan, E., Forbes, D. A., & Ponman, T. J. 2001,
MNRAS, 328, 461, doi: 10.1046/j.1365-8711.2001.04890.x
Li, J.-T., Hodges-Kluck, E., Stein, Y., et al. 2019, ApJ, 873,
Pandge, M. B., Sebastian, B., Seth, R., & Raychaudhury,
27, doi: 10.3847/1538-4357/ab010a
S. 2021, MNRAS, 504, 1644, doi: 10.1093/mnras/stab384
Liu, T., Tozzi, P., Tundo, E., et al. 2013, A&A, 549, A143,
Plšek, T., Werner, N., Topinka, M., & Simionescu, A. 2024,
doi: 10.1051/0004-6361/201219866
MNRAS, 527, 3315, doi: 10.1093/mnras/stad3371
Lonsdale, C. J., Lonsdale, C. J., Smith, H. E., & Diamond, Ptak, A., Heckman, T., Levenson, N. A., Weaver, K., &
P. J. 2003, ApJ, 592, 804, doi: 10.1086/375778 Strickland, D. 2003, ApJ, 592, 782, doi: 10.1086/375766
Ma, J., Elvis, M., Fabbiano, G., et al. 2023, ApJ, 948, 61, Rees, M. J., & Ostriker, J. P. 1977, MNRAS, 179, 541,
doi: 10.3847/1538-4357/acba8d doi: 10.1093/mnras/179.4.541
 5 2
25
Events τexp = 5 · 10 s cm τexp = 105 s cm2 τexp = 5 · 104 s cm2

Final

False positive/negative

Figure 15. (Continuation of Fig. 14) SAUNAS processing test using the double jet model as a function of the equivalent exposure
time (see Sec. 2.3.1 and Table 1). Top row shows the simulated event images. Middle row shows the final recovered surface
brightness maps after processing with SAUNAS. Dashed contours represent the 3σ and dotted contours the 2σ detection level of
X-ray emission. Bottom row represents the false positive (red) and false negative (blue) detection maps for each simulation (see
Sec. 2.3.1). From left to right columns, the equivalent exposure times are τexp = 5 × 105 , 105 , and 5 × 104 s cm2 . See the labels
in the panels. Colorbars represent the number of events per pixel (event images) and the surface brightness flux (final mosaics).
Rothberg, B., & Joseph, R. D. 2006, AJ, 131, 185, Seibert, M., Wyder, T., Neill, J., et al. 2012, in American
doi: 10.1086/498452 Astronomical Society Meeting Abstracts, Vol. 219,
American Astronomical Society Meeting Abstracts #219,
Sanders, D. B., Soifer, B. T., Elias, J. H., et al. 1988, ApJ,
340.01
325, 74, doi: 10.1086/165983 She, R., Ho, L. C., & Feng, H. 2017, ApJ, 835, 223,
Sandin, C. 2014, A&A, 567, A97, doi: 10.3847/1538-4357/835/2/223
Silk, J. 1977, ApJ, 211, 638, doi: 10.1086/154972
doi: 10.1051/0004-6361/201423429
Smith, B. J., Campbell, K., Struck, C., et al. 2018, AJ, 155,
—. 2015, A&A, 577, A106, 81, doi: 10.3847/1538-3881/aaa1a6
doi: 10.1051/0004-6361/201425168 Smith, B. J., Swartz, D. A., Miller, O., et al. 2012, AJ, 143,
144, doi: 10.1088/0004-6256/143/6/144
Sato, S., Akimoto, F., Furuzawa, A., et al. 2000, ApJ, 537,
Smith, B. J., Wagstaff, P., Struck, C., et al. 2019, AJ, 158,
L73, doi: 10.1086/312772 169, doi: 10.3847/1538-3881/ab3e72
26

ObsID: 2038 ObsID: 7851

Figure 16. Point Spread Functions (PSF) of NGC 3079 images processed with SAUNAS. Left panel: 1) PSF of the visit 2038
to NGC 3079 on the broad 0.3-2.0 keV band. 2) PSF of the visit 7851 to NGC 3079 on the same broad 0.3-2.0 keV band. The
binning (pixelscale) for the NGC 3079 images and the PSFs is 8 × 8 (3.936 arcsec px−1 ). Notice the logarithmic color scale in
the bottom of the panels.

