A&A 489, 849-883 (2008)
DOI: 10.1051/0004-6361:200809685

A revised catalogue of EGRET ${\sf\gamma}$-ray sources

J.-M. Casandjian - I. A. Grenier

Laboratoire AIM, CEA/DSM, CNRS, Université Paris Diderot, Service d'Astrophysique, CEA Saclay, 91191 Gif-sur-Yvette, France

Received 29 February 2008 / Accepted 23 May 2008

Aims. We present a catalog of point $\gamma$-ray sources detected by the EGRET detector on the Compton Gamma Ray Observatory. We used the entire $\gamma$-ray dataset of reprocessed photons at energies above 100 MeV and new Galactic interstellar emission models based on CO, HI, dark gas, and interstellar radiation field data. Two different assumptions are used to describe the cosmic-ray distribution in the Galaxy to analyse the systematic uncertainties in source detection and characterization.
Methods. We applied a 2-dimensional maximum-likelihood detection method similar to that used to analyze the 3rd EGRET catalogue.
Results. The revised catalogue lists 188 sources, 14 of which are marked as confused, in contrast to the 271 entries of the 3rd EGRET (3EG) catalogue. We do not detect 107 sources discovered previously because additional structure is present in the interstellar background. The vast majority of them were unidentified and marked as possibly extended or confused in the 3EG catalogue. In particular, we do not confirm most of the 3EG sources associated with the local clouds of the Gould Belt. Alternatively, we found 30 new sources that have no 3EG counterpart. The new error circles for the confirmed 3EG sources largely overlap the previous ones, but several counterparts of particular interest discussed before, such as Sgr A*, radiogalaxies, and several microquasars are now found outside the error circles. We cross-correlated the source positions with a large number of radio pulsars, pulsar wind nebulae, supernova remnants, OB associations, blazars and flat radiosources and we found a surprising large number of sources (87) at all latitudes that have no counterpart among the potential $\gamma$-ray emitters.

Key words: catalogs - gamma rays: observations - ISM: structure

1 Introduction

The Energetic Gamma-Ray Experiment Telescope (EGRET), which operated onboard the Compton-Gamma Ray Observatory from April 1991 to May 2000, detected photons in the 20 MeV to 30 GeV range. The observation program made use of the large instrumental field of view (25$^{\circ }$ in radius) to cover the entire sky and complete in-depth studies of specific regions. The corresponding exposure and flux sensitivity to point sources are therefore not uniform across the sky. The sensitivity threshold also varies because of the intense background emission that arises from cosmic-ray interactions with the interstellar gas and photon fields in the Milky Way. The minimum flux detectable by EGRET rises steeply with decreasing Galactic latitude. To be able to detect point sources and assess their significance in these varying conditions, a 2-dimensional maximum-likelihood method using binned maps was developed for analyzing the COS-B data (Pollock et al. 1981) and implemented in the study of EGRET data set (Mattox et al. 1996). A first catalog generated with this method was published for the first 1.5 years of data (Fichtel et al. 1994), followed by the second one (Thompson et al. 1995), and its supplement (Thompson et al. 1996) after 3 years of data. Lamb & Macomb (1997) presented a catalog of sources detected above 1 GeV. The last EGRET catalog (hereafter 3EG, Hartman et al. 1999) comprised reprocessed data from April 1991 to October 1995 and the interstellar emission model from Hunter et al. (1997) and extragalactic background from Sreekumar et al. (1998). This version contained 271 point sources including a solar flare, the Large Magellanic Cloud, five pulsars, one radiogalaxy detection (Cen A), 66 high-confidence identifications of blazars (BL Lac objects and flat-spectrum radio quasars), and 27 lower-confidence blazar identifications. Because of the wide tails of the instrument point-spread function, seven potential artifacts were noted around the brightest sources and many sources were marked as confused or possibly extended.

