A Combined Radio Multi-Survey Catalog of Fermi Unassociated Sources

Abstract

Approximately one-third of existing γ -ray sources identified by the Fermi Gamma-Ray Space Telescope are considered to be unassociated, with no known counterpart at other frequencies/wavelengths. These sources have been the subject of intense scrutiny and observational effort during the observatory’s mission lifetime, and here we present a method of leveraging existing radio catalogs to examine these sources without the need for specific dedicated observations, which can be costly and complex. Via the inclusion of many sensitive low-frequency catalogs we specifically target steep-spectrum sources such as pulsars. This work has found steep-spectrum radio sources contained inside 591 Fermi unassociated fields, with at least 21 of them being notable for having pulsar-like γ -ray properties as well. We also identify a number of other fields of interest based on various radio and γ -ray selections.

Full text

A Combined Radio Multi-Survey Catalog of Fermi Unassociated Sources
S. Bruzewski
1
, F. K. Schinzel
2,3
, and G. B. Taylor
1
1
Department of Physics and Astronomy, University of New Mexico, Albuquerque, NM 87131, USA; bruzewskis@unm.edu
2
National Radio Astronomy Observatory, P.O. Box O, Socorro, NM 87801, USA
Received 2022 October 6; revised 2022 December 6; accepted 2022 December 6; published 2023 January 24
Abstract
Approximately one-third of existing γ-ray sources identified by the Fermi Gamma-Ray Space Telescope are
considered to be unassociated, with no known counterpart at other frequencies/wavelengths. These sources have
been the subject of intense scrutiny and observational effort during the observatory’s mission lifetime, and here we
present a method of leveraging existing radio catalogs to examine these sources without the need for specific
dedicated observations, which can be costly and complex. Via the inclusion of many sensitive low-frequency
catalogs we specifically target steep-spectrum sources such as pulsars. This work has found steep-spectrum radio
sources contained inside 591 Fermi unassociated fields, with at least 21 of them being notable for having pulsar-
like γ-ray properties as well. We also identify a number of other fields of interest based on various radio and γ-ray
selections.
Unified Astronomy Thesaurus concepts: High energy astrophysics (739);Surveys (1671);Radio source catalogs
(1356);Spectral index (1553);Radio astronomy (1338);Active galaxies (17);Gamma-ray astronomy (628);Radio
continuum emission (1340);Pulsars (1306)
Supporting material: data behind figure, machine-readable tables
1. Introduction
During its mission lifetime, the Fermi Gamma-Ray Space
Telescope has provided substantial insights into the high-
energy regime of the sky. Of particular note are the resolved γ-
ray sources described in each subsequent data release from the
Large Area Telescope (Atwood et al. 2009). The most recent
release (4FGL-DR3; see Abdollahi et al. 2022)features some
6659 sources, many falling into categories one might expect,
such as pulsars or active galactic nuclei (AGNs). Beyond these
sources though, the catalog has presented a persistent mystery
in the form of its unassociated sources: γ-ray sources that are
detected by Fermi, but cannot be definitively associated with or
identified as a source in any other electromagnetic regime.
Currently the number of such sources stands at 2157, making
up 32% of the entire catalog. As γ-ray properties alone are
generally insufficient to determine the physical nature of a
source, this means that nearly one-third of all presently known
astrophysical γ-ray sources have an unknown origin. One
should also note that the existence of these sources cannot be
completely explained away by invoking signal-to-noise argu-
ments. While present unassociated sources do generally have
somewhat lower fluxes, there are a noninsignificant number of
bright unassociated sources that have persisted across catalog
updates over the years. As an example, the source 4FGL
J1801.6-2326 was originally published in the very first Fermi-
LAT catalog (0FGL), and has remained unassociated even
though it is a “bright”source by any criteria that one might
imagine. This source is, at most, loosely associated with the
supernova remnant W28, although there are other γ-ray sources
that are more definitively linked to that object (Abdo et al.
2010).
The existence of such persistent sources implies that some
fraction of the unassociated sources will continue to remain
inscrutable against our traditional methods of association and
identification, and thus necessitates the use of new methods that
might be able to provide insight into these unassociated fields.
The radio regime has long served as a primary testing ground
for such methods, as most sources of γ-rays (e.g., AGN and
pulsars)would typically be expected to appear in the radio
(with the exception of a few exotic systems), and the smaller
sky density of sources compared to optical or near-optical
makes the problem somewhat more tractable. The techniques
used in radio observations thus far can largely be divided into
two categories: association via statistical likelihood arguments
based on source properties (common for AGN), and identifica-
tion via correlated variability (common for pulsars). Both of
these methods have found remarkable success, with a majority
of sources in both 4LAC and 2PC (the Fermi catalogs of AGN
and pulsars; see Ajello et al. 2020 and Abdo et al. 2013
respectively)having been originally identified/associated via
radio observations.
For previous searches (such as those described in Schinzel
et al. 2015 and Petrov et al. 2013), each new list of
unassociated sources was targeted using the Jansky Very
Large Array (VLA)at 5 and 7 GHz, looking for bright,
compact sources inside the positional uncertainty ellipse. These
candidate sources could then be targeted using very long
baseline interferometry (VLBI), typically via the Very Long
Baseline Array (VLBA). If these follow-up observations
detected the source, one could then apply a statistical argument:
what is the likelihood this source is the γ-ray emitter, as
opposed to a chance background source, knowing that we have
near completeness on said compact radio sources at these
frequencies thanks to the calibrator surveys such as Petrov
(2021)? If this quantified likelihood surpasses some threshold,
The Astrophysical Journal, 943:51 (9pp), 2023 January 20 https://doi.org/10.3847/1538-4357/acaa33
© 2023. The Author(s). Published by the American Astronomical Society.
3
An Adjunct Professor at the University of New Mexico.
Original content from this work may be used under the terms
of the Creative Commons Attribution 4.0 licence. Any further
distribution of this work must maintain attribution to the author(s)and the title
of the work, journal citation and DOI.
1

the source may be considered a candidate for association, and
passed along to other groups for further analysis.
