HUNTING FOR TREASURES AMONG THE FERMI UNASSOCIATED SOURCES: A MULTIWAVELENGTH APPROACH

To limit number of potential multiwavelength counterparts to explore, we selected bright unassociated γ-ray sources (average significance in all energy bands given in the 2FGL σ > 10), as these usually have smaller position uncertainty. We selected sources located at high latitudes (|b| > 5°) where the density of potential counterparts is lower. We removed all unassociated sources labeled as "c" in 2FGL (those found in a region with bright and/or possibly incorrectly modeled diffuse emission). Those different selection criteria reduced the sample of unassociated Fermi sources in the 2FGL from 555 to 85 (σ > 10), then 33 (|b| > 5°), and finally 32 (no "c" flag).

Among this sample of bright, high-latitude, unconfused unassociated 2FGL targets, we selected the sources that had been observed in the X-rays by the Swift satellite, whose XRT FOV typically encompasses the Fermi uncertainty ellipse. We excluded sources observed by other recent X-ray missions (i.e., XMM, Suzaku, and Chandra) because these data sets have already been studied extensively. In 2010, Swift began a systematic search¹⁰ for X-ray counterparts of Fermi LAT unassociated sources (see Stroh & Falcone 2013). For this reason, the large majority of those bright, high-latitude unassociated 2FGL targets have X-ray coverage. We selected only observations with at least a 3.5 ks exposure time. Our final list includes 22 γ-ray sources.

The Swift observations were analyzed (see Section 3.1) to identify potential X-ray counterparts within the Fermi R₉₅ error ellipse. Among the 22 γ-ray sources, 16 had at least one X-ray counterpart detected with a significance greater than 3σ. The remaining six γ-ray sources¹¹ had no X-ray counterpart and represent a puzzling sample that would require deeper X-ray observations. The spatial resolution of the Swift XRT (18'' half power diameter) allows a localization at the level of a few arcseconds. Once this precise position in the X-ray is obtained, multiwavelength follow-ups with radio, IR, or optical observatories (whose FOVs are typically smaller) are possible.

The combination of X-rays and follow-up radio observations provides a powerful tool to test the two most commonly known scenarios (AGN or pulsar and MSP) for our sample of Fermi unassociated sources. Sources that do not show typical characteristics of these two populations are therefore promising candidates for new discoveries.

Multifrequency radio observations of X-ray counterparts are useful to search for radio flat spectrum sources, a typical signature for an AGN. If no radio counterpart to the X-ray source is detected (or is faintly detected only at the lowest radio frequency), the γ-ray source might be a good MSP candidate. Obtaining the precise X-ray positions of the potential counterparts was a key element in optimizing our radio observation strategy. These X-ray positions were used to carry out follow-up observations with the Australia Telescope Compact Array (ATCA) radio telescope and to minimize the number of radio pointings per Fermi source. This approach allowed deeper observation of each X-ray counterpart. The typical Fermi error ellipse is 01 in the semimajor axis, and the primary beam of the ATCA radio telescope at the observed frequencies ranges from about 10 arcmin at 5.5 GHz to about 1 arcmin at 40 GHz.

From the list of 16 γ-ray sources with at least one X-ray counterpart, radio observations of seven sources were obtained with the ATCA. These objects are presented in Table 1.

We analyzed all the archival clean event files obtained in the photon counting mode by the XRT on board Swift and covering the uncertainty ellipse of each γ-ray source. In some cases several exposures of the same γ-ray source have been taken. After combining them within XSelect, we used Ximage to perform source detection. The resulting images are shown in Figure 1.