The 3EG catalogue also contained 170 sources with no attractive counterpart at lower energy. About 130 of them remain unidentified (see Grenier (2004) and references therein). Candidate counterparts included pulsars and their wind nebulae, supernova remnants, massive stars, X-ray binaries and microquasars, blazars and nearby radiogalaxies, luminous infrared and starburst galaxies, and galaxy clusters. It was also noticed (Grenier 1995; Grenier 2000; Gehrels et al. 2000) that the most stable unidentified sources were correlated significantly with the nearby Gould Belt, a system of massive stars and interstellar clouds that surrounds the Sun at a distance of hundreds of parsecs. The offset position of the Sun with respect to the Belt centre and the Belt inclination of 17$^{\circ }$ with respect to the Galactic plane provides a useful spatial signature across the sky (Perrot & Grenier 2003).

EGRET continued to observe for an additional 4.5 years following the 4 cycles used in the 3EG analysis. Its sensitivity was reduced because of the ageing gas in the spark chamber, but it gathered almost an additional ten percent of photons and detected several new variable sources. Several authors (Sowards-Emmerd et al. 2005; Nolan et al. 2003), however, noticed discrepancies between their studies and at least five 3EG sources. They failed to confirm some and found additional. The entire $\gamma$-ray dataset and final instrument response functions were also reprocessed significantly by the EGRET team in 2001. Furthermore, the spatial coverage of CO surveys has reached higher latitudes since 1999, finding new small CO clouds (Dame et al. 2001). In parallel, new HI surveys (Kalberla et al. 2005) have been completed to correct for the significant contamination of stray radiation present in the older surveys. Finally, an additional `dark' gas component was discovered in the Gould Belt clouds that increased their estimated mass and spatial extent significantly (Grenier et al. 2005). The additional mass is structured into large envelopes around the dense CO cores, and does not follow the HI and CO maps commonly used to trace atomic and molecular column-densities. The dark gas therefore provides both $\gamma$-ray intensity and structure that were not accounted for in the 3EG background model.

For all of these reasons and in preparation for the new GLAST mission, the interstellar background model had to be revised and the EGRET detection method was applied to the full nine years of data to build a revised catalogue of sources above 100 MeV. To study the systematic uncertainties in source locations and fluxes due to our limited knowledge of the intense interstellar background, we applied our analysis to two different background models, exploiting the same new interstellar data, but using independent approaches to constrain the cosmic-ray gradient across the Galaxy.

2 The Galactic interstellar emission models

\par\includegraphics[width=16.5cm,clip]{9685fig1.eps}\end{figure} Figure 1: The top figure is the longitude profile of all photon counts observed by EGRET above 100 MeV at all latitudes (black error bars), compared with the diffuse counts predicted by the 3EG model (blue curve) and the Ring model (red curve). The bottom figure is the residual expressed in number of standard deviation, colors are the same as above, we added the Galprop residuals in purple. Counts from bright sources have been added to the diffuse component. For more visibility the plot are presented with a binning of 4$^{\circ }$.
Open with DEXTER

The high-energy Galactic emission is produced by the interaction of energetic cosmic-ray electrons and protons with interstellar nucleons and photons. The decay of neutral pions produced in hadron collisions accounts for most of the emission above 300 MeV. Inverse Compton (IC) scattering of the interstellar radiation field by electrons and their bremsstrahlung emission in the interstellar gas are the other main contributors to the Galactic emission. The observed intensity therefore scales with the integral along the line-of-sight of the product of the cosmic-ray density and the gas or soft-photon density.

The diffuse model used to produce the 3EG catalogue (Hunter et al. 1997) was based on a 3D-distribution of matter, cosmic-ray, and soft-photon densities in the Galaxy, where the cosmic-ray density was assumed to be coupled to the gas density over a given length scale. This length scale and the CO-to-H2 conversion factor (X ratio) were adjusted to the data. The 3D gas map was obtained from the HI and CO line surveys and from kinematical distances derived for circular rotation. Distance ambiguities in the inner Galaxy were solved by dividing the gas into the far and near sides according to its expected scale height. Gas with velocities in excess of the tangent values was attributed to the tangent point; the gas emission within 10$^{\circ }$ of both the Galactic centre and anticentre was interpolated from the regions just outside these boundaries and normalized to match the total emission observed along the line-of-sight. The resulting map is, however, still strongly biased to our side of the Galaxy, particularly for the atomic gas. This bias is reflected in the cosmic-ray density via the coupling length.