A new approach began to be implemented with the release of
the 8 yr catalog (4FGL; Abdollahi et al. 2020). By cataloging
sources in the early data products of the recently completed first
epoch of the Very Large Array Sky Survey (VLASS; Lacy
et al. 2020)we were able to link any new 5 and 7 GHz data we
acquired to a new data point at 3 GHz, effectively doubling our
spectral coverage, and allowing us access to certain Fermi
unassociated sources that we previously would not have
targeted due to observational constraints. A full description
of these efforts can be found in Bruzewski et al. (2021), but to
summarize, sources at each frequency are linked using an
uncertainty weighted distance metric, such that a graph is
generated for each unassociated Fermi source. We can then
extract all connected component subgraphs, considering them
multifrequency sources, and fit for their spectral features.
This work represents the logical extension of that effort to
include a larger number of catalogs, thus spanning both an
extended range of frequency and a larger portion of the entire
sky. In particular, this approach incorporates an increased
number of low-frequency radio catalogs, which is ideal if one is
searching for sources with steeply negative spectral indices
(positive convention, S
ν
∝ν
+α
). We highlight these sources
because it is well known that continuum emission from pulsars
typically occurs in the range α<−1.4 (Bates et al. 2013), quite
outside the standard range of spectral indices for extragalactic
objects Condon et al. (1971).
This continuum based analysis, inspired by similar but more
limited efforts such as Frail et al. (2018)and Massaro et al.
(2014), has so far served as the basis of many different
observational techniques toward Fermi unassociated sources,
and has provided a significant number of newly identified γ-ray
pulsars and newly associated blazars (see Giroletti et al. 2016).
Furthermore, it should be noted that compared to pulsation
searches, this method of pulsar identification is much less
sensitive to the effects of pulse scattering by the interstellar
medium. A population of scattered γ-ray pulsars is of particular
interest given current questions about the nature of the Galactic
Center Excess (Hooper & Goodenough 2011). Initially this
anomalous excess of γ-rays was ascribed to the theoretical
annihilation of dark matter particles in the galactic center
region (Berlin et al. 2014), but more recent analysis has
produced some contention. While dark matter remains a
popular explanation for its origin (see Ackermann et al.
2017; Grand & White 2022), it seems this excess could also be
explained by the existence of an unresolved population of γ-ray
pulsars in that region. As such, confirmation of this existence
(or nonexistence)of these objects would be likely to lead to
significant progress in our understanding of the Galactic Center
Excess.
The paper is organized as follows: in Section 2we describe
the process of assembling the various catalogs into a combined
system. In Section 3we provide analysis of the connected
component sources that we have identified. Finally in Section 4
we discuss broader implications of this work.
2. Methodology
The novel approach this paper describes involves the
introduction of several new catalogs to an existing source-
finding framework established in Bruzewski et al. (2021).In
particular, we sought out any sky survey that covered a
substantial portion of the sky, and provided some new
information in frequency or sky coverage space. The end goal
was to produce reasonable sky coverage at as many frequencies
as possible.
To this end we identified 14 catalogs that could be added to
our existing dedicated observations. A list of these catalogs,
along with their general properties, can be seen in Table 1.In
total we have collected approximately 7 million point sources
spanning the frequency range from 74 MHz to 20 GHz. While
many of these surveys covered the entire sky available to the
instruments involved, some were more specifically targeted,
giving us increased coverage in some areas, particularly the
galactic plane. It should also be noted that because of the
comparative number of instruments in the Southern Hemi-
sphere, our coverage in that region of the sky is somewhat less
than the northern sky. Figure 1shows a summary map of our
sky coverage.
At this stage we perform a similar friends-of-friends
graph analysis to identify entries in catalogs that are likely
associated with the same physical source. To do this we narrow
Table 1
Selected Catalogs
Catalog Center Frequency Number of Sources Sky Coverage Reference
AT20G 5, 8, and 20 GHz 3797, 3795, and 5877 Southern sky Massardi et al. (2008)
CGPS 1.4 GHz 72787 Northern galactic Taylor et al. (2003)
Dedicated 5 and 7 GHz 12999 and 10247 Selected northern sources Bruzewski et al. (2021)
FIRST 1.40 GHz 946432 Northern extragalactic Becker et al. (1995)
GLEAM 200 MHz 329487 Southern extragalactic Hurley-Walker et al. (2017)
LOTSS 144 MHz 300098 Selected northern field Shimwell et al. (2022)
NVSS 1.40 GHz 1773484 Northern sky Kimball & Ivezic (2006)
PMN 4.85 GHz 50814 Southern sky Wright et al. (1994)
SUMSS 843 MHz 211047 Southern extragalactic Mauch et al. (2003)
TGSS 1.50 GHz 623604 Northern sky Intema et al. (2017)
VLASS 3.00 GHz 2232725 Northern sky Lacy et al. (2020)
VLITE 364 MHz 39957 Selected northern sources Clarke et al. (2016)
VLSSR 74.0 MHz 92965 Northern sky Lane et al. (2014)
WENSS 325 MHz 229418 Northern sky Rengelink et al. (1997)
WISH 352 MHz 90357 Selected southern field De Breuck et al. (2002)
Note. A brief summary of the various catalogs included in our analysis. For a full description of each catalog, one should see the various provided references.
4
Radio Fundamental Catalog: http://astrogeo.org/rfc/.
2
The Astrophysical Journal, 943:51 (9pp), 2023 January 20 Bruzewski, Schinzel, & Taylor

each catalog to only entries inside of the Fermi positional
uncertainty ellipse, checking its normalized radius

ˆ()
qq
=+
⎜⎟⎜⎟
⎛
⎝⎞
⎠⎛
⎝⎞
⎠
rd
ab
cos sin 1, 1
S
H
F
H
F
22
where d
S
is the distance from the center of the ellipse to the
source, θ
H
is the angle between the ellipse major axis and the
source, and a
F
and b
F
are the semimajor and semiminor axes of
the ellipse, respectively. It is necessary to select for interior
sources at this stage because the graphing analysis scales
roughly as ()

nsources
2, making a whole-sky analysis of 7
million sources computationally prohibitive. We instead select
for nearby sources (a relativity inexpensive calculation)and
then perform the analysis on at most a few dozen sources inside
the ellipse.