0.3 – 1.0 keV 1.0 – 2.0 keV 2.0 – 8.0 keV

Figure 17. Point Spread Functions (PSF) of UGC 5101 images processed with SAUNAS. From left to right: 1) PSF of the 0.3–1.0
keV (soft) band. 2) PSF of the 1.0–2.0 keV (medium) band. 3) PSF of the 2.0–8.0 keV (hard) band. The binning (pixelscale)
for the UGC 5101 images and the PSFs is 1 × 1 (0.492 arcsec px−1 ). Notice the logarithmic color scale in the bottom of the
panels.
Sofue, Y. 1995, PASJ, 47, 527, Su, M., Slatyer, T. R., & Finkbeiner, D. P. 2010, ApJ, 724,
doi: 10.48550/arXiv.astro-ph/9508110 1044, doi: 10.1088/0004-637X/724/2/1044
Soifer, B. T., Neugebauer, G., Matthews, K., et al. 2000,
Surace, J. A., Sanders, D. B., & Evans, A. S. 2000, ApJ,
AJ, 119, 509, doi: 10.1086/301233
529, 170, doi: 10.1086/308247
Spergel, D. N., Bean, R., Doré, O., et al. 2007, ApJ, 170,
377, doi: 10.1086/513700 Veilleux, S., Kim, D. C., Sanders, D. B., Mazzarella, J. M.,
Springob, C. M., Haynes, M. P., Giovanelli, R., & Kent, & Soifer, B. T. 1995, ApJ, 98, 171, doi: 10.1086/192158
B. R. 2005, ApJ, 160, 149, doi: 10.1086/431550
Wang, Q. D. 2012, in The Spectral Energy Distribution of
Strickland, D. K., Heckman, T. M., Colbert, E. J. M.,
Galaxies - SED 2011, ed. R. J. Tuffs & C. C. Popescu,
Hoopes, C. G., & Weaver, K. A. 2004, ApJ, 151, 193,
doi: 10.1086/382214 Vol. 284, 183–192, doi: 10.1017/S1743921312009039
 27

NGC 3079 - ObsID: 2038 - 0.3–2.0 keV NGC 3079 - ObsID: 7851 - 0.3–2.0 keV

Figure 18. Event maps of the NGC 3079 observations, before processing with SAUNAS. Left: Observation ID 2038, 0.3–2.0 keV.
Right: Observation ID 7851, 0.3–2.0 keV. The binning (pixelscale) for the NGC 3079 images is 8 × 8 (3.936 arcsec px−1 ). Solid
black contours represent the 3σ and dashed contours the 2σ detection level of X-ray emission.

UGC 5101 - 0.3–1.0 keV UGC 5101 - 1.0–2.0 keV UGC 5101 - 2.0–8.0 keV

Figure 19. Event maps of the UGC 5101 observations, before processing with SAUNAS. Left to right: 1) UGC 3079 on the
0.3-1.0 keV band. 2)UGC 3079 on the 1.0-2.0 keV band. 3) UGC 3079 on the 2.0-8.0 keV band. The binning (pixelscale) for
the UGC 5101 images is 1 × 1 (0.492 arcsec px−1 ). Solid black contours represent the 3σ and dashed contours the 2σ detection
level of X-ray emission.
Weisskopf, M. C., Tananbaum, H. D., Van Speybroeck, White, S. D. M., & Frenk, C. S. 1991, ApJ, 379, 52,
L. P., & O’Dell, S. L. 2000, in Society of Photo-Optical doi: 10.1086/170483
Instrumentation Engineers (SPIE) Conference Series, White, S. D. M., & Rees, M. J. 1978, MNRAS, 183, 341,
Vol. 4012, X-Ray Optics, Instruments, and Missions III, doi: 10.1093/mnras/183.3.341
ed. J. E. Truemper & B. Aschenbach, 2–16, Yuan, T. T., Kewley, L. J., & Sanders, D. B. 2010, ApJ,
709, 884, doi: 10.1088/0004-637X/709/2/884
doi: 10.1117/12.391545