For the present analyses, we assumed an axisymmetric Galaxy for the cosmic-ray density and we used gas column-density distributions in Galactocentric rings that are less affected by biases caused by the strategy adopted to measure the cloud distance in the inner Galaxy. The radial velocity information in the HI and CO line surveys together with the rotation curve of Clemens (1985) and the solar motion parameters (v = 220 km s-1 at R = 8.5 kpc), were used to partition the gas into 6 rings bounded by 3.5, 7.5, 9.5, 11.5, and 13.5 kpc in Galactocentric distance (Digel et al., in preparation). Gas within 10$^{\circ }$ of the Galactic centre and anticentre was interpolated as before. The all-sky Leiden-Argentina-Bonn (LAB) composite survey (Kalberla et al. 2005) was used for the HI data. Column densities, $N({\rm HI})$, were derived by assuming a constant spin temperature of 125 K. The velocity-integrated CO brightness temperature, $W({\rm CO})$, was taken from the Center for Astrophysics compilation of observations at $\vert b\vert \leq 32 ^{\circ}$(Dame et al. 2001). The regions outside the survey boundaries should be free of bright CO emission.

We used two different approaches to account for the cosmic-ray density gradient. One is based on the Galprop model for cosmic-ray propagation developed by Strong et al. (2007, 2004a, 2004b), using run number 49-6002029RB to derive the $\gamma$-ray maps from pion decay, $I_{\pi^0}$, bremstrahlung radiation, $I_{\rm brem}$, and inverse Compton radiation, $I_{\rm IC}$. This version includes secondary electrons and positrons, an optimized cosmic-ray spectrum to reproduce the GeV excess in the EGRET data, a cosmic-ray source distribution matching the radial profile of pulsars and supernova remnants, a radial gradient in the X factor, and the new HI and CO gas rings.

The second model, hereafter referred to as the Ring model, is based on the simpler, but realistic hypothesis that, if energetic cosmic rays penetrate uniformly all gas phases, the $\gamma$-ray intensity in each direction can be modelled by a linear combination of gas column-densities in the different rings, plus the IC intensity map (as predicted by Galprop), and an isotropic intensity ( $I_{\rm iso}$) that accounts for local IC emission and extragalactic emission. This assumption was used to derive gas emissivities in several rings from the COS-B and EGRET data (Strong et al. 1988; Strong & Mattox 1996). We repeated these analyzes to derive gas emissivities for the new HI and CO rings using 9 years of EGRET data in three energy bands (>100 MeV, 0.3-1 GeV, >1 GeV). Both the Ring and Galprop models used the revised distribution of the interstellar radiation field (Moskalenko et al. 2006; Porter & et al. 2005) to calculate the IC intensity map. The Galprop IC map is common to both diffuse models.

As indicated in the introduction, we also included in the local ring the large column-densities of ``dark'' gas associated with cold and anomalous dust at the transition between the atomic and molecular phases (Grenier et al. 2005). This transitional phase is not traced in the radio. After removing from total dust column-density maps the part that correlates linearly with $N({\rm HI})$and $W({\rm CO})$, large envelopes of excess dust remain surrounding nearby CO. The fact that the excess dust correlates spatially with significant diffuse gamma radiation indicates that cosmic rays pervade gas not accounted for in HI or CO. As inferred from the excess dust and correlated $\gamma$-ray data, the gas-to-dust ratio in this phase is normal. This phase appears to form an extended layer at the transition between the dense CO cores and the densest parts of the outer HI envelope of a cloud complex. It is observed most clearly in total dust maps such as the reddening E(B-V) map (Schlegel et al. 1998), or low-frequency thermal emission at 93 GHz for WMAP (Finkbeiner et al. 1999), or anomalous emission close to 20 GHz (Lagache 2003). We constructed a ``dark'' gas column-density template, $NH_{\rm dark}$, by removing from the E(B-V) map the part that was correlated linearly with $N({\rm HI})$ and $W({\rm CO})$. This template was converted into gas column-densities by fitting the all-sky $\gamma$-ray maps with this template as well as $N({\rm HI})$ and $W({\rm CO})$ rings, IC and isotropic components. Because of its column-densities, clumpiness, and large spread across the sky (see Fig. 4 in Grenier et al. 2005), the ``dark'' gas component may strongly affect source detectability. This template was also added to the Galprop 49-6002029RB background model.