Once we have our list of interior sources we can begin to
network them. For catalogs generated using the PyBDSF
Python package (here VLASS and our dedicated observations)
we are already provided intracatalog connections, which can be
drawn immediately. This is especially powerful given that these
particular catalogs represent the deepest observations of many
of these fields, giving us extra insight into sources with more
complex morphology. From here, we look for intercatalog
connections using a likelihood ratio metric, where sources are
connected if their uncertainty normalized distance
¯⪅()
a
s
dd
s
=D+D
ad
⎜⎟⎜⎟
⎛
⎝⎞
⎠⎛
⎝⎞
⎠
dcos 3.71, 2
R
22
where Δαand Δδrepresent the difference in R.A. and decl.
respectively, and the uncertainties here are the quadrature sums
s
ss=+
ii i
2,1
2,2
2. Note that the fractional uncertainty on R.A.
is scaled by the cosine of the mean decl. to remove coordinate
effects. This criterion originally comes from de Ruiter
et al. (1977), and is derived by assuming the sources are
scattered as a Rayleigh distribution. The choice of cutoff at
»2ln10 3.71
3, inspired by a similar cutoff applied in the
creation of the LOTSS DR1 catalog (Shimwell et al. 2017),is
set such that we would miss only 1 in 1000 sources randomly
scattered in such a distribution, while also not making the
search radius so large that we have a large chance of
encountering a separate source.
Treating each source inside the Fermi ellipse as a node in a
graph and each connection as an edge between two nodes, we
can then extract connected component subgraphs using the
networkx python package. An example field, featuring two
simple sources and one more complicated networked source, is
shown in Figure 2. Every one of these subgraphs represents a
multifrequency source that we can extract the spectrum of. The
general function we fit to our data is of the form
() ()
()
n=a+
SSe ,3
xcx
000
2
where ()nn=xln 0. In this case we are fitting for the
reference flux S
0
, the spectral index α
0
, and the spectral
curvature c
0
. We can choose the reference frequency ν
0
, which
we perform the fit at, and so we adopt the logarithmic midpoint
of the source data such that ·
n
nn=
0 min max , as this seems to
minimize the final uncertainty. Of course not every source will
contain enough unique frequency points to be fit by this
function, so we apply the following procedure:
1. Two frequencies—calculate values and uncertainties
directly from the data points.
2. Three frequencies—fit a version of Equation (3)where
c
0
=0.
3. Four frequencies or more—fit the full version of Equation (3)
to the data.
These fits are performed using the curve_fit method in
Scipy, which also provides the parameter uncertainties. These
fit parameters and their uncertainties are then recorded along
with the catalog entry names from their respective catalogs.
Figure 3shows an example of a well-constrained fit of a source
with fairly obvious spectral curvature, which appears in a
variety of catalogs.
Where multiple sources from the same catalog are present
(e.g., sources linked by PyBDSF in VLASS), we calculate the
sum of their fluxes, and provide the name of the source with the
lowest positional uncertainty. We note that this should be a
sufficient solution for point-like or minimally resolved sources,
which are the primary targets of this study. Finally we generate
a weighted average of the sky coordinates, and from that
produce a unique IAU-style name for each source.
These data, along with various metadata on the ellipse
containing each source, are then combined into our final data
product, which we call the Multi-Survey Catalog (MSC). This
catalog is provided alongside this article, and can also be found
(along with diagnostic plots such as that shown in Figure 2)at
www.cv.nrao.edu/F357/MSC/. Table 2describes the structure
of the catalog. The intent of this catalog is to provide a high-
level overview of radio sources in a given Fermi field, which
can then be used to select for targets of interest based on
properties such as flux density, spectral index, or curvature in a
given band. Transformation of the given values to a new
reference frequency is straightforward, with flux being scaled
as in Equation (3), and spectral index scaling like
() ()an a=+cx2. 4
00
Note that because spectral curvature is the highest-order
spectral term we fit for, the curvature term will be the same
Figure 1. An approximate map of our sky coverage in equatorial coordinates.
For each 5°×5°bin, we count catalogs having at least one source in that
region. Note that only a small portion of the southern sky is not covered by our
catalogs.
5
https://www.astron.nl/citt/pybdsf/
6
https://docs.scipy.org/doc/scipy/reference/generated/scipy.optimize.
curve_fit.html
3
The Astrophysical Journal, 943:51 (9pp), 2023 January 20 Bruzewski, Schinzel, & Taylor

regardless of the choice of frequency. If further details are
needed, our catalog then identifies the relevant sources in the
various radio catalogs which may be of interest. This catalog
may then be used as the basis for further analysis of these
unassociated fields, some of which we detail in the section to
follow.
3. Analysis
As a test of our methods, we performed the above processing
on all sources in the 4FGL-DR3 catalog, including those which
have an existing association or identification. In total we
identify 37041 distinct radio sources inside of the positional
error ellipses of 5746 Fermi fields (see Section 4for a
discussion why this number is lower than the 6659 total objects
in the DR3 catalog). Looking at just the unassociated fields we
find 12746 radio sources inside 1622 (out of 2157)fields. Of
these 3598 are identified at more than one frequency, allowing
to fit for spectral index, and 1329 have enough data points to fit
for spectral curvature.
In the sections to follow, we attempt to illustrate some of the
ways in which this catalog may be used to select for interesting
targets. We provide some degree of analysis and discussion on
each method, and where possible we also provide the relevant
subset catalogs alongside the primary catalog. We by no means
intend for this list of methods to be definitive, and would
encourage others to find novel ways to extrapolate targets from
the information provided.
3.1. Pulsar Candidates
The most obvious way we can begin to find interesting
sources from our data is to look at the radio properties of the
identified sources. In particular the most obvious choice is the
use of the spectral index, which in this study is particularly
powerful given the extended range of our spectral coverage
compared to our prior efforts, especially at lower frequencies.
From studies such as Condon et al. (1971)we know that the
bulk of extragalactic radio sources (such as AGNs)have
spectral indices of α≈−0.5. Furthermore we know that
pulsars most often have spectral indices well outside the norm,
typically in the range α<−1, with the distribution peaked at
α=−1.4 (Bates et al. 2013).