Zoom In Zoom Out Reset image size

We then used those coordinates to run "the Swift-XRT data products generator," available at the University of Leicester Web site,¹² to perform the data reduction and the early data analysis. This facility allowed the creation of (1) three combined event files of all the observations, one in the full energy band 0.3–10 keV and two in the sub-bands 0.3–1.5 keV (S band) and 1.5–10 keV (H band); (2) a combined 0.3–10 keV spectral file for both the source and the background; and (3) a 0.3–10 keV light curve, binned by observation. We used the 0.3–10 keV combined event file to estimate the significance of detection (σ_det), the combined spectral file to perform a spectral analysis for those sources with a high level of detection significance, the sub-bands event files to estimate the hardness ratios (defined as (H − S)/(H + S), where the counts are typically extracted from a 20'' circular region) for the less significant sources, and the light curve to check the presence of variability. The wide range of separation between Swift snapshots (from a few days to many months) has allowed us to test the X-ray variability of potential, bright counterparts by comparing the X-ray flux in the 0.3–10 keV energy band for each available observation. The significance of the X-ray variability was calculated as $\sigma _{\rm var}= ({\rm CR}_{\rm max} - {\rm CR}_{\rm min })/\sqrt{\vphantom{A^A}\smash{{{{\rm CR}_{\rm min,err}^{2}+{\rm CR}_{\rm max,err}^{2} }}}}$, where CR is the net count rate. We note that all the detected sources are within 10 arcmin of the center of the images, where the XRT is best calibrated, and the estimated count rates are not affected by issues at the edges of the CCD. The values of σ_var and σ_det are reported in Table 1 for X-ray sources found within the R₉₅ Fermi error ellipse at a detection significance >3σ. The position of the X-ray sources presented in Section 4.1 were obtained with the task XRTcentroid.

The Ultraviolet/Optical Telescope (UVOT) provides coverage simultaneous to the XRT by using the UVOT "filter of the day." This gives us partial, random coverage in the U, W1, M2, and W2 filters (no observations with V or B filters were performed). We analyzed the public data by using the standard tools from the Swift analysis Web page of HEASARC¹³.

For each filter, we combined all the exposures within a single observation to estimate the monochromatic flux (corrected for the finite aperture of the extraction region, coincidence loss, and large scale sensitivity). For sources with multiple observations, we kept them separated to check for variability in the bright sources. For the dim sources, the observations were combined to increase the signal-to-noise ratio or to lower the upper limit in case of nondetection. We summed both the sky images and the exposure maps by using uvotimsum. The photometry has been obtained by running uvotsource and using a circular source extraction region (with a radius varying between 3''and 5'', depending on the source intensity) and an annular background centered on the source (with an inner radius not less than 15''). In Table 1 we report if the source has been detected in each available filter. The detection threshold in uvotsource has been set to the canonical 3σ.

Regular observations with the ATCA are a key component of the TANAMI program (Ojha et al. 2010), which monitors southern hemisphere Fermi LAT detected sources at a number of radio frequencies and resolutions. Every few weeks, "snapshot" observations are made at the frequencies 5.5, 9, 17, 19, 38, and 40 GHz, where each frequency is the center of a 2 GHz wide band and the fluxes are calibrated against the ATCA primary flux calibrator PKS 1934−638 (Stevens et al. 2012).

For some candidates, there were multiple X-ray counterparts within the Fermi LAT error circle, and in many cases these counterparts were too far from each other to be observed in one pointing of the ATCA. Each observation was for 8 minutes at each band, and the first observation was made at the lowest frequency, where an AGN counterpart would likely have the highest flux density. In cases where no detection was made at the lower frequencies, no observations were made at the higher frequencies.

These positions and error ellipses were obtained by first imaging each source (that had at least some detected flux at 5.5 GHz) in a standard manner using MIRIAD. Because of the elongated beams (in most cases), the beam fitting was done using the task IMFIT. IMFIT was used to obtain the position and error ellipse after checking that the source was realistically extracted from the image.

While some of the γ-ray sources have only one detected counterpart, for the majority of them multiple X-ray counterparts have been found within, or just outside, the R₉₅ γ-ray position uncertainty reported in the 2FGL catalog. Only a handful of them have a detection significance above 5σ. The typical X-ray uncertainty position is of the order of 3''–6''.

Here we report the results of the analysis of the XRT and UVOT instruments. Whenever the statistics were sufficient, a double-absorbed power law model was fitted to the data to derive basic spectral parameters for the sources detected with the XRT. The first absorption corresponds to the Galactic value (from the Leiden/Argentine/Bonn Survey of Galactic H i; Kalberla et al. 2005), and the second is intrinsic to the source and is left free to vary in the fitting. Both absorptions were modeled using the phabs model available in XSPEC.

