Radio Follow-up on all Unassociated Gamma-ray Sources from the Third Fermi Large Area Telescope Source Catalog

TL;DR

This study conducted radio follow-up observations of unassociated gamma-ray sources from the 3FGL catalog, identifying new associations with active galactic nuclei and refining source positions, thereby enhancing understanding of gamma-ray origins.

Contribution

It presents a comprehensive radio survey and VLBI follow-up that significantly increased the number of gamma-ray source associations and improved positional accuracy.

Findings

142 new AGN associations with gamma-ray sources

144 improved source positions to milliarcsecond accuracy

36 extended radio sources as candidate gamma-ray counterparts

Abstract

The third Fermi Large Area Telescope (LAT) $\gamma$-ray source catalog (3FGL) contains over 1000 objects for which there is no known counterpart at other wavelengths. The physical origin of the $\gamma$-ray emission of those objects is unknown. Such objects are commonly referred to as unassociated and mostly do not exhibit significant $\gamma$-ray flux variability. We performed a survey of all unassociated $\gamma$-ray sources found in 3FGL using the Australia Telescope Compact Array and Very Large Array in the range of 4.0-10.0 GHz. We found 2097 radio candidates for association with $\gamma$-ray sources. The follow-up with very long baseline interferometry for a subset of those candidates yielded 142 new AGN associations with $\gamma$-ray sources, provided alternative associations for 7 objects, and improved positions for another 144 known associations to the milliarcsecond level of accuracy. In addition, for 245 unassociated $\gamma$-ray sources we did not find a single compact radio source above 2 mJy within 3$\sigma$ of their $\gamma$-ray localization. A significant fraction of these empty fields, 39%, are located away from the galactic plane. We also found 36 extended radio sources that are candidates for association with a corresponding $\gamma$-ray object, 19 of which are most likely supernova remnants or HII regions, whereas 17 could be radio galaxies.

Tables

Table 1. Table 1: List of observations with connected interferometers.

T	C	Code	Start	Dur.	Tune	B/w	#	Identifier
			(UTC)	(h)	(GHz)	(GHz)
ATCA	H214	C2624	2012 Sep 19 10:00	29	5.5/9.0	2.0	411	A1
VLA	A	S5272	2012 Oct 26 11:19	2	5.0/7.3	1.0	41	V1A
VLA	A	S5272	2012 Nov 03 21:01	7	5.0/7.3	1.0	175	V1B
ATCA	H214	C2624	2013 Sep 25 21:30	45	5.5/9.0	2.0	997	A2
ATCA	H168	C2624	2014 Apr 07 16:00	30	5.5/9.0	2.0	405	A3
ATCA	H214	C2624	2014 Sep 23 04:00	66	5.5/9.0	2.0	703	A4
VLA	B	S7104	2015 Mar 16 21:33	1	5.0/7.0	2.0	30	V2A
VLA	B	S7104	2015 Mar 16 22:33	3	5.0/7.0	2.0	107	V2B
VLA	B	S7104	2015 Mar 17 09:46	1	5.0/7.0	2.0	21	V2C
VLA	B	S7104	2015 Mar 21 12:37	4	5.0/7.0	2.0	156	V2D
ATCA	H214	C2624	2015 Apr 04 04:00	8	5.5/9.0	2.0	139	A5
VLA	B	15A-466	2015 Apr 16 20:00	1	5.0/7.0	2.0	19	V2E

Table 2. Table 3: The first 8 rows of the list of 291 associations of 3FGL objects that have a likelihood ratio of their association of more than 8 based on their parsec-scale radio emission detected with follow-up VLBI observations. The full table is available in the electronic attachment.

3FGL name	VLBI Name	Right Ascension	Declination	${\sigma }_{\alpha }$	${\sigma }_{\delta }$	Corr	D	N$\sigma$	S	$\Lambda$	3FGL ass
		h   min   s	∘   ′   ′′	mas	mas		′		mJy
3FGL J0000.2$-$3738	RFC J0000$-$3738	00 00 08.414182	-37 38 20.67354	0.54	1.10	0.178	1.36	0.86	22.0	207.2
3FGL J0003.2$-$5246	RFC J0003$-$5247	00 03 19.600260	-52 47 27.28128	19.44	10.13	-0.272	1.04	0.70	15.0	194.3	RBS 0006
3FGL J0006.2$+$0135	RFC J0006$+$0136	00 06 26.924724	+01 36 10.38555	1.07	2.16	-0.413	2.64	1.48	14.0	33.0
3FGL J0007.4$+$1742	RFC J0007$+$1745	00 07 18.873920	+17 45 34.56806	0.16	0.26	-0.201	3.72	0.84	22.0	28.2
3FGL J0007.9$+$4006	RFC J0007$+$4008	00 07 41.666525	+40 08 29.93735	0.42	0.66	0.258	3.43	1.19	14.0	20.1
3FGL J0008.3$+$1456	RFC J0008$+$1456	00 08 25.399845	+14 56 35.79096	0.18	0.32	-0.049	1.19	0.40	21.0	104.5
3FGL J0008.6$-$2340	RFC J0008$-$233A	00 08 35.399660	-23 39 28.00793	0.32	0.80	0.133	1.70	1.09	29.0	202.1	RBS 0016
3FGL J0010.5$-$1425	RFC J0010$-$1420	00 10 41.940201	-14 20 20.33807	0.26	0.56	-0.149	5.49	1.24	21.0	8.4

Table 3. Table 4: Objects with alternative associations based on parsec-scale emission.

3FGL name	type	old assoc	suggested assoc.
3FGL J2021.9$+$0630	agn	87GB 201926.8$+$061922	RFC J2021$+$0629
3FGL J0332.0$+$6308	agn	GB6 J0331$+$6307	RFC J0331$+$6308
3FGL J0150.5$-$5447	agn	PMN J0150$-$5450	RFC J0150$-$5450
3FGL J0415.7$-$4351	agn	SUMSS J041605$-$435516	RFC J0416$-$4350
3FGL J1838.5$-$6006	agn	SUMSS J183806$-$600033	RFC J1838$-$6005
3FGL J0930.9$-$1904	psr	PSR J0931$-$19	RFC J0930$-$1903
3FGL J2017.6$-$1616	psr	PSR J2017$-$1618	RFC J2017$-$1618

Table 4. Table 5: First eight rows of 158 extended sources found in radio observations targeting 3FGL unassociated γ 𝛾 \gamma -ray objects. The full table is available in the electronic attachment.