Beginning with the pulsars, it is fairly straightforward to
select for targets with the correct spectral index. As our fits
include curvature, one must calculate the spectral index at a
particular choice of frequency; here we choose 1.4 GHz, the
frequency of the VLA Lband, as that is a likely band to use in
the case of follow-up for pulsar candidates. We then select for
sources with an L-band spectral index in the range
−3<α<−1, as that roughly matches the distribution shown
in Bates et al. (2013)and similar works, while also excluding
the majority of flatter spectrum sources and any outliers that
have artificially steeper indices. This selection alone produces a
total of 1213 radio sources.
These sources are contained in 591 unassociated Fermi
fields, implying approximately two pulsar candidates per field,
Figure 2. An unassociated field with a few well-defined sources, one of which can easily be identified in multiple catalogs. On the left we show the coordinates of each
radio source, along with the properties of the Fermi ellipse. On the right is a more abstract view of the source graph used to illustrate the catalog composition of each
source. We provide these plots for every Fermi source whose positional uncertainty ellipse contains at least one radio source, along with the small guide table that one
can use to find the sources in the larger Multi-Survey Catalog or their respective original radio catalogs. Note that for Source 1, there are in fact a number of sources
shown in the left-hand image, they are simply plotted over the top of each other.
Figure 3. A sample spectrum that covers nearly our entire range of frequencies,
in good agreement with the fit. The Fermi source containing this radio source is
unassociated.
4
The Astrophysical Journal, 943:51 (9pp), 2023 January 20 Bruzewski, Schinzel, & Taylor

but in actuality the difference between these numbers is largely
driven by a small number of fields containing a larger number
(in one case as many as 40)candidates. We note that such fields
do not appear to be correlated to the size of the Fermi ellipse,
but do seem to most often appear in or near the galactic plane,
which could easily cause an increase in noise and the
generation of spurious sources. A simple solution is then to
only target sources in fields having less than some number of
candidates, assuming fields with more than this number cannot
be trusted. As an example, we find 907 steep-spectrum sources
in 566 unassociated fields each containing at most five of these
radio candidates.
One would typically then take the further step of selecting
for sources that would be bright enough to be easily observable
at the chosen frequency. Interestingly, for the choice of
1.4 GHz, all radio sources in unassociated fields within this
range of spectral index have spectral flux densities about 1 mJy,
our typical threshold for such a cut. One such radio source can
be seen in Figure 3, showing a wide range of spectral data and
easily bright enough to be considered a good choice for follow-
up. We highlight these as sources that may have been passed
over by traditional timing surveys due to highly scattered pulse
profiles.
A further step we can take is to make use of γ-ray properties
that have been known to be indicative of pulsars. One primary
example comes from the 4FGL-DR3 paper, where it is noted
that pulsars seem to have significant spectral curvature in the γ-
ray band, and can be disentangled from blazars by their
location in the significance-curvature space (see Figure 15 of
Abdollahi et al. 2022). With this in mind we select for
unassociated sources having an average significance (column
Signif_Avg in the catalog)greater than 10, and falling above
the line
() ()>LP_SigCurv 0.6 Signif_Avg . 5
0.7
This line was selected by eye to divide the associated and
identified populations of blazars and pulsars. The cutoff in
significance keeps our selection out of the region of the
parameter space where the distinction is more ambiguous. This
particular selection is of course hardly novel on its own, but
gains more leverage when collated with the radio selected
pulsar-like sources. Here we enforce the further criteria that we
are only interested in sources that are moderately well defined
spectrally, such that their fractional spectral index error is less
than 50%. With these γ-ray and radio criteria in hand, we
identify 21 sources that we designate as pulsar candidates of
high interest, making ideal targets for follow-up. These 21
sources are noted in the MSC and Table 3with the “psr-
candidate”note, as well as illustrated in Figure 4.
3.2. Blazar Candidates
We also note that the above methodology can be extended to
flat-spectrum sources if one is searching instead for AGNs. By
selecting for bright and flat-spectrum radio sources, primarily
outside the galactic plane, one might be able to identify a
suitable list of potential AGN candidates worthy of follow-up
via VLBI or a similar method. As flat-spectrum sources make
up the bulk of radio sources, one would likely require further
criteria to generate a reasonably small list of targets.
For example, one could easily make us of a machine
learning–generated catalog such as those in Chiaro et al. (2016)
or Saz Parkinson et al. (2016). Here we compare our data
against the 134 blazar candidates identified by Kaur et al.
(2019), which used γ-ray and likely X-ray properties as inputs
for both a decision tree and random forest machine-learning
classification system. We identify 26 of these blazar-like
candidate fields as containing a flat-spectrum radio source, all
of which would thus be prime candidates for immediate follow-
up, and notably we are able to do this using only existing
catalogs, without the need to further survey these fields. All
26 are illustrated in Figure 4. Twenty of these sources are noted
in the MSC and Table 3with the “blz-candidate”note.
We note the remaining six sources separately, as they also
appear in Kaur et al. (2022). This more recent work performed
further multiregime classification in an attempt to identify
blazar candidates as one of two primary subtypes, either flat-
spectrum radio quasars or BL Lac–type objects. All six of the
flat-spectrum radio sources we found are located inside fields
Table 2
MSC Column Description
Name Format Unit Description
Name 23 s Generated positional name
FermiName 17 s Fermi field which contains this
source
FermiClass 05 s Fermi object classification
RA deg .4f Weighted average of R.A. (J2000)
RAErr deg .2e Weighted error on average RA
Dec deg .5f Weighted average of decl. (J2000)
DecErr deg .2e Weighted error on average Dec
Name_AT20G_05 20 s
Name_AT20G_08 20 s
Name_AT20G_20 20 s
Name_CGPS 19 s
Name_D5GHZ 05 s
Name_D7GHZ 05 s
Name_FIRST 16 s
Name_GLEAM 14 s
Name_LOTSS 22 s
Name_NVSS 14 s
Name_PMN 15 s
Name_SUMSS 20 s
Name_TGSS 24 s
Name_VLASS 23 s
Name_VLITE 07 s
Name_VLSSR 22 s
Name_WENSS 16 s
Name_WISH 18 s
RefFreq Hz .4e Reference frequency at the log-mid-
point of data
RefFlux mJy .3f Fit flux value at RefFreq
RefFluxErr mJy .3e Fit uncertainty on RefFlux
Alpha .3f Fit spectral index, positive
convention
AlphaErr .3e Fit uncertainty on Alpha
Curve .3f Fit spectral curvature
CurveErr .3e Fit uncertainty on Curve
Notes 13 s Notes on sources
Note. Here the format column refers specifically to the Python string formatting
code used to generate the final output table.