J0143.6−5844. In the R₉₅ Fermi LAT ellipse we found three X-ray sources in a 4.4 ks exposure. While A2 and A3 (R.A. = 01^h43^m4985; decl. = −58°43'192; and R.A. = 01^h43^m3861; decl. = −58°41'502) have a low detection significance and are moderately soft (hardness ratio HR = −0.6 and −0.3, respectively), A1 (R.A. = 01^h43^m4757; decl. = −58°45'516) is a bright object with 1590 counts. The spectral analysis revealed an absorption along the line of sight in slight excess over the Galactic value (n_{H, Gal} = 2.04 × 10²⁰ cm⁻², $n_{{\rm H},{\rm int}}=3.3_{-0.2}^{+6.5} \times 10^{20}$ cm⁻²) with a photon index Γ = 2.29 ± 0.12. Swift observed the FOV of this γ-ray source three times. While the hardness ratio of A1 does not change over time (indicating no spectral evolution), the net, corrected count rate varies between 0.503 ± 0.024 and 0.671 ± 0.026, with a significance of variability of σ_var ∼ 4.8. Both the brightness changes and the shape of the spectrum are typical characteristics of an AGN. All three sources were detected at UV frequencies in the U and W2 filters. In the first and last observation, both performed with the W2 filter, no sign of flux variation is seen for A1, while only one observation was performed with the U filter and no variability can be assessed.

J0523.3−2530. The only possible X-ray counterpart detected by Swift in a 4.8 ks exposure is a dim object with 11 counts (R.A. = 05^h23^m1711; decl. = −25°27'319), not found in the UV band.

J0803.2−0339. Also, in this case only one X-ray source has been found (R.A. = 08^h03^m1211; decl. = −03°36'14) but at relatively high significance. The 80 counts detected in 3.9 ks were distributed following a power law of spectral index $\Gamma = 2.10_{-0.31}^{+0.60}$. Swift performed two observations on this object, and the light curve analysis shows some sign of variability (σ_var = 3.5) with the count rate changing from 0.018 ± 0.003 to 0.039  ±  0.005. Also, in this case no significant optical variability has been found in the two observations with the W1 filter.

J1036.1−6722. The FOV of this source has been observed extensively (35.5 ks, spanning two years), but no bright X-ray sources have been detected. Only two marginal sources can be found above the 3σ threshold: D1 (R.A. = 10^h35^m4598; decl. = −67°25'154) and D2 (R.A. = 10^h36^m2213; decl. = −67°22'285). The sources are very faint also in the UV band and are detected only at longer wavelengths because of the long exposures. Because of the low statistics, no variability analysis can be performed. The number of X-ray counts for the two sources is 34 and 10, respectively. But while in D1 the counts are evenly split below and above 1.5 keV ($\Gamma =1.9_{-0.8}^{+0.7}$), they are only in the soft band in D2.

J1129.5+3758. In four pointings (for a total of 4.8 ks), only two faint X-ray sources were detected: E1 (R.A. = 11^h29^m0342; decl. = +37°56'586) and E2 (R.A. = 11^h29^m3112; decl. = +38°01'596). The total counts are 11 and 10, distributed in a 2:1 ratio between the soft and hard band (HR = −0.3 and −0.4). Both sources are detected in the UV range.

J1231.3−5112. Also, in this case only dim objects were detected. F1 (R.A. = 12^h31^m5131; decl. = −51°19'399), F2 (R.A. = 12^h30^m5213; decl. = −51°19'171), and F3 (R.A. = 12^h31^m2962; decl. = −51°09'323) have 16, 5, and 15 counts observed in a 7.3 ks exposure. F1 is a hard source (HR = 0.25), while the other two sources have counts in the soft band only. While F1 and F2 have dim optical counterparts, F3 is a bright star, making this source an unlikely association of the γ-ray emission.

J1844.3+1548. Both the X-ray and UV analyses suggest that the likely counterpart of the γ-ray source is an AGN. The X-ray position (R.A. = 18^h44^m2542; decl. = +15°46'443) coincides with a known radio source (NVSS J184425+154646). The spectral analysis of the 4.2 ks exposure indicates a very steep photon index ($\Gamma = 3.00_{-0.39}^{+0.42}$), an intrinsic absorption well above the Galactic value (n_{H, Gal} = 1.73 × 10²¹, and n_{H, int} = 2.63 ± 0.06 × 10²¹), and a 0.3–10 keV unabsorbed flux $F_X = 9.5_{-3.2}^{+6.7} \times 10^{-12}$ erg cm⁻² s⁻¹. The X-ray evolution indicates that the source count rate varied between 0.110 ± 0.007 on 2011 November 18 and 0.019 ± 0.007 on 2011 November 30, and it remained constant at (0.020 ± 0.006) on 2011 December 4. The UV light curve shows variation as well: in the M2 filters the source brightness declined by a magnitude from the first (m_M2 = 19.00 ± 0.08) to the last observation (m_M2 = 19.99 ± 0.25).