3FGL Name	RA	Dec	D	Nd	C
	hms	dms	′
J0000.1$+$6545	23 59 41.68	$+$65 42 3.456	4.14	3.3	2
J0003.2$-$5246	00 03 11.60	$-$52 41 37.20	5.04	5.5	2
J0003.4$+$3100	00 03 20.13	$+$30 55 25.01	5.24	2.1	1
J0020.9$+$0323	00 20 50.30	$+$03 23 43.10	1.29	1.6	2
J0026.2$-$4812	00 25 37.47	$-$48 16 39.11	7.27	4.5	2
J0154.1$+$4642	01 53 56.09	$+$46 38 51.47	3.89	1.7	1
J0225.8$+$6159	02 25 37.82	$+$62 05 53.76	6.92	5.8	1
J0026.2$-$4812	00 25 53.27	$-$48 16 43.19	5.34	3.3	2

Table 5. Table 6: The first 8 rows of the list of 245 3FGL unassociated objects that have no radio source brighter 2 mJy within the 3- σ 𝜎 \sigma Fermi localization error ellipse. In the 98 cases where a TGSS ADR1 was found within the 3- σ 𝜎 \sigma Fermi localization error ellipse, the corresponding identifier, separation, and σ 𝜎 \sigma are listed.

3FGL	TGSS ADR1	D	Nd
		(′)
3FGL J0001.6+3535	J000051.8+352741	12.0	150.6
3FGL J0003.5+5721	J000332.3+572711	05.6	167.9
3FGL J0004.2+6757	-	-
3FGL J0017.1+1445	-	-
3FGL J0022.7+4651	J002212.5+464402	09.0	122.5
3FGL J0032.5+3912	-	-
3FGL J0039.3+6256	-	-
3FGL J0051.6+6445	-	-

Equations

$$\Lambda =\frac{e^{-n^2/2}}{N\left(1\right)S^{{}^p}{d}^2/4},$$

Full text

galaxies: active; catalogs; surveys; gamma rays: general; radio continuum: general

††slugcomment: final.1.0††facilities: VLA,ATCA,VLBA,LBA,Fermi/LAT

1 Introduction

The latest generation of $\gamma$-ray observatories, in particular the Large Area Telescope (LAT) aboard the Fermi satellite have revolutionized the field of $\gamma$-ray astronomy. The previous major satellite mission, the Energetic Gamma-Ray Experiment Telescope (EGRET) on board the Compton-Gamma Ray Observatory (CGRO) only detected a couple hundred point sources (Hartman et al., 1999; Casandjian & Grenier, 2008). Since its launch in August 2008, Fermi/LAT has detected thousands of distinct $\gamma$-ray emitting objects which have been reported in multiple catalogs. For all of these catalogs a significant fraction, typically $>30\%$, of the detected point sources have no known counterpart at any other wavelength and their nature is unknown. This makes the $\gamma$-ray sky the least understood in all of observational astronomy.

There are two ways to associate objects from different catalogs: 1) to match light curves if a source manifests variability; 2) to associate by proximity. Due to errors in the measurement of flux or localization, both methods are statistical. Association by proximity is based on two probabilities: the probability that entries in two catalogs are related to the same object with the differences in positions due to uncertainties in localization, and the probability that the second catalog contains a coincident background object, unrelated to the object in the first catalog. The second probability is crucial for effective association. If the catalog selected for association contains too many objects such that the probability to find a background object within the localization error ellipse is not small, then the effectiveness of association with such a catalog will be too low to be practical.

The method for source associations used in the 3FGL is addressed in section 5 of (3FGL; Acero et al., 2015). A detailed description for the method can be found in Ackermann et al. (2012). We found in our previous work (Petrov et al., 2013; Schinzel et al., 2015) that association based on proximity of Fermi sources to radio sources with emission at 8 GHz at parsec scales is very effective. In this paper we will present the results from our method that is complimentary to the approach taken by the Fermi science team. Following the statistical criteria described in Petrov et al. (2013), the detection of a compact radio source brighter than 12 mJy at 8 GHz found within the 2-$\sigma$ Fermi localization error ellipse establishes an association. More than 50% of all Fermi associations are made on the basis of this approach. There are two reasons why this association method based on VLBI detection is so powerful. Firstly, the number of compact radio source is rather limited: according to the $\log N$–$\log S$ distribution, there are only $\!100,000$ sources brighter than 10 mJy at parsec scales at 8 GHz (Petrov et al., 2013). Secondly, variability of both Fermi $\gamma$-ray sources and compact radio sources suggests that emission is contemporaneous and comes from regions on parsec scales. In contrast, catalogs from connected interferometers, such as NVSS (Condon et al., 1998) and SUMSS (Bock et al., 1999; Mauch et al., 2003), typically probe emission at kiloparsec scales that is related to the interaction of particles in the jet with surrounding media thousands of years ago.

Our approach is first to observe with connected interferometers fields within the Fermi localization error ellipse at 5–9 GHz and then follow-up detected sources with VLBI. At these frequencies emission from a compact core usually dominates in contrast to frequencies below 2 GHz where emission from extended jets or radio lobes generally dominates. Therefore detection of a source at 5–9 GHz at arcsecond scales increases the probability of finding detectable emission at milliarcsecond scales at 8 GHz. This contrasts to the case when only emission at 0.8–1.4 GHz at arcsecond scale is known, which generally does not result in detection of compact emission. Observations with connected interferometers at 5–9 GHz serve as a screening tool for potential targets for follow-up VLBI observations to find new associations.