(This table is available in its entirety in machine-readable form.)
5
The Astrophysical Journal, 943:51 (9pp), 2023 January 20 Bruzewski, Schinzel, & Taylor

classified as BL Lac candidates, and as such are denoted as
“bll-candidates”in the MSC and in Table 3.
3.3. Superempty Fields
The concept of empty fields in this context was first
discussed in Schinzel et al. (2017), which noted that a number
of Fermi fields contained no significant (S
ν
>2 mJy)radio
sources between 4 and 10 GHz. These fields represent an
interesting subset of unassociated sources where instead of any
sort of ambiguity in potential associations, we simply have no
likely sources which could be producing the γ-rays.
For the MSC it would be more difficult to establish a specific
criterion for significance in terms of spectral flux density, as our
Table 3
Targets of Interest
Name FermiName Spectral Index LP SigCurv Signif Avg Candidate Type
MSC J015905.20+331257.6 4FGL J0159.0+3313 −0.55 0.36 8.84 bll
MSC J040921.74+254442.0 4FGL J0409.2+2542 −0.86 1.21 7.39 bll
MSC J080056.56+073235.0 4FGL J0800.9+0733 −0.66 3.56 9.12 bll
MSC J083855.81+401736.1 4FGL J0838.5+4013 −0.84 0.43 4.13 bll
MSC J091429.68+684508.4 4FGL J0914.5+6845 −0.68 0.37 8.56 bll
MSC J155734.66+382030.1 4FGL J1557.2+3822 −0.51 1.50 4.00 bll
MSC J003655.87-265215.2 4FGL J0037.2-2653 −0.71 0.41 4.20 blz
MSC J013724.50-324039.3 4FGL J0137.3-3239 −0.16 1.16 5.18 blz
MSC J040607.94+063922.0 4FGL J0406.2+0639 −0.93 0.12 4.22 blz
MSC J072310.80-304754.0 4FGL J0723.1-3048 −0.44 0.34 8.24 blz
MSC J073725.92+653637.8 4FGL J0737.4+6535 −0.83 1.23 6.66 blz
MSC J075615.95-051254.5 4FGL J0755.9-0515 −0.67 1.76 8.26 blz
MSC J090605.54-100907.6 4FGL J0906.1-1011 −0.93 0.58 6.84 blz
MSC J093420.67+723045.1 4FGL J0934.5+7223 −0.96 0.76 7.43 blz
MSC J104705.92+673758.0 4FGL J1047.2+6740 −0.39 1.36 7.85 blz
MSC J104938.78+274213.0 4FGL J1049.8+2741 −0.67 0.05 6.51 blz
MSC J111147.61+013907.2 4FGL J1111.4+0137 −0.70 0.00 4.44 blz
MSC J111444.14+122618.1 4FGL J1114.6+1225 −0.61 3.08 4.48 blz
MSC J112213.70-022913.8 4FGL J1122.0-0231 −0.20 1.93 7.75 blz
MSC J122438.56+701649.7 4FGL J1224.6+7011 −0.72 0.97 8.28 blz
MSC J125636.31+532549.4 4FGL J1256.8+5329 −0.51 0.79 5.64 blz
MSC J162331.34-231334.3 4FGL J1623.7-2315 −0.78 1.08 4.68 blz
MSC J164825.28+483656.4 4FGL J1648.7+4834 −0.98 1.65 5.71 blz
MSC J181805.97+253548.0 4FGL J1818.5+2533 −0.60 2.80 6.64 blz
MSC J185559.92-122329.2 4FGL J1856.1-1222 +0.75 1.37 9.30 blz
MSC J232718.97-413437.5 4FGL J2326.9-4130 −0.68 0.67 6.60 blz
MSC J020509.38+665340.5 4FGL J0204.7+6656 −1.78 4.67 10.97 psr
MSC J053314.10+594509.3 4FGL J0533.6+5945 −1.56 8.50 16.77 psr
MSC J075149.66-293021.7 4FGL J0752.0-2931 −1.17 6.55 11.18 psr
MSC J075158.77-293602.9 4FGL J0752.0-2931 −1.22 6.55 11.18 psr
MSC J075451.90-395313.3 4FGL J0754.9-3953 −1.43 6.80 12.37 psr
MSC J120325.67-175001.9 4FGL J1203.5-1748 −1.05 3.31 10.71 psr
MSC J120342.65-174918.7 4FGL J1203.5-1748 −1.24 3.31 10.71 psr
MSC J135638.90+023722.6 4FGL J1356.6+0234 −1.41 4.73 10.52 psr
MSC J140715.01-301456.6 4FGL J1407.7-3017 −1.51 5.35 10.17 psr
MSC J140725.95-301445.7 4FGL J1407.7-3017 −1.54 5.35 10.17 psr
MSC J152953.52-151835.0 4FGL J1530.0-1522 −1.34 5.18 10.01 psr
MSC J171109.30-300536.7 4FGL J1711.0-3002 −1.39 6.72 13.21 psr
MSC J173526.34-071803.9 4FGL J1735.3-0717 −1.33 4.14 13.51 psr
MSC J173928.23-253112.0 4FGL J1739.3-2531 −1.23 4.16 11.95 psr
MSC J174652.52-350528.7 4FGL J1747.0-3505 −1.41 3.88 10.14 psr
MSC J180550.22+340116.8 4FGL J1805.7+3401 −1.81 4.38 14.30 psr
MSC J181341.68+282008.7 4FGL J1813.5+2819 −1.20 4.60 11.88 psr
MSC J190835.87+081523.4 4FGL J1908.7+0812 −1.57 7.56 13.74 psr
MSC J194019.07-251552.0 4FGL J1940.2-2511 −1.13 3.90 13.59 psr
MSC J202631.53+143054.0 4FGL J2026.3+1431 −1.30 4.52 11.35 psr
MSC J210744.58+515736.6 4FGL J2108.0+5155 −2.20 4.81 14.29 psr
MSC J210803.98+515253.8 4FGL J2108.0+5155 −1.30 4.81 14.29 psr
MSC J211437.06+502155.0 4FGL J2114.3+5023 −1.02 3.74 12.60 psr
MSC J211654.13+134155.8 4FGL J2117.0+1344 −1.65 4.31 12.74 psr
MSC J211709.60+134416.4 4FGL J2117.0+1344 −1.22 4.31 12.74 psr
MSC J225020.87+330429.7 4FGL J2250.5+3305 −1.20 7.51 17.10 psr
(This table is available in machine-readable form.)