To summarize, two diffuse backgrounds were constructed by fitting different components to the EGRET photon maps, in $0.5^{\circ}
\times 0.5^{\circ}$ bins, in the three energy bands that we use for source detection (>100 MeV, 0.3-1 GeV, >1 GeV). With the Ring model, the predicted count rates are calculated as:

  \begin{eqnarray*}N_{\rm pred}(l,b) &=& [\Sigma_{i={\rm rings}} q_{{\rm HI},i} N_...
... \Sigma_{j={\rm sources}} \epsilon(l_j,b_j)~ f_{j}~ PSF(l_j,b_j)

and the Galprop model as:
                           $\displaystyle N_{\rm pred}(l,b)$ = $\displaystyle [q_{\pi^0} I_{\pi^0}(l,b) + q_{\rm brem}
I_{\rm brem}(l,b) + q_{\rm dark} NH_{\rm dark}(l,b)$  
    $\displaystyle + q_{\rm IC} I_{\rm IC}(l,b) + I_{\rm iso}] \times \epsilon(l,b)$  
    $\displaystyle + \Sigma_{j={\rm sources}} \epsilon(l_j,b_j)~ f_{j}~ PSF(l_j,b_j).$ (1)

In both models, $\epsilon(l,b)$ and fj represent the EGRET exposure map and source fluxes respectively. The product of the diffuse maps and the exposure were convolved with the EGRET PSF for an input E-2.1 spectrum before adding the source maps. The EGRET count and exposure maps, the 3EG diffuse model, as well as the latest instrument response functions, were downloaded from the CGRO Science Support Center. They differ from those used for 3EG since they were reprocessed in 2001. The q parameters (gas emissivities or relative contributions of different radiation components) were fitted to the data by means of a maximum likelihood with Poisson statistics. To avoid biasing the interstellar parameters, the model included the brightest sources detected during a first source detection iteration with a significance >5$ \sigma$, with fixed fluxes. Changing these fluxes within their statistical uncertainties do not significantly change the diffuse results.

The resulting emissivities corresponding to the local gas are fully consistant with Table 1 of Grenier et al. (2005). The emissivity gradient in the Galactic plane will be described in a separate paper. The quality of the fit can be seen in Fig. 1. The top figure displays the longitude profile of all the EGRET photon counts above 100 MeV. The error bars are only statistical. The plot compares the best fit that can be obtained using the former 3EG diffuse model with the longitude profile resulting from the present Ring model. The bottom plot shows the longitude profile of the residuals and the improvement of the ring model over the 3EG one. It also shows the residuals for the best fit Galprop model. All modelled profiles include the brightest sources. Systematic differences can be seen in various places where the 3EG model significantly over predicts and under predicts the data while the new models describe the data more accurately. Because of its larger flexibility (the gas emissivity gradient due to cosmic-ray variations is measured, not inferred from propagation properties or gas coupling), the Ring model was found to best fit the data. We note than even if the agreement is excellent, there still exists small deviations that can significantly impact source detection and characterization.

\par\includegraphics[width=9cm,clip]{9685fig2.eps} .
\end{figure} Figure 2: Map in Galactic coordinates of the residuals (expressed in $\sigma = \sqrt {N_{\rm pred}}$ values) between the E > 100 MeV photon counts (in 0.5$^{\circ }$ bin) and the best-fit Ring model solution given by Eq. (1).
Open with DEXTER

The residual count map obtained above 100 MeV with the Ring model is presented in Fig. 2. It displays the statistical difference $(N_{\rm obs}-N_{\rm pred})/ \sqrt{N_{\rm pred}}$ between the observed counts and those predicted from the diffuse background and bright sources using Eq. (1). The model globally fits the data very well. The extended blue fan-like structures with negative residuals are correlated with the edge of several observing periods. They are probably caused by an incorrect estimate of the exposure at large angles from the instrument axis. They are visible independently of the choice of diffuse model (Ring, Galprop, or 3EG). Their spatial extent is sufficiently large compared to the PSF size not to severely affect source detection, yet source fluxes in these directions are underestimated. Uncertain knowledge of the off-axis instrument exposure is also reflected in the small model deficit (orange edge) bordering the fan-like excesses. We checked for suspicious strings of faint sources that would correlate with these instrumental features.