In 2012, we started a campaign to image all regions covered by the Fermi localization error ellipse containing unassociated Fermi sources first with connected interferometers at 4–10 GHz, then with Very Long Baseline Interferometry (VLBI). The results covering the second Fermi point source catalog (2FGL; Abdo et al., 2013) were reported in Petrov et al. (2013) and Schinzel et al. (2015). The 2FGL presented 1873 sources of which 575 were considered unassociated. Follow-up on those unassociated sources revealed 865 radio sources at arcsec scales as candidates for association. We then obtained new associations using VLBI for 76 of the unassociated $\gamma$-ray sources with radio-loud active galactic nuclei (AGN). We found that 129 out of 588 observed $\gamma$-ray sources at arcmin scales did not have a single radio continuum source detected above our sensitivity limit within the 3$\sigma$ $\gamma$-ray localization. These “empty” fields were found to be particularly concentrated at low Galactic latitudes.

Since then, a third source catalog was released by the Fermi/LAT collaboration (3FGL; Acero et al., 2015) covering the first 4 years of operations and listing 3033 sources among which 1010 were reported to have no plausible counterpart at other wavelengths. Most of the identified or associated $\gamma$-ray sources are active galaxies. Here we present an update of our work that includes radio observations for new 3FGL unassociated sources and presents them in the context of our previous work.

The 2FGL catalog covers the first two years of the Fermi/LAT mission and lists all point sources found over that time period with about 5$\sigma$ sensitivity. At the time of publication it listed 575 unassociated point sources. However, since its release the catalog was modified and the version released on May 18, 2015 contains 651 unassociated $\gamma$-ray sources with a median $\gamma$-ray flux of $1.21\times 10^{-12}$ ph cm*-2* s*-1* and $\gamma$-ray spectral index of 2.31. In 3FGL, covering the first four years of the Fermi/LAT mission, the number of listed unassociated sources as of May 18, 2015 grew to 1,010 with a median $\gamma$-ray flux of $6.63\times 10^{-13}$ ph cm*-2* s*-1* and $\gamma$-ray spectral index of 2.37. Cross-referencing the catalogs, only 300 unassociated $\gamma$-ray sources from 2FGL are represented in 3FGL, which leaves 710 new unassociated $\gamma$-ray sources that have a median $\gamma$-ray flux of $8.96\times 10^{-13}$ ph cm*-2* s*-1* and index of 2.31.

In Section 2 new radio observations are described that have been performed between 2014 and 2015. This is followed by Section 3 where we describe the observational results and provide the complete catalog of radio counterpart candidates for all 3FGL unassociated sources. This is followed by a discussion of the subclass of steady $\gamma$-ray emitters in Section 4, which are half of all the detected $\gamma$-ray point sources. In Section 4 we discuss the results in the context of understanding the population of unassociated $\gamma$-ray sources. Finally in Section 4 we provide a summary and conclusions for our reported findings.

2 Observations

The 3FGL catalog covers the entire sky, thus we performed follow-up observations at two radio interferometric arrays: The Australia Telescope Compact Array (ATCA) in the southern hemisphere for observing sources with declinations in a range of $[-90^{\circ},+10^{\circ}]$ and the Jansky Very Large Array (VLA) in the northern hemisphere for observing sources with declinations $[0^{\circ},+90^{\circ}]$. The overlap in sky area was used to cross-check VLA and ATCA calibration procedures. Where a source was detected in both ATCA and VLA data, the VLA results were used.

Combining all observing campaigns with VLA and ATCA, we observed 960 fields around all Fermi unassociated sources. We excluded 80 sources for which we had previously found an association from our analysis of radio sources, which fall within the Fermi localization error ellipse of 3FGL sources and exhibit parsec scale emission.

2.1 Australia Telescope Compact Array

We have identified 713 objects marked as unassociated in the 3FGL catalog with declinations $<+10^{\circ}$. Together with our previous observations we were able to associate 80 objects with AGN found to have radio emission on parsec scales using the method described in Petrov et al. (2013) and Schinzel et al. (2015). This is based on the computation of the likelihood ratio between the probabilities to find an unrelated background source within the Fermi localization error ellipse and the probability to find a radio counterpart within a given position difference due to random localization errors. The remaining 633 objects were selected as the primary targets for the observations described here. We added 169 secondary targets to the list. These included 122 sources associated on the basis of X-ray emission, 14 sources associated with the low frequency catalogs NVSS (Condon et al., 1998) and SUMSS (Mauch et al., 2003), and 30 unassociated sources that were in the preliminary version of the 3FGL catalog but not included in the final release. We also included three sources that were marked as unassociated in the preliminary version of the 3FGL but were later associated with pulsars. In total, 805 fields with declinations $<+10^{\circ}$ were in the target list.

Observations were made in three campaigns: A3 started on 2014 April 07 and lasted for 30 hours, A4 that started on 2014 September 23 and lasted for 66 hours, and A5 that started on 2015 April 04 and lasted for 8 hours. Campaign A3 was observed in array configuration H168 with baselines ranging from 61–192 m between the inner five antennas and $\sim$4.4 km between CA06 and the inner antennas, and campaigns A4 and A5 were observed in H214 configuration with baselines ranging from 92–247 m between the inner five antennas111http://www.narrabri.atnf.csiro.au/operations/array_configurations/configurations.html. Observations in all three campaigns were recorded simultaneously in two bands centered at 5.5 and 9.0 GHz with a bandwidth of 2 GHz in both linear polarizations, with circular polarizations derived during processing. The primary flux calibrator was PKS 1934-638. A summary of all observations is listed in Table 1.

A list of 405 target sources were observed in the A3 campaign. In the A4 campaign we observed the remaining 228 primary targets, 172 secondary targets and we re-observed 303 fields observed in the A3 campaign. There are two reasons for re-observations: 1) we detected a radio source within the $3\sigma$ Fermi localization error ellipse, but at a distance of more than $4^{\prime}$ from the pointing direction – the distance where the primary beam power drops below 20% with respect to the pointing direction at 9 GHz (205 fields); and 2) no source was detected (98 fields). In the latter case we re-observed the field in a 7-element mosaic mode to search for radio counterparts farther from the Fermi position, since typically the Fermi localization error ellipse is wider than the ATCA field of view. During the A5 campaign we re-observed 13 primary targets that were missed in A4, 6 secondary targets that were recorded only in 1 scan in A4 and re-observed 120 fields where a source was found at distance more than $4^{\prime}$ from the pointing direction.