Seven of our 16 Fermi unassociated sources were observed over two epochs on 2012 August 27 and 2012 September 4. Altogether, a total of 13 radio pointings were required to observe all the possible X-ray counterparts.

Six of our seven Fermi sources were detected in at least one radio frequency at one of their X-ray candidate counterpart locations. The flux density of the sources detected ranges from 8 mJy to 97 mJy (sources B1 and F1, respectively) at 5.5 GHz. All the radio sources reported in Table 1 are within the 1σ error ellipse of their corresponding X-ray source. Unique radio counterparts to the X-ray detection were found for the four sources (A1, C1, E1, G1) that were detected in multiple radio bands.

In the case of the sources J0523.3−2530 and J1036.1−6722, the error ellipse of our radio observation includes three discrete sources identified by earlier radio surveys. Thus, we cannot determine which of these three sources (or which two or all three) are being detected. Higher resolution observations and/or multi-epoch variability observations will be required to firmly establish the counterpart to these two Fermi candidates.

For J0143.6−5844 in addition to A1, two alternate nearby radio sources (A2 and A3) were detected with a much lower radio flux density, almost certainly because of the large beam¹⁴ at 5.5 GHz. J1129.5+3758 was observed to have a flat spectrum between two observed frequencies (5.5 and 9.0 GHz). Observations at higher frequencies were not performed because of its low elevation at the ATCA, which would make such observations problematic. We will confirm that its flat spectrum extends to higher frequencies in future observations with a more northern telescope. The only γ-ray source where no radio sources were detected is J1231.3−5112, and an upper limit of ∼10 mJy at 5.5 and 9 GHz can be placed on all three radio pointings.

In follow-up observations on 2012 August 26 with the two element Ceduna Hobart Interferometer (CHI; Blanchard et al. 2012) that has a resolution of 6.6 mas at the observed frequency of 6.7 GHz, source A1 in J0143.6−5844 was detected with a flux density of 25 ± 4 mJy. This compares to a flux density of 26 ± 3 mJy measured by the ATCA at the adjacent frequency of 5.5 GHz. Thus, we conclude that J0143.6−5844 is a very compact object. CHI observations of the other detected sources are also planned.

An independent radio follow-up of unassociated 2FGL sources by Petrov et al. (2013) found the same candidate counterparts for J0143.6−5844 and J0803.2−0339 and confirmed our nondetection of J1231.3−5112.

In order to discuss the multiwavelength properties of the counterparts presented in the previous section, spectral energy distributions (SEDs) were built for the γ-ray sources where a radio counterpart was detected in at least two frequencies (2FGL J0143.6−5844, J0803.2−0339, J1129.5+3758, and J1844.3+1548), making these sources strong candidates to be AGNs. To generate the SEDs for each candidate, we used the ASDC SED Builder (Stratta et al. 2011), a web-based program developed at the ASI Science Data Center that combines data from several missions and experiments together with catalogs, archival, and proprietary data. We used fluxes from our radio campaign, as well as from the UVOT images and the XRT events. The optical/UV fluxes were corrected for Galactic extinction, while we used the unabsorbed X-ray flux in the 0.3–1.5 and 1.5–10 keV energy ranges (or 0.3–10 keV only for sources with a limited number of counts). In the case of J1129.5+3758, where no spectral information could be derived because of the low statistics, the X-ray flux was derived assuming an absorbed power law with an index of 2.0 (similar to what is observed for the other AGN candidates), and the value of the Galactic n_H along the line of sight was used. The SEDs are shown in Figure 2 and are discussed individually below.