The use of two different background models allowed us to study their impact on source detection and characterization. Given its higher likelihood value and locally flatter residuals, the Ring model was used to derive the default source flux and location. The values obtained with the Galprop background are used to illustrate the amplitude of the systematic uncertainty due to the background modelling. When searching for sources, we used the diffuse emission parameters calculated from this global fit. We adjusted the source flux and a free normalization of the total diffuse flux within 15$^{\circ }$ around each pixel, and a free isotropic flux. This procedure is the same as used for 3EG (Gmult and Gbias). These two parameters correct for small local mismatches between the diffuse model and the data. Gmult fluctuates around 1.

Table 1: List of individual or short periods used in the analysis in addition to the summed cycles.

3 Source detection

As for deriving the 3EG catalogue, we used the LIKE code (Mattox 1996, version 5.61) to compute the 2-dimensional binned Poisson likelihood of detecting a source at a particular location on top of the diffuse background. LIKE calculates the Test Statistic (TS) value that compares the likelihood of detecting a PSF-like excess above the background to the null hypothesis - a random background fluctuation - for a given position. The likelihood (Li) is calculated as the product, for all pixels within 15$^{\circ }$ of a specific position, of the Poisson probabilities of observing photons in a pixel where the number of counts is predicted by the model (background + source). The likelihood ratio test statistic is defined to be TS = -2 (Ln L0 - Ln L1), where the likelihood values L1 and L0 are optimized respectively with and without a source in the model. Asymptotically, the TS distribution follows a $\chi^{2}$ one. The detection significance of a source at the given position is $\sqrt{TS} \sigma$ (Mattox 1996).

Sources were searched for in the summed maps corresponding to cycle 1, 2, 3, 4, 1+2, 3+4,1+2+3+4, 5+6, 7+8+9, 1+2+3+4+5+6+7+8+9. In addition, we analyzed the 46 individual periods listed in Table 1 for which flaring 3EG sources were detected. As for the summed maps, the individual period maps retained only photons with inclinations within 30$^{\circ }$ from the instrument axis, or $19^{\circ}$ for cycle 6, 7, 8, and 9. Photons and exposure maps were binned to $0.5^{\circ}
\times 0.5^{\circ}$.

To build the 3EG catalogue, sources were detected only in the integrated E > 100 MeV band. TS maps were then constructed in three energy bands (>100 MeV, 0.3-1 GeV, and >1 GeV) from the observation (single or summed) with highest TS and a source final position was obtained from the smallest error contours. Given the modern computer performance, we directly searched for sources independently in the three energy bands.

At 100 MeV, the EGRET PSF is wide and discrepancies exists between its real shape, as observed in bright sources, and its modelled one. In practice, differences may be caused because the source spectrum is more complex than the single power-law assumed to integrate the PSF. A choice of 300 MeV instead of 100 MeV for the lower analysis threshold might have provided a more effective compromise between count rates for detection and systematic uncertainties in the PSF. We, however, retained a lower limit of 100 MeV, as in 3EG, to account for soft sources and to allow comparison with the 3EG results. We assumed a spectral index of 2.0 for all sources, apart from 11 bright sources with a 3EG spectral index that differed significantly from 2.0, for which we integrated the PSF using their 3EG index.