Among primary targets 45% of the fields were observed in 3 scans of 24 s each, 39% were observed in 4 scans, and 16% were observed in 5 scans. Among secondary targets, 37% of the sources were observed in 2 scans; other sources were observed in 3 or more scans.

The data analysis procedure we used for the A3, A4, and A5 campaigns is described in detail in Petrov et al. (2013). We obtained radio images of size $1024\times 1024$ with a pixel size of 2.4 arcsec and a synthesized beam with a typical FWHM size $20^{\prime\prime}\times 30^{\prime\prime}$. The majority of sources found look point-like. Two examples of images with extended structure are shown in Figures 1 and 2 222Compare the image with http://cornish.leeds.ac.uk/cgi-bin/public/summary_src.py?name=G012.8050-00.2007, http://atlasgal.mpifr-bonn.mpg.de/cgi-bin/ATLASGAL_SEARCH_RESULTS.cgi?text_field_1=G012.8057-0.1994&catalog_field=GaussClump, though these are the rare exceptions. We determined the flux density of the detected sources, spectral index within a band, and positions. Typical position uncertainty is around $1^{\prime\prime}$ and typical uncertainty in flux density is 0.1–0.4 mJy. The detection limit is 1.0–1.5 mJy for sources in the center of the field-of-view. For sources detected at both 5.5 and 9.0 GHz bands we determined the spectral index across the bands.

Since target sources are observed in 2–3 scans of 18 s length with a 6-element interferometer, the estimates of flux densities of extended objects may have a strong bias and underestimate the peak flux density by up to one order of magnitude. Scrutinizing the data, we found 1–2% of the reported objects to have questionable flux density estimates. The poor sampling of the Fourier plane due to snapshot observations makes it difficult to filter outliers reliably. When in doubt, we tended to keep questionable flux density estimates, considering it better to keep problematic flux density estimates than to eliminate a potential candidate for association. Follow-up VLBI observations that are sensitive to the compact components of the objects will filter out galactic objects that do not have detectable emission on milliarcsecond scales.

We found a radio object within at least one band and within the area of the 3$\sigma$ Fermi localization error ellipse of 497 3FGL unassociated sources out of 633 (79%). For the remaining 21% of the sources no radio counterpart brighter than 2 mJy was found.

2.2 Very Large Array

A list of 322 unassociated 3FGL fields with declinations above 0∘ were selected for observations with NRAO’s Jansky Very Large Array (VLA) in this campaign (V2). Additionally, we observed the location of 2FGL J0423.4+5612, for which no data was recorded in our previous VLA survey. We reanalyzed our previous campaign V1 (Schinzel et al., 2015) and found that 51 out of 169 (30%) of observed fields lie within $4^{\prime}$ of 3FGL sources. Among these 51 sources, 18 are marked as associated in 3FGL, and 5 others were observed in V2. Therefore, combining V1 and V2, we get 491 fields. Of these, 327 are centered at 3FGL sources marked as unassociated, 18 are marked as associated, and 146 are pointed at directions farther than $4^{\prime}$ from a 3FGL source.

The observations were conducted using the C-Band receiver covering a frequency range of 4–8 GHz, recording a total bandwidth of 4 GHz in both left and right handed circular polarizations. The VLA array was in B configuration, thus an integration time of 2 s was chosen. Real-time correlation was performed using VLA’s WIDAR (Wideband Interferometric Digital ARchitecture) correlator. The instantaneous bandwidth was split into two parts, with one half centered at 5.0 GHz (4.0–6.0 GHz) and the other centered at 7.0 GHz (6.0–8.0 GHz). This provides a simultaneous observation of two separate frequency bands. The observing time of 10 hours was split into five segments to be able to observe fields at all LST ranges and to increase the likelihood of scheduling (see Table 1). The first four segments were observed between 2015 March 16 and 21 under time approved through the NASA Fermi Guest Investigator program, an additional hour to complete the program was approved as director’s discretionary time observed on 2015 April 16.

At the beginning of each observing segment either 3C 48 or 3C 286 was observed to provide a bright flux density/bandpass calibrator. In most cases each target source was observed only once with a total integration time of $\sim$30 s each. Nearby phase calibrators were added with typical integration times of 15 s each in order to solve for changes in the complex gains during the target observations. For all segments the VLA was in B array configuration, providing baseline lengths from 0.21 to 11 km, which results in 1 arcsecond resolutions and a field-of-view of up to about 7*′. The largest angular scale of extended radio structures is about 29′′*.

The data calibration and analysis procedure is similar to what is described in Schinzel et al. (2015). Major differences were that the newer Common Astronomy Software Applications (CASA) release 4.5.2 and the 31DEC2015 release of the the Astronomical Image Processing System (AIPS; Greisen, 1990) were used.333for more information on CASA see http://casa.nrao.edu/ and for AIPS see http://www.aips.nrao.edu/index.shtml The data calibration was performed entirely within CASA.

The imaging was performed in CASA using the clean task. This task uses a Clark based clean algorithm (Clark, 1980), applies the $w$-projection needed for widefield imaging using 100 projection planes, and visibility weights are determined using the Briggs weighting scheme (Briggs, 1995), minimizing sidelobes and noise levels. We used “$\mathrm{robust}=0$”, which corresponds to a weighting scheme in between uniform and natural. In addition, we used multifrequency synthesis with two terms in order to compensate for spectral changes over the 2 GHz instantaneous bandwidth. This provides both a Stokes I map combining the two polarizations together and a spectral index map. The deconvolution was run with 5000 iterations, a default loop gain of 0.1, and a flux density threshold at which to stop cleaning of 0.06 mJy corresponding to the approximate thermal noise limit. The primary beam attenuation was corrected using new measurements described in Perley (2016).