Zoom In Zoom Out Reset image size

It is interesting to note that Mirabal et al. (2012) assign probabilities of 0.962 and higher for these four objects to be AGNs. Using an improved version of their infrared colors prediction method, a list of new γ-ray AGN candidates is presented in Massaro et al. (2013). When a prediction is available, a comparison of the results is presented.

J0143.6−5844. Among these four sources, the SED of A1 (Figure 2, top right panel) most closely resembles what has been observed in other blazar candidates: there is a flat radio spectrum that rises to a peak in the UV/soft X-ray range and declines in the X-rays. This might be interpreted as the synchrotron component of the emission observed in the high-peaked (BL Lacertae) blazars. The flat spectrum observed in the Fermi LAT data suggests that the high-energy photons belong to a different emission process, likely inverse Compton upscatter of photons either in the relativistic jets (synchrotron self-Compton) or from outside the jets (external Compton). Other supporting evidence is found in Massaro et al. (2013), where the infrared source WISE J014347.39−584551.3 (spatially coincident with the X-ray source A1) is classified as an AGN and more specifically as a BL Lac candidate.

J0803.2−0339. The combination of archival and new radio data might indicate that the source C1 has evolved with time, moving from a flat to a steep radio spectrum. However, these data are not simultaneous and should be interpreted with caution. This kind of behavior has been seen in other blazars (e.g., PKS 0521−36; Tornikoski et al. 2002) when the steep spectrum is seen during an active period. Furthermore, the observations of a steep radio spectrum in a quasar by LAT is not uncommon (Abdo et al. 2010b). The emission from radio to X-rays is again dominated by synchrotron radiation, although in the optical a sign of the contribution from the host galaxy is evident. Also, in this case the high-energy photons do not seem to be due to the same radiation mechanism and may be produced by inverse Compton upscattering. In addition, we note that the variability index reported in the 2FGL catalog for this source (TS_var = 48.58; Nolan et al. 2012) indicates a γ-ray variability with a >99% confidence level¹⁵, supporting an AGN nature for this object. For this source, no γ-ray AGN candidate was found in Massaro et al. (2013).

J1129.5+3758. The SED of E1 is very puzzling; while the galaxy contribution is an evident feature in the near-IR to the UV, the radio, X-ray, and γ-ray emissions are more difficult to reconcile. The relatively steep spectrum in the GeV range and the very low value of the X-ray flux indicate that a peak of emission must be found between these two ranges. Considering that the hardness ratio in the X-rays is −0.3, this might indicate a somewhat flat spectrum and, consequently, that the X-ray emission might be produced by the same mechanism as the γ-rays. Unfortunately, the lack of a detailed X-ray spectrum does not allow us to draw a firm conclusion. In any case, the parabola that would fit the inverse Compton emission in the X-rays and the γ-rays would be very narrow. The radio emission, meanwhile, cannot be an extrapolation of the higher energy emission. If this counterpart is indeed a blazar, then the radio emission might be produced by synchrotron with a very low peak emission, particularly when compared with the inverse Compton peak. The ratio between the synchrotron and the inverse Compton peak might differ by a factor of 2–3 orders of magnitude, a relatively uncommon feature among blazars (see, e.g., Giommi et al. 2012). We also note that there is a hint of γ-ray variability in the aperture photometry light curve provided by the Fermi Science Support Center¹⁶. However, this variability is probably contamination from the nearby (angular separation of 17) flaring γ-ray source 2FGL J1127.6+3622.

J1844.3+1548. The SED of G1 is another intriguing case; the near-IR to the UV is thermal radiation produced by the stars in the host galaxy. The very steep X-ray spectrum is a feature observed in narrow-line Seyfert 1 galaxies (NLSy1, see, e.g., Grupe et al. 2004, and references therein). In those objects, the steep spectrum may be due to intense soft X-ray flux cooling the accretion disk corona (Maraschi & Haardt 1997). The soft flux is not observed in this source because of the relatively high intrinsic absorption found in the X-ray analysis. If G1 is really an NLSy1, then it belongs to the rare class of radio-loud NLSy1 galaxies, a class that have been found to be a γ-ray emitter (Abdo et al. 2009; Foschini et al. 2011; D'Ammando et al. 2012). In these sources, the emission in the radio and the γ-ray is produced by a blazar-like relativistic jet that is dissipating most of its energy beyond the broad-line region (Ghisellini & Tavecchio 2008). We note that the infrared source WISE J184425.36+154645.9 (spatially coincident with the X-ray source G1) is also classified as a BL Lac candidate in Massaro et al. (2013).