Each of the 10 all-sky summed maps was divided, both in Galactic and equatorial coordinates, in 45 zones with a large overlap. The use of both coordinates systems is required since source images are deformed in rectangular projection at high latitude or declination. For each zone, each individual period, and each of the 3 energy bands (>100 MeV, 0.3-1 GeV, and >1 GeV), we calculated a TS map for excesses above the background. Sources were iteratively detected from high TS to low TS in successive TSmaps. Between each steps, the detected sources were included in the background model until no excess with $\sqrt{TS} > 3$ remained in the final TS map. An example of the iteration around Geminga is given in Fig. 3. Peaks in the TS map were automatically detected with SExtractor (Bertin & Arnouts 1996) and converted into source positions by taking the TS-weighted centroid in the region enclosed by the 95% confidence contour around this position. Source positions were recalculated at each iteration to take into account the influence of the neighbouring sources. More than 1100 TS-maps were thus calculated at the CCIN2P3 Computing Center.

\par\includegraphics[width=9cm,clip]{9685fig3.eps}\end{figure} Figure 3: An example of the iterative source detection with the 2D binned likelihood around Geminga at energies above 100 MeV. 4 consecutive TS maps are shown. Sources are detected, then are included in the background for the next step until no significant one is left. The colourbar gives TS.
Open with DEXTER

4 Catalogue construction

To account for real versus modelled PSF discrepancies in extremely bright sources, for instance to account for the splitting in two of the bright pulsar sources or for the artifacts in the Vela tails, we removed all source candidates within $3.5^{\circ}$ of the intense sources (that exhibit more than 800 photons in a map). For less intense sources, we checked the probability of having a double versus single source with a specific likelihood calculation, using the likelihood ratio between the 2 cases to keep or reject the double source.

At the end of this stage, most sources have two possible positions per energy band and observation, one from the Galactic coordinate map and one from the equatorial one. We cross-compared the two and selected the position from the least deformed projection. Sources detected only once were not included in the list unless their latitude or declination were higher than $40^{\circ}$ or their longitude or right ascension were less than $5^{\circ}$ from the map edges.

At this stage, most sources have three possible positions (with energy) for a given observation. We chose among the three the position corresponding to the smallest 95% confidence contour, unless its peak $\sqrt{TS}$ was a factor of 1.5 smaller than found in another energy band. The latter condition reduces the risk of incorrect source assignment during the cross-comparison phase. Sources found at low energy, but not at high energy were included in the list, as well as sources found only at high energy.

Table 2: The EGR catalogue. The three first sources are shown. The full catalogue is available after the references.

We used the same criteria to cross-compare the source positions for individual periods and summed cycles to obtain a final list of candidate sources with the most accurately determined position from the different energy bands and periods/cycles. We followed the entire procedure for both the Ring and Galprop interstellar backgrounds. We obtained respectively 1192 and 1225 candidate sources with the Ring and Galprop models. Source fluxes and $\sqrt{TS}$ values above 100 MeV were calculated for these sets of positions for the different periods and cycles. Unlike in 3EG, we did not adjust the position of the identified sources (AGN or pulsars) to that of their radio counterpart.

We adopted the same detection threshold as for the 3EG catalogue ( $\!\!\sqrt{TS} > 5$ at $\vert b\vert < 10^{\circ}$ and $\sqrt{TS} > 4$elsewhere) and found 188 and 208 significant sources for the Ring and Galprop models, respectively. We manually checked the TS maps of all sources that barely passed the detection threshold with the Ring model and for which $\sqrt{TS} \sim 3$ with the Galprop one.

We emphasize that the order and criteria applied to cross-correlate positions between the excesses detected in different energy bands and time periods can strongly affect the catalog list close to the detection threshold. Several strategies were tested before adopting the present one, but one must remember that a faint source can pass or drop below the threshold by slightly changing its position or that of its neighbours. Given the steep increase in source numbers with decreasing TS, we emphasize that a small change in the TS threshold, alternatively in the background over which the source TS is calculated, produce a large change in the number of catalogue entries. For instance, lowering the $\sqrt{TS}$ threshold by 0.1 would add 27 sources.

5 Catalogue description

The EGR acronym was adopted for the EGret Revised source list presented in Tables 2 and A.1 in a format similar to the 3EG one. As explained above, the source characteristics (position and flux, and their uncertainties) were determined with the Ring model because of its higher flexibility, better fit, and flatter residual map. A secondary position and flux was measured with the Galprop model and is listed in Tables 2 and A.1 to illustrate the amplitude of the systematic uncertainties due to the choice of interstellar model.