All images were then searched using an automatic procedure reading images into AIPS and using the task SAD to identify point sources. Custom python scripts were applied to analyze and inspect the resulting images and point sources found. Sometimes the algorithm applied in SAD identifies image artifacts near strong point-sources, which were manually flagged in order to retain a clean list of point sources that includes position, flux density, and spectral index for each of the two 2 GHz sub-bands which are used in subsequent analysis.

2.3 VLBI follow-up observations

We should stress that estimates of flux densities of emission at kiloparsec scales determined from analysis of ATCA and VLA observations are not sufficient to provide high-confidence associations for Fermi sources. Firstly, emission at arcsec resolutions can have a significant contribution from extended regions of an AGN that often dominate over emission from the AGN’s parsec scale core. Variability time scales of $\gamma$-ray fluxes strongly suggest $\gamma$-ray emission to be generated at parsec scales. Secondly, the number of sources with flux density at arcsecond resolution above some limit is significantly greater than the number of sources with flux density above the same limit at milliarcsecond resolutions. Given the large number of weak radio sources located within a typical Fermi localization error ellipse at arcsecond resolution, this prevents association of sources weaker than 30–50 mJy. Thus, we suggest that ATCA and VLA observations provide only initial candidates for associations.

Here we report results of our follow-up observations of candidate sources with VLBI. We included 144 target sources detected in ATCA and VLA observations in the ongoing VLBA Calibrator Survey Densification campaign (VCS8/9) at 4.4 and 7.6 GHz (Petrov, 2016)444http://astrogeo.org/vcs9. The goal of this program is to increase the density of calibrator sources. New AGN candidates for $\gamma$-ray association turn out to be good targets for this program. We included a number of target sources in the ongoing LBA Calibrator Survey (LCS-2) at 8.4 GHz that has similar goals as the VLBA program but is focused on sources in the southern hemisphere. The data analysis procedures of the observations is similar to previous VCS and LCS campaigns and is described in Petrov et al. (2008) and Petrov et al. (2011). The detection limit of these programs was in a range of 10–12 mJy for sources at elevations above $20^{\circ}$ at the telescopes.

We also ran a dedicated VLBA program at 7.6 GHz in 2015 and in 2016 for observing 561 target sources detected by the ATCA and VLA (codes BS241 and S7104) with flux density at 9 GHz exceeding 10 mJy or at 5.5/7 GHz exceeding 20 mJy. Target sources were observed in two scans of four minutes each using a bandwidth of 480 MHz, and dual polarization. We ran two LBA experiments in 2014 and 2016 at 8.4 GHz for observing 140 target sources at declinations below -$40^{\circ}$. These target sources were observed in two scans of 210 s each within the spanned bandwidth of 320 MHz. The detection limit for dedicated VLBI observations is 6 mJy when a source is at elevations above $20^{\circ}$ at the telescopes. All VLBI observations were correlated with accumulation periods of 0.1 s and a spectral resolution of 62.5–125 kHz in order to have a wide field-of-view with sensitivity reduced by no more than 20% within $1^{\prime}$ of the pointing direction.

Additional VLBI observations to cover the remaining $\gamma$-ray counterparts brighter than 10 mJy in any band are ongoing and will be presented in a future publication.

3 Results

3.1 Catalog of Radio Candidates

Similar to Schinzel et al. (2015), we combined the results from the ATCA and VLA observing campaigns into a single uniform dataset. We combined results of three ATCA campaigns, A3, A4, A5, the VLA campaign V2 that we described in the previous sections and two ATCA and one VLA campaign. This continues our program of observing unassociated sources in 1FGL and 2FGL catalogs described in our previous publications (Schinzel et al., 2015; Petrov et al., 2013), A1, A2, and V1. We retained 2097 sources, 50% of the total number of detected sources, that are within the $3\sigma$ Fermi localization error ellipse and have flux densities above 1 mJy in at least one band. The catalog presented here for the fields observed in the previous A1, A2, and V1 campaigns are not identical to those previously published and is only valid for 3FGL localizations. Firstly, we re-analyzed the observations. Secondly, since positions of $\gamma$-ray sources in 3FGL and their uncertainties are slightly different than positions in the 1FGL and 2FGL catalogs, some radio sources absent in the published versions are now closer to the 3FGL positions than to 1FGL or 2FGL and are now included, while some radio sources with positions beyond the 3FGL 3$\sigma$ Fermi localization error ellipse are now excluded.

If a source was detected at distances from the pointing direction beyond where the total power drops to 20% of the pointing center, which is $6.2^{\prime}$ at 5 GHz and $4.2^{\prime}$ at 7 GHz, we report only a lower limit on the flux density. The detection limit at the center of the field-of-view is 1 mJy and is 5 mJy at the edge. A number of sources are detected only at one band. This may be due to the source spectrum or because a source was too distant from the pointing direction. We report spectral index estimates only if a source was detected at both bands within $4.2^{\prime}$ of the pointing direction.

All radio sources found within the 3-$\sigma$ Fermi localization error ellipse were cross-matched against the reprocessed TIFR GMRT 150 MHz Sky Survey (TGSS) alternative data release 1 (ADR1: Intema et al., 2016), NRAO VLA Sky Survey (NVSS: Condon et al., 1998), Sydney University Molonglo Sky Survey (SUMSS, version 2.1 of 2012 February 16: Bock et al., 1999; Mauch et al., 2003), the Molonglo Galactic Plane Survey 2nd Epoch (MGPS-2: Murphy et al., 2007), the Gaia Data release 1 (Gaia: Gaia Collaboration et al., 2016), and the Wide-field Infrared Survey Explorer (WISE) catalog555http://wise2.ipac.caltech.edu/docs/release/allwise/ (ALLWISE, November 13, 2013: Wright et al., 2010; Mainzer et al., 2011), which combines the data from the WISE cryogenic and post-cryogenic survey phases providing the most comprehensive view of the full mid-infrared sky currently available. We obtained 837 matches with the ALLWISE catalog, 1291 matches with the Gaia DR1 catalogue, 1051 matches with NVSS, 461 matches with SUMSS and MGPS-2, and 507 matches with TGSS ADR1. We evaluated the probability of a false detection for Gaia counterparts in the following way. We counted the number of sources in a uniform $0.25^{\circ}\times 0.25^{\circ}$ grid and converted it to the density of Gaia sources per steradian. We then defined the probability of false detection as the product of the area in the search radius and local Gaia sources density (Petrov & Kovalev, 2016). The search radius for sources associated with VLBI is $0.2^{\prime\prime}$, and $3.0^{\prime\prime}$ otherwise. A search radius of $0.2^{\prime\prime}$ for VLBI associated sources was selected to accommodate possible position errors in Gaia.