6
The Astrophysical Journal, 943:51 (9pp), 2023 January 20 Bruzewski, Schinzel, & Taylor

frequency range is much wider and we are particularly
interested in sources with steeper spectra. Instead we choose
to highlight what we dub superempty fields: unassociated
Fermi fields that contain no radio sources whatsoever. We
identify 537 such fields, a catalog of which is provided among
our data products (MSC_empty.fits; see Figure 4).
There are several explanations for these sorts of sources, all
of which make them somewhat tantalizing targets for follow-
up. To begin with, a large fraction likely exist owing to a lack
of coverage in the Southern Hemisphere, as can be illustrated
by examining the large number residing at low decl. and
unreachable by northern observatories. These fields provide
motivation toward surveys of unassociated sources by southern
observatories (e.g., ATCA, MeerKAT, ASKAP), which should
shed new light on a significant number of these fields.
Of course not all of these fields are due to gaps in coverage,
as shown by the numerous fields comfortably reachable in the
northern sky. This implies the existence of a population of γ-
ray sources across the sky, which remain resistant against
traditional search methods. We can explain these sources two
ways: either the sources do not produce radio emission, or the
emission is in some way obscured or complicated.
Toward the first explanation, there has been interest in these
fields as possible sites of dark matter annihilation. It is believed
that WIMP-like dark matter would self-annihilate along
channels, which would produce γ-rays in a range observable
to Fermi and/or HESS (see Abdallah et al. 2018; Coronado-
Blazquez et al. 2019, respectively). As the annihilation signal is
confined to high energies, one would not expect any sort of
complimentary signal at lower energies, especially in the radio.
As these sources would be by their nature effectively invisible
to radio telescopes, the main insight we can provide is
identifying and associating radio/γ-ray sources toward com-
pleteness of the catalog, leaving only dark matter candidates to
study.
If, however, these fields simply have had their radio
emission missed by searches thus far, then there are a few
potential origins. For fields in the galactic plane, it is possible
that these could represent pulsars that have had their pulsed
emission scattered by intervening medium, effectively wiping
out the pulsed signal that would typically identify them in
pulsation searches. These scattered pulsars would be obser-
vable by other methods, such as their steep spectrum, and so
can be targets by dedicated low-frequency observations, or by
archival methods such as those illustrated in this paper. It is
also known that a majority of pulsars exhibit high circular
polarization in their continuum emission (Kazbegi et al. 1991),
and thus this could be used as further evidence toward a pulsar
target. Once such targets are found, dedicated timing observa-
tions can be performed and used to definitively identify the γ-
ray source.
We note that all of the above methods have found some
degree of success in follow-up observations. One particular
example to highlight is the recent discovery of PSR J0002
+6216 (Schinzel et al. 2019), a cannonball pulsar having been
found in one of the empty fields listed in Schinzel et al. (2017),
which has not beforehand been identified by pulsation searches.
This system is set to provide significant insight into the
evolution of pulsar wind nebulae (P. Kumar et al. 2022, in
preparation), as well as the pulsar natal-kick velocity distribu-
tion (S. Bruzewski et al. 2023, in preparation).
Pulsars can also provide an explanation for some fields
outside the galactic plane. It is generally noted that γ-ray
pulsars typically fall into one of two populations: either
recycled MSPs, or young pulsars (Abdo et al. 2013). For the
first type, the binary in which these have been recycled will
have had time to migrate out of the plane, and can be easily
missed by blind pulsation searches, as binaries significantly
complicate the process of searching for periodic signals. As
such we compare our superempty fields and MSC sources
Figure 4. A galactic sky map of sources of interest discussed in Section 3. We have overlaid these sources on the Fermi 4FGL-DR3 sensitivity map for the sake of
illustration. The dashed white line represents the minimum observable decl. of the VLA, meaning sources inside the circle will generally have been out of reach of
northern observatories. The plot features 21 pulsar candidates, 26 blazar candidates, and 97 empty PSR-like fields. The data providing the locations of the empty MSC
fields are available.
(The data used to create this figure are available.)
7
The Astrophysical Journal, 943:51 (9pp), 2023 January 20 Bruzewski, Schinzel, & Taylor

against Gaia DR3 binaries identified in Gomel et al. (2022)as
having a compact component. While we do not find any cross
matches to MSC radio sources, we do note 22 superempty
fields containing at least one of these binaries, and highlight
them in the catalog.
Finally, these superempty fields, especially those outside the
galactic plane, may harbor high-redshift radio galaxies
(HzRGs). While the exact magnitude of this effect is still in
question (see Meyer et al. 2019; Connor et al. 2021; Hodges-
Kluck et al. 2021, for such discussion), it is thought that at
higher redshifts inverse Compton scattering of accelerated
particles off photons from the cosmic microwave background
(CMB)may become the dominant cooling mechanism over the
synchrotron emission typically seen in more local populations
of AGNs. This would generally lead to a dimming of radio
emission seen from these objects at increasing redshifts, as well
as an enhancement in the X-rays (where the CMB photons are
up-scattered to). This effect has been noted as a possible
explanation for the apparent lack of radio-loud AGNs at high
redshifts (Volonteri et al. 2011). With this is mind, a γ-ray
source outside of the galactic plane could be explained by such
an HzRG, where the γ-rays have reached us unimpeded, but the
radio emission one would typically expect has been quenched.