Sources found within a radius of 1.5 PSF FWHM from a very bright source, and/or with very asymmetric TS map contours are not included in Tables 2 and A.1. We remark, however, that they represent significant excesses of photons above the background that may be due to extended sources, or structures not properly modelled in the interstellar emission, or artifacts due to incorrect PSF tails. This list of 14 confused sources is given in Table B.1, under the acronym EGRc for EGret Revised confused.

For both tables, the description for each column follows:

Num: source number in order of increasing right ascension.
Name: source name based on J2000 coordinates.
RA and Dec: J2000 equatorial coordinates in degrees.
l and b: Galactic coordinates in degrees.
$\theta_{95}$: angular radius, in degrees, of a circular cone that contains the same solid angle as the 95% confidence contour.
F: flux in 10-8 photon cm-2 s-1 for E > 100 MeV and for each time period.
$\sigma_F$: $1 \sigma$ statistical flux uncertainty in 10-8 photon cm-2 s-1.
Cnts: number of photons detected with E > 100 MeV.
$\sqrt{TS}$: statistical significance of the detection.
vp: short viewing period as defined in Table 1 or summed cycles noted px for cycle x, pijkl for the sum of cycles i, j, k, and l, and p19 for the total of 9 cycles.
$l_{\rm sys}$ and $b_{\rm sys}$ : Galactic longitude and latitude obtained with the Galprop background model.
$F_{\rm sys}$: flux obtained with the Galprop background model, in 10-8 photon cm-2 s-1.
3EG: third EGRET catalog counterpart source name if one exists within a radius of 1 PSF FWHM ($2^{\circ}$ for E > 100 MeV) from the EGR source and if the nearest neighbour relation between the EGR and 3EG sources is univocal (the nearest neighbour of the EGR source is the 3EG one and vice versa).

6 Comparison with the 3EG catalogue

\par\includegraphics[width=9cm,clip]{9685fig4.eps}\end{figure} Figure 4: Spatial distribution, in Galactic coordinates, of the EGR sources. The confused sources are marked as open circles.
Open with DEXTER

The revised catalogue contains 174 sources plus 14 confused sources compared to the 265 entries of the 3EG catalogue (excluding the Vela artifacts). Their spatial distribution across the sky looks different from that of the 3EG sources, as illustrated in Figs. 4 and 5. The accumulation of faint 3EG sources within $30^{\circ}$ of the Galactic centre is strongly reduced in the new results and fewer sources are seen below $30^{\circ}$ in general. These changes at low and mid latitudes are primarily due to the increase in background intensity from new HI, CO, and dark gas structures. At high latitude, the use of more $\gamma$-ray observations and of a revised large-scale IC component in the background may also explain why a handful of 3EG sources have fallen below the detection threshold whereas new sources are now detected.

The names of the 107 unconfirmed 3EG sources are listed in Table 3 and they are displayed in Fig. 6. They comprise only six sources firmly identified as AGN by Hartman et al. (1999), but flagged as extended or confused sources by the EGRET team. In fact, the proportion of these extended or confused cases among the unconfirmed 3EG sources is overwhelming (95%) and significantly larger than among the confirmed ones. The unconfirmed and confirmed 3EG groups show 69% and 33% of possibly extended 'em' sources respectively. Figure 6 also shows that the vast majority of unconfirmed 3EG sources were unidentified and spatially correlated with the Gould Belt system of nearby clouds. They follow the characteristic trace of the inclined Belt across the sky, gathering at $\vert b\vert < 30^{\circ}$, more at positive latitudes toward the Galactic centre, and below the plane at the anticentre. The EGR source sky distribution in Fig. 4 does not exhibit the Gould Belt signature anymore.