Table 6 presents the catalog of detected sources within 3-$\sigma$ of the Fermi localization error ellipse. The table of the remaining 1842 sources detected outside of the 3-$\sigma$ Fermi localization error ellipse can be found in the online attachment.

3.2 New Associations

In order to establish the association of a parsec scale source with its $\gamma$-ray counterpart we run a likelihood ratio test using Poisson statistics. The likelihood ratio of association $\Lambda$ is defined as the probability that the $\gamma$-ray and radio source found at separation $d$ is physically the same object, and their position difference is due to statistical errors only, to the probability that the radio source is a background, unrelated object. Simple geometric consideration results in

[TABLE]

with $d$ being the angular separation in radians between the radio and the $\gamma$-ray position, $S$ is the radio flux density in Jansky, $n$ is the normalized distance between the radio and $\gamma$-ray source777The distance between the radio and $\gamma$-ray localization divided by the 1$\sigma$ Fermi localization error., $N\left(1\right)=374$, and $p=-1.2088$. The numerical values were determined from $\log N$–$\log S$ relation of correlated flux densities. A detailed description for determining the association probabilities and the numerical value of the $\log N$–$\log S$ diagram are given in Petrov et al. (2013) and Schinzel et al. (2015).

In total, we performed VLBA or LBA observations in the fields that contain 798 sources detected in our VLA and ATCA programs within $1^{\prime}$ of the pointing direction. We employed such a wide field-of-view because a compact source with emission at parsec scales is not always coincident with the peak brightness of the radio structure on kiloparsec scales. Petrov (2013) demonstrated a number of such examples.

We have detected 451 compact sources from these VLBI observations. Among 2097 sources detected in our ATCA and VLA programs, there are 744 objects with flux densities greater 10 mJy in either 5.5 or 9 GHz. Of these, 630 or 85% have been followed up with VLBI. The remaining 114 objects, primarily in the southern hemisphere, will be observed in the near future.

We have computed the likelihood ratio among 451 VLBI sources detected within $1^{\prime}$ of objects observed in VLA and ATCA programs. We consider the sources with $\Lambda>8$ and a normalized arc length of $n<3$ as associations. The probability that the normalized arc-length exceeds 3 due to random errors of the $\gamma$-ray source position is 1.1%. That means setting this criteria we will have false negative associations for 1% of the sources. We selected the $\Lambda>8$ criteria to have approximately the same number of false positive associations. In total, we have 19 double VLBI associations for 1530 $\gamma$-ray sources. Excluding two gravitational lenses 3FGL J0221.1+3556 and 3FGL J1833.6-2103, we get a false positive probability of 1.1%. We performed an additional test: we rotated the VLBI catalog at random angles in a range of [0.2, 100] degrees, ran 256 test associations, and found on average of 18.9 associations. Thus, we conclude that the false association probability of this criteria, both positive and negative, is around 1%.

We found that 286 sources out of 451 (63%) have , and therefore are listed as associations based on their parsec-scale emission. Of these, 144 have associations reported in the 3FGL catalog. Of those, 3 Fermi sources have two associations. Association based on parsec-scale emission confirms them and allows us to improve their position accuracy to milliarcsecond scales. We have further established new associations with AGNs for 142 sources. Of these, 2 Fermi sources have two associations.

In the majority of fields observed with ATCA and VLA we found more than one point source that is a candidate for association. VLBI observations allow us to find which of these point sources have parsec-scale detectable emission and therefore identify the likely counterpart. Figure 3 provides an illustration of this. There are 6 point sources in the field. Of them, two show parsec scale emission at $>$20 mJy level. The first, J0905-2120 has a $\Lambda$ of 2.5, which is below the threshold of 8 for a reliable association, and the second J0905-2131 which is a double at $28^{\prime\prime}$ has a $\Lambda$ of 22.9 and thus, we considered it to be associated with the 3FGL object J0905.8-2127. Figure 4 illustrates another situation. There are two sources detected with ATCA $4.7^{\prime}$ apart within the 3FGL J1323.2-3901 localization error ellipse: J1323-3859 and J1323-3903. The first source has a flux density of 6.3 mJy at 5.5 GHz and a spectral index of +0.5, while the second one has flux density of 29.0 mJy and a spectral index of $-$0.8. One may expect that the strongest source is a probable association. However, VLBI observations revealed compact emission from the weaker source J1323-3859 and no compact emission from J1323-3903888It is interesting that the VLBI flux density of J1323-3903 at 8 GHz, 23 mJy at epoch 2015.02.24 and 31 mJy at epoch 2015.11.10, is significantly higher than its ATCA flux density 8 mJy at 9 GHz at epoch 2014.04.07. The flux density of J1323-3903 found in the same field extrapolated to 1.4 GHz is $89\pm 6$ which is rather close to the value from NVSS of 70 mJy. The flux density of J1323-3859 extrapolated to 1.4 GHz would be $3.1\pm 0.4$ mJy and it was not detected in the NVSS with a limit of 2.4 mJy. These comparisons with NVSS demonstrate that there is no gross oversight in the ATCA data calibration. Therefore, we attribute the discrepancy between ATCA and VLBI flux density estimates to source variability.

Finally, we present Table 3 which provides the list of our proposed 291 associations based on parsec-scale emission at 8 GHz detected with follow-up VLBI of point sources found in our ATCA and VLA observations. We note that association of J1413-6520 in this list is problematic. The ATCA observations provide the position of the Circinus galaxy, while with VLBI we detected only emission from the supernova remnant 1996cr, $20.86^{\prime\prime}$ away.