The degree of this quenching is again a matter of active
discussion, but is not expected to wholly remove the radio
luminosity in its entirety. Thus the best way to probe such
fields toward these objects would be deep observations, which
could potentially pick out the weaker-than-expected radio flux.
It is of some note that the expected continuum spectrum for
these objects likely falls into a similar steep-spectrum range as
that expected for pulsars, and as such there are ongoing efforts
to determine ways to adequately disentangle the two. As an
example, the γ-ray properties of these populations might be
expected to separate similarly to what was described previously
for blazars/pulsars, and so one could conceivably look for
superempty fields with pulsar-like γ-ray properties, meeting the
criteria established in Section 3.1. We identify 97 of these
pulsar-like superempty fields, which we then illustrate in
Figure 4.
4. Discussion and Conclusions
The objective of this work has been to highlight the utility of
existing data toward the study of unassociated γ-ray sources. In
this process we have generalized an approach for networking
catalogs into multifrequency sources, providing immediate and
extended insight into the characteristics of the sources inside
these fields. This method catalogs nearly 13,000 unique radio
sources among the unassociated fields, a large number of which
are detected in multiple catalogs. This spectral information has
been used to generate various lists of interesting targets, and it
is our intent that our MSC serves as a stepping-off point for
others to generate their own targets of interest. In the prior
sections we also particularly highlight the utility of our
enhanced low-frequency coverage toward picking out steep-
spectrum objects, which is of particular note if one seeks to find
pulsar associations for a number of the unassociated fields (or
in some cases HzRGs, as discussed above).
There is also interesting insight to be gleaned from the
networking itself. Figure 5shows the interconnectivity of the
various catalogs that were used in our processing, showing
which catalogs appear most frequently (which effectively
probes the relative scale and depth of said surveys)as well as
which catalogs appear most frequently together (probing
overlaps in coverage, ideally covering different frequencies
so as to provide unique information). The dominance of
northern-sky catalogs, as well as their high degree of
connectivity, illustrates one of the larger gaps in our under-
standing of these unassociated sources, namely the lack of all-
sky surveys in the Southern Hemisphere. We also note the
relative lack of complete high-frequency all-sky catalogs, with
AT20G being the singular exception. For a number of the
sources described in the sections prior, high-frequency data
points or even a confirmed nondetection at those frequencies
can provide significant leverage over the spectral index of a
source. Future surveys with coverage in the Southern Hemi-
sphere, such as C-BASS (Jones et al. 2018)or RACS (Hale
et al. 2021), should lend significant insight into these fields.
RACS especially could serve as a comparable lever-arm in the
Southern Hemisphere, perhaps analogous to VLASS in the
Northern Hemisphere, and as such further analysis incorporat-
ing this survey is planned.
Our analysis was confined to the positional uncertainty
ellipses of Fermi unassociated sources, which would rarely
contain more than 100 sources across the various catalogs.
Generation of the network graph is effectively an (
)
n2
operation, as it calculates the distances between all pairs of
sources. One could conceivably extend this methodology to
larger regions (or even the entirety)of the sky, at the cost of
greatly increased computational complexity, or via the use of
more advanced pair-finding algorithms (such as some sort of
uncertainty aware k-d tree). As astronomy moves toward larger
data products and more complete catalogs of the radio sky,
such further refinements and extensions to this methodology
Figure 5. Connection network showing which catalogs feature most
prominently and how often certain catalogs appear together. Mostly a selection
of the deepest catalogs, biased toward all-sky coverage in the Northern
Hemisphere. The color of each catalog node represents the number of
occurrences of a source from that catalog in the MSC, while the lines between
the nodes are shaded to approximate the number of of times a given pair of
catalogs appear together.
8
The Astrophysical Journal, 943:51 (9pp), 2023 January 20 Bruzewski, Schinzel, & Taylor

may prove invaluable, both to the subject of unassociated fields
and beyond.
We thank Dale Frail for useful discussions in the context of
this paper, as well as Emil Polisensky for his contribution of
early VLITE data toward this analysis. S.B., F.K.S., and G.B.
T., acknowledge support by the NASA Fermi Guest Investi-
gator program, grants 80NSSC19K1508, NNH17ZDA001N,
NNX15AU85G, NNX14AQ87G, and NNX12A075G. The
National Radio Astronomy Observatory is a facility of the
National Science Foundation operated under cooperative
agreement by Associated Universities, Inc. Support for this
work was provided by the NSF through the Grote Reber
Fellowship Program administered by Associated Universities,
Inc./National Radio Astronomy Observatory.
This work has made use of data from the European Space
Agency (ESA)mission Gaia (https://www.cosmos.esa.int/
gaia), processed by the Gaia Data Processing and Analysis
Consortium (DPAC, https://www.cosmos.esa.int/web/gaia/
dpac/consortium). Funding for the DPAC has been provided
by national institutions, in particular the institutions participat-
ing in the Gaia Multilateral Agreement.
This research has made use of NASA’s Astrophysics Data
System and has made use of the NASA/IPAC Extragalactic
Database (NED), which is operated by the Jet Propulsion
Laboratory, California Institute of Technology, under contract
with the National Aeronautics and Space Administration. This
research has made use of data, software and/or web tools
obtained from NASA’s High Energy Astrophysics Science
Archive Research Center (HEASARC), a service of Goddard
Space Flight Center and the Smithsonian Astrophysical
Observatory, of the SIMBAD database, operated at CDS,
Strasbourg, France.
Software: Astropy (Astropy Collaboration et al. 2022;
http://www.astropy.org), matplotlib (Hunter 2007;http://
www.matplotlib.org), networkx (Hagberg et al. 2008;http://
networkx.org), Numpy (Harris et al. 2020;http://www.
numpy.org), Scipy (Virtanen et al. 2020;http://www.
scipy.org).