J1036.1−6722. For this object, only two X-ray sources with significance >3σ are detected in a 35 ks observation, the deepest for our sample. Counterparts in the radio were found for only one X-ray source (D1) and were only detected at 5.5 GHz, indicating a steep radio spectrum. In γ-rays, the best-fit spectral model reported in the 2FGL is a LogParabola. In addition, no infrared AGN candidates are reported in Massaro et al. (2013) for this γ-ray source. Those characteristics are reminiscent of what is seen for the pulsar/MSP population observed with the Fermi LAT. This γ-ray source is therefore a prime candidate to search for γ-ray pulsation at the position of the newly detected counterpart. The much reduced position error ellipse (from 01 to ∼4'') will allow exploration of the parameter space (e.g., pulsar period) in considerably greater detail. We note that in the case of a binary system, the measurement of the orbital period (through, e.g., optical observations) will greatly enhance the probability of detecting the pulsation.

J0523.3−2530. Although J0523.3−2530 is the brightest γ-ray source in our sample, only one faint X-ray source was detected, with a radio counterpart only at the lowest radio frequency. While the radio, UV, and X-ray observations might suggest a pulsar origin, the γ-ray spectrum is best represented by a power law as seen for other AGN candidates. However, Massaro et al. (2013) report no AGN candidate for this source, and interestingly, the probability to be an AGN is P (AGN) = 0.738 (Mirabal et al. 2012), a value that places this source neither in the pulsar nor in the AGN class.

J1231.3−5112. The two interesting X-ray counterparts (F1 and F2) have dim UV counterparts and no detection in the radio. This unidentified source has intriguing γ-ray properties, as the best-fit spectral model, a LogParabola, may suggest a pulsar origin while the light curve is not constant, with a γ-ray variability index of (TS_var = 39.04; Nolan et al. 2012). As for the previous source, the probability to be an AGN is at an intermediate value, P (AGN) = 0.554.

Further radio and X-ray follow-up observations of these sources are therefore required to understand their nature.

Although the X-ray to radio connection, combined with the detection of X-ray variability and the presence of a flat radio spectrum, is a step forward in pinpointing AGNs, a simple detection in the optical/UV/X-ray regimes is insufficient in determining the correct counterpart for pulsar-like γ-ray sources. For this category of sources, the probability of chance coincidence was investigated. Among the three non-AGN candidates, two of them (J0523.3−2530 and J1231.3−5112) had no Swift XRT detection outside the R₉₅ error ellipse. The X-ray sources presented in Table 1 are therefore promising counterpart candidates. For J1036.1−6722, whose Swift exposure time was longer (35.5 ks, compared to a standard ∼4 ks for the other sources), 13 other sources were observed outside the R₉₅ region. The majority (eight) of these sources have a detection significance just above 3σ, and no additional information can be obtained. We tentatively tried to look for peculiar behaviors for the remaining five sources, whose detection significance is in the range 4.5–6.3σ, by using the XRT and the UVOT data. The analysis of the 26 XRT observations does not show any sign of significant variability (above 1σ). Inspection of the five UVOT filters shows that of the five X-ray sources, one does not have any optical/UV counterpart, two have very weak counterparts, and the other two have strong counterparts. Unfortunately, we are not able to estimate the variability of the latter two sources, since they are located outside the UVOT FOV¹⁷ most of the time. We estimated the chance coincidence of finding X-ray sources within the R₉₅ region following the method explained in Bloom et al. (2002). The probability of chance association P can be expressed as P = 1 − e^{−A × ρ}, where A is the area of the R₉₅ region (a 0062 × 0058 ellipse), while ρ is the sky density of objects with equal or greater X-ray brightness. Since there are 15 sources in the XRT FOV (a circular shape 12' in radius) detected above the 3σ level (D2 in Table 1), the chance probability is 0.74.

While this value is high, D1 has a slightly higher chance to be the γ-ray counterpart than other X-ray detected sources because it is located within the R₉₅ region, is the second brightest object in the XRT field, and is detected at 5.5 GHz. However, other sources detected in the XRT field can still be considered as candidates for the association.