3.3 Previously reported 3FGL associations

Among 170 sources with associations reported by the 3FGL team we overturned 5 associations for radio sources that are AGN and provide alternative associations for two pulsars (see Table 4). For all 5 AGN we suggest new associations that are closer to the 3FGL position than the associations suggested by the Fermi team. Laffon et al. (2015) report $\gamma$-ray pulsation with the same period as millisecond pulsar PSR J0931-19. The pulsar PSR J0931-19 and the AGN RFC J0930-1903 are two different objects. One of the explanations why two methods, detection of parsec scale emission and detection of radio pulses, led to different sources is that errors of 3FGL J0930.9-1904 Fermi localization may be larger than their formal uncertainties.

We have detected radio counterparts within $1^{\prime}$ of all 122 observed X-ray sources associated with Fermi sources. The median position differences between ATCA and X-ray positions is $6.8^{\prime\prime}$ with 7 objects detected at distances in a range of $30$–$51^{\prime\prime}$. ATCA observations provided improvement in positions for 85% of the sources. We observed 84 out of 122 sources with the VLBA and detected 82. Among the remaining 38 objects, 28 have a flux density from ATCA observations above 10 mJy. They will be observed with VLBI in the future.

Based on these observations, we can conclude that association of 3FGL sources based on the X-ray emission matches very well with association based on radio emission from parsec scales. We strongly confirmed two out of three X-ray associations, and found no X-ray sources from our sample of 122 objects that have not been detected at 5.5 or 9 GHz.

3.4 Extended Sources

All target fields observed with ATCA and VLA were imaged and visually inspected. We found 158 fields where sources showed emission larger than the synthesized beam. Two classes of such extended sources were identified: Large Angular Structures (LAS) that were either resolved (or over-resolved) and double sources or point sources with extensions. Table 5 lists the 3FGL source in those fields and their corresponding localization within that field, distance to the 3FGL position, and structural classification.

Most sources found are either sources exhibiting a shell like structure, as shown in Figures 1, 6, and 8, while a second group of sources shows a morphology that resembles double lobed sources or one-sided jets, as in Figures 5 and 7. In 36 cases an extended source was found within the 2$\sigma$ Fermi localization error ellipse, of which 19 were classified as LAS and 17 as double source. Fig. 9 shows the sky distribution of the extended sources. From this plot it is evident that LAS sources are primarily located within the Galactic plane, whereas double sources are primarily found outside of the Galactic plane. Judging from the morphologies and radio spectral properties the objects shown in Figures 6 and 8 we see that they resemble supernova remnant shells or HII regions.

• •

In the case of 3FGL J2004.4+3338 the shell is found within the 1-$\sigma$ localization of the $\gamma$-ray source and is most likely associated with such. An infrared counterpart is found for this structure in the ALLWISE catalog (Wright et al., 2010; Mainzer et al., 2011), J200423.63+333904.2.

• •

In the case of 3FGL J0225.8+6159 the shells are found more than 5-$\sigma$ away from the $\gamma$-ray localization. This object also has an infrared (IR) counterpart in the ALLWISE catalog, J022537.37+620553.3. This indicates the presence of a HII region in both cases.

• •

The case of Figure 5 shows an extended steep spectrum radio source that could be a compact symmetric object (Wilkinson et al., 1994; Readhead et al., 1996), but more likely is a radio galaxy, due to the lack of evidence for the presence of a single compact core. Higher resolution observations are required to understand the nature of this object. It is found within the 2-$\sigma$ localization of 3FGL J0154.1+4642.

• •

Figure 7 shows an example of two extended sources separated by $\sim$8*′′*. Both of these have a steep spectral index and if related resemble the structure of a radio galaxy. In this case this structure is located within 1-$\sigma$ of the $\gamma$-ray localization.

Overall, the above defined source classes seem to separate a Galactic and an extragalactic population of extended radio sources.

3.5 Empty Fields

The list of 245 3FGL sources with no radio counterpart brighter than 2 mJy at 5 GHz within the 3-$\sigma$ Fermi localization error ellipse is presented in Table 6. Of these, 237 are marked as unassociated in the 3FGL catalog, four are associated with pulsars, one is associated with NVSS J224604+154437. The latter association was not confirmed. The remaining three are associated with supernova remnants, namely 3C 391, MSH 17-39 and 3CTB 37A. Among those 245 “empty” fields we found evidence for extended radio emission in 39 cases listed in Table 5. In those cases there is evidence for extended emission with a total flux density of $>$2 mJy, however no individual bright spots exceed $>$2 mJy. This leaves 206 empty fields in which we did not detect a radio source above 2 mJy with the VLA and ATCA. We call these fields “empty” in the sense there are no radio sources brighter than 2 mJy within the observed frequency range. As Frail et al. (2016) noted, some of these fields contain weak objects listed in low-frequency NVSS and SUMSS catalogs. The locations of the empty fields on the sky are illustrated in Fig. 9, which shows no apparent concentration of empty fields in the Galactic plane. However, counting the number of empty fields, 27% (65/245) lie within 1∘ of the Galactic plane. There were 54 matches of TGSS ADR1 sources (Intema et al., 2016) that were found within 3$\sigma$ of the $\gamma$-ray localization. Such sources are most likely steep spectrum radio sources and are good candidates for Pulsar searches.

Notable are 95/245 (39%) empty fields that are found $\pm$10∘ outside of the galactic plane. Their median $\gamma$-ray flux density is $5.96\times 10^{-13}$ph cm*-2* s*-1* MeV*-1* and median $\gamma$-ray spectral index is 2.4. Their median variability index is 47.0 (Ackermann et al., 2011), which indicates that most of the sources can be assumed to be non-variable. The spectral index corresponds to the median of the total population of $\gamma$-ray point sources.