ORCID iDs
S. Bruzewski https://orcid.org/0000-0001-7887-1912
F. K. Schinzel https://orcid.org/0000-0001-6672-128X
G. B. Taylor https://orcid.org/0000-0001-6495-7731
References
Abdallah, H., Abramowski, A., Aharonian, F., et al. 2018, PhRvL,120, 201101
Abdo, A. A., Ackermann, M., Ajello, M., et al. 2010, ApJ,718, 348
Abdo, A. A., Ajello, M., Allafort, A., et al. 2013, ApJS,208, 17
Abdollahi, S., Acero, F., Ackermann, M., et al. 2020, ApJS,247, 33
Abdollahi, S., Acero, F., Baldini, L., et al. 2022, ApJS,260, 53
Ackermann, M., Ajello, M., Albert, A., et al. 2017, ApJ,840, 43
Ajello, M., Angioni, R., Axelsson, M., et al. 2020, ApJ,892, 105
Astropy Collaboration, Price-Whelan, A. M., Lim, P. L., et al. 2022, ApJ,
935, 167
Atwood, W. B., Abdo, A. A., Ackermann, M., et al. 2009, ApJ,697, 1071
Bates, S. D., Lorimer, D. R., & Verbiest, J. P. W. 2013, MNRAS,431, 1352
Becker, R. H., White, R. L., & Helfand, D. J. 1995, ApJ,450, 559
Berlin, A., Hooper, D., & McDermott, S. D. 2014, PhRvD,89, 115022
Bruzewski, S., Schinzel, F. K., Taylor, G. B., & Petrov, L. 2021, ApJ,914, 42
Chiaro, G., Salvetti, D., La Mura, G., et al. 2016, MNRAS,462, 3180
Clarke, T., Kassim, N., Polisensky, E., et al. 2016, arXiv:1603.03080
Condon, J. J., Jauncey, D. L., & Niell, A. E. 1971, BAAS, 3, 447
Connor, T., Banados, E., Stern, D., et al. 2021, ApJ,911, 120
Coronado-Blazquez, J., Sanchez-Conde, M. A., Dominguez, A., et al. 2019,
JCAP, 2019, 020
De Breuck, C., Tang, Y., de Bruyn, A. G., Rottgering, H., & van Breugel, W.
2002, A&A,394, 59
de Ruiter, H. R., Willis, A. G., & Arp, H. C. 1977, A&AS, 28, 211
Frail, D. A., Ray, P. S., Mooley, K. P., et al. 2018, MNRAS,475, 942
Giroletti, M., Massaro, F., D?Abrusco, R., et al. 2016, A&A,588, A141
Gomel, R., Mazeh, T., Faigler, S., et al. 2022, arXiv:2206.06032
Grand, R. J. J., & White, S. D. M. 2022, MNRAS,511, L55
Hagberg, A. A., Schult, D. A., & Swart, P. J. 2008, in Proc. of the 7th Python
in Science Conf., ed. G. Varoquaux, T. Vaught, & J. Millman, 11
Hale, C. L., McConnell, D., Thomson, A. J. M., et al. 2021, PASA,38, e058
Harris, C. R., Millman, K. J., van der Walt, S. J., et al. 2020, Natur,585, 357
Hodges-Kluck, E., Gallo, E., Ghisellini, G., et al. 2021, MNRAS,505, 1543
Hooper, D., & Goodenough, L. 2011, PhLB,697, 412
Hunter, J. D. 2007, CSE,9, 90
Hurley-Walker, N., Callingham, J. R., Hancock, P. J., et al. 2017, MNRAS,
464, 1146
Intema, H. T., Jagannathan, P., Mooley, K. P., & Frail, D. A. 2017, A&A,
598, A78
Jones, M. E., Taylor, A. C., Aich, M., et al. 2018, MNRAS,480, 3224
Kaur, A., Falcone, A. D., Stroh, M. D., Kennea, J. A., & Ferrara, E. C. 2019,
ApJ,887, 18
Kaur, A., Kerby, S., & Falcone, A. D. 2022, arXiv:2208.10015
Kazbegi, A. Z., Machabeli, G. Z., & Melikidze, G. I. 1991, MNRAS,253, 377
Kimball, A. E., & Ivezic, Z. 2006, AAS/AAPT Meeting, 209, 97.09
Lacy, M., Baum, S. A., Chandler, C. J., et al. 2020, PASP,132, 035001
Lane, W. M., Cotton, W. D., van Velzen, S., et al. 2014, MNRAS,440, 327
Massardi, M., Ekers, R. D., Murphy, T., et al. 2008, MNRAS,384, 775
Massaro, F., Giroletti, M., D’Abrusco, R., et al. 2014, ApJS,213, 3
Mauch, T., Murphy, T., Buttery, H. J., et al. 2003, MNRAS,342, 1117
Meyer, E. T., Iyer, A. R., Reddy, K., et al. 2019, ApJL,883, L2
Petrov, L. 2021, AJ,161, 14
Petrov, L., Mahony, E. K., Edwards, P. G., et al. 2013, MNRAS,432, 1294
Rengelink, R. B., Tang, Y., de Bruyn, A. G., et al. 1997, A&AS,124, 259
Saz Parkinson, P. M., Xu, H., Yu, P. L. H., et al. 2016, ApJ,820, 8
Schinzel, F. K., Kerr, M., Rau, U., Bhatnagar, S., & Frail, D. A. 2019, ApJL,
876, L17
Schinzel, F. K., Petrov, L., Taylor, G. B., et al. 2015, ApJS,217, 4
Schinzel, F. K., Petrov, L., Taylor, G. B., & Edwards, P. G. 2017, ApJ,
838, 139
Shimwell, T. W., Hardcastle, M. J., Tasse, C., et al. 2022, A&A,659, A1
Shimwell, T. W., Rottgering, H. J. A., Best, P. N., et al. 2017, A&A,
598, A104
Taylor, A. R., Gibson, S. J., Peracaula, M., et al. 2003, AJ,125, 3145
Virtanen, P., Gommers, R., Oliphant, T. E., et al. 2020, NatMe,17, 261
Volonteri, M., Haardt, F., Ghisellini, G., & Della Ceca, R. 2011, MNRAS,
416, 216
Wright, A. E., Griffith, M. R., Burke, B. F., & Ekers, R. D. 1994, ApJS,
91, 111
9
The Astrophysical Journal, 943:51 (9pp), 2023 January 20 Bruzewski, Schinzel, & Taylor