A

4 Discussion & Summary

With the newer Fermi/LAT catalogs providing longer averaging time one can see that from 2FGL to 3FGL the median fluxes of unassociated $\gamma$-ray sources have dropped by a factor of 1.8. Since the two catalogs 2FGL and 3FGL were derived using different diffuse background models and slightly different methods, we use in the following analysis only values from 3FGL and when referring to 2FGL sources the values of the 300 cross-listed objects are chosen. Comparing the group of previously found unassociated $\gamma$-ray sources with the group of newly found ones and using the 3FGL reported variability indices we find that 3.0% of previously reported objects are variable compared to 2.4% of newly found objects. This shows that the majority of unassociated $\gamma$-ray sources do not show significant $\gamma$-ray variability and could be considered steady $\gamma$-ray emitters. Since most of the unassociated sources are steady emitters this cannot be used to distinguish different populations such as fields associated with AGN or empty fields. An additional complication arises from the fact that finding variability in fainter $\gamma$-ray sources is very difficult given the poor photon statistics.

Despite pushing toward lower $\gamma$-ray flux densities in the newest $\gamma$-ray source catalogs we continue to increase the number of AGN associations using radio observations. Eventually, this will lead to a reduction of the radio flux density completeness limit of $\gamma$-ray radio counterparts by a factor of 10, from 100 mJy to around 10 mJy. With the results presented here, together with previous observations, we have firmly associated 144 new $\gamma$-ray objects with newly discovered AGN and improved positions of 170 previously associated sources to the milliarcsecond level of accuracy. We expect to increase this number after all VLBI follow-up observations have been completed. This marks the single largest reduction in unassociated sources by any follow-up program to date and demonstrates the strength of this effort in comparison to solely relying on the analysis of existing multi-wavelength catalogs. In the global context this corresponds to a reduction of 3FGL unassociated sources by 14%, leaving 901 sources without a multi-wavelength counterpart. Note that the total number of known $\gamma$-ray pulsars is currently 220999https://confluence.slac.stanford.edu/display/GLAMCOG/Public+List+of+LAT-Detected+Gamma-Ray+Pulsars, against 1550 $\gamma$-ray AGN with radio emission from parsec scales. A recent publication discussing potential new $\gamma$-ray pulsar candidates highlights the challenges involved in finding more $\gamma$-ray pulsars (Frail et al., 2016).

The most remarkable result presented here are the so-called empty fields, which are not necessarily empty, but for which no radio point source brighter than several mJy was found in the region of the $\gamma$-ray localization. We first pointed these fields out in Schinzel et al. (2015). For 2FGL unassociated sources those empty fields were primarily localized within the galactic plane. However, 39% of empty fields were found outside of the galactic plane. This hints at an extragalactic distribution in addition to the previously found galactic one. There is still a possibility that a large number of empty fields are related to pulsars, which are also found outside of the galactic plane. However, another possibility could be the presence of radio quiet AGN that are not found in optical/IR catalogs. It is expected that high-frequency peaked AGN have very faint radio flux densities, due to the shifting of their overall spectral energy distribution to higher energies.

We found in at least 16% (154) of observed radio fields extended radio sources, of which 36 lie within 2$\sigma$ of the $\gamma$-ray source. These extended sources were broken down into two sub-classes, one indicating a large angular structure object and the second indicating the presence of a double or jet-like extended source. The first subclass could be related to galactic SNR or HII regions, which should be further investigated with more compact interferometer configurations or radio observations at lower frequencies, as well as a systematic comparison to infrared observations which could distinguish the SNR origin from HII regions. The majority of sources found in the second subclass are most likely radio galaxies, but could also contain compact symmetric objects (CSOs). High resolution VLBI spectral imaging observations are needed to determine the nature of those objects, which has been done in the case of 2234$+$282 (An et al., 2016). However no convincing CSO $\gamma$-ray emitter has been found to date at redshifts beyond z$>$0.006. Establishing a firm association of a CSO or distant radio galaxy with a $\gamma$-ray source would challenge existing interpretations of the physics of the production of $\gamma$-ray emission.

In summary, we present an update of our work on new associations for $\gamma$-ray unassociated sources found by Fermi. For all 3FGL unassociated sources we performed radio observations with compact interferometers, which provided 2097 radio objects as candidates for association. We have observed a large fraction of those providing milliarcsecond scale detections. For 144 of those detections we provide firm associations of newly found AGN based on statistical likelihood. More remarkably, we found a new population of $\gamma$-ray fields devoid of any compact radio source brighter than 2 mJy that lies outside of the galactic plane. We also provide a list of fields for which we found large scale radio structures that are indicative of SNRs, HII regions, and radio galaxy counterparts.

High resolution radio follow-up observations are underway to cover all newly reported association candidates with which we expect to double the current number of new AGN associations.

We thank the anonymous referee for a thorough review which improved the quality of this publication. FKS, GBT, and LP acknowledge support by the NASA Fermi Guest Investigator program, grants NNX12A075G, NNX14AQ87G and NNX15AU85G. The National Radio Astronomy Observatory is a facility of the National Science Foundation operated under cooperative agreement by Associated Universities, Inc. The Australia Telescope Compact Array & Long Baseline Array are part of the Australia Telescope National Facility which is funded by the Commonwealth of Australia for operation as a National Facility managed by CSIRO. The authors also thank Jamie Stevens and Elaine Sadler with their support of the ATCA observations. FKS dedicates this paper to his wife Mónica for her support and understanding leading to this publication. This publication makes use of data products from the Wide-field Infrared Survey Explorer, which is a joint project of the University of California, Los Angeles, and the Jet Propulsion Laboratory/California Institute of Technology, and NEOWISE, which is a project of the Jet Propulsion Laboratory/California Institute of Technology. WISE and NEOWISE are funded by the National Aeronautics and Space Administration. This work made use of the Swinburne University of Technology software correlator, developed as part of the Australian Major National Research Facilities Programme and operated under licence. 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, and the TOPCAT software version 4.1101010http://www.star.bris.ac.uk/m̃bt/topcat/ (Taylor, 2005). The authors made use of the database CATS (Verkhodanov et al., 2007) of the Special Astrophysical Observatory.