\par\includegraphics[width=9cm,clip]{9685fig5.eps}\end{figure} Figure 5: Spatial distribution, in Galactic coordinates, of the 3EG sources.
Open with DEXTER

\par\includegraphics[width=9cm,clip]{9685fig6.eps}\end{figure} Figure 6: Spatial distribution, in Galactic coordinates, of the 3EG sources with no counterpart in EGR: the unidentified sources as circles and the identified AGN as stars. The filled circles and stars mark the sources that were flagged as extended or confused in the 3EG catalogue.
Open with DEXTER

\par\includegraphics[width=9cm,clip]{9685fig7.eps}\end{figure} Figure 7: Spatial distribution, in Galactic coordinates, of the new EGR sources with no 3EG counterpart. The confused sources are marked as open circles.
Open with DEXTER

Table 3: Names of the 3EG sources with no EGR counterpart.

Table 4: Names of the new EGR sources with no 3EG counterpart.

\includegraphics[width=9cm,clip]{9685fi15.ps}\end{figure} Figure 8: Second stage of the iterative source detection around Geminga (see Fig. 3) obtained using the 3EG model ( left) and map of the Ring model intensity divided by the 3EG one ( right). The excess in the TS map assigned in 3EG to the 3EG J0556+0409 point source corresponds to a local underestimation of the diffuse emission in the 3EG model. Maps are given in 0.5$^{\circ }$ bins and galactic coordinates.
Open with DEXTER

The fact that many 3EG sources are unconfirmed by the present analyses should not cast doubts on the detection method from a statistical point of view. They did correspond to significant photon excesses above the background in the 3EG analyses, but, in the absence of some structures in the predicted interstellar background, an ensemble of point sources with the wide EGRET PSF would compensate for the missing clouds and yield an excellent fit to the data. Figure 8 illustrates this fact for the unidentified source 3EG J0556+0409 detected at 7.2$ \sigma$ in 3EG. The left side shows the TS-map corresponding to the second stage of the iterative source detection around Geminga above 100 MeV. It is the same as in Fig. 3 but we have used here the 3EG diffuse emission model instead of the Ring one. The same sources are detected apart from 3EG J0556+0409, which is not seen in Fig. 3. Instead an excess of diffuse emission appears in the ratio of the Ring to 3EG background intensities (Fig. 8, right). The photons attributed to a point source in 3EG in fact originate in a gas cloud within the Galaxy. This is probably still the case in the present analysis, although to a lesser degree, in particular at very low latitude where the optical thicknesses in HI and CO severely limit our knowledge of the true column densities. Other sources may also be due to increased cosmic-ray densities in specific clouds with respect to the local Galactic average. Over-irradiated clouds close to cosmic-ray sources would be detected as a single or cluster of point sources, depending on their angular scale.

\par\resizebox{7cm}{5cm}{\includegraphics{9685fig8.eps}}\end{figure} Figure 9: Histogram of the relative flux differences $\vert F_{\rm EGR} - F_{\rm 3EG}\vert / \sigma _{F_{\rm EGR}}$ measured between the EGR and 3EG counterparts in units of the statistical error on flux for each source. All fluxes are measured above 100 MeV.
Open with DEXTER

For the 81 EGR sources that do have a 3EG counterpart, we find reasonable agreement in position and flux from both analyses. On average, we find 3% lower fluxes in the EGR analysis with respect to the 3EG one because of the increase in Galactic background. Figure 9 shows the histogram of ratios of the EGR and 3EG flux difference over the statistical error in flux for each source: $\vert F_{\rm EGR}- F_{3{\rm EG}}\vert / \sigma_{F_{\rm EGR}}$. We considered the EGR flux for the observation with the highest $\sqrt{TS}$ and compared it with the 3EG counterpart flux for the same time period if available. Average P19 fluxes were compared to the 3EG P1234 average for non flaring sources. The flux differences are modest (17% rms dispersion) and in most cases smaller than the statistical uncertainties on flux estimates. Similarly, Fig. 10 indicates that the angular separations between EGR and 3EG counterparts are often consistent with the $\theta_{95}$ error radii. Thirty sources were, however, found as distant as $0.5^{\circ}$ from the 3EG position. This will have a large impact on counterpart searches and identification at other wavelengths.

On the other hand, we find 30 new EGR sources with no 3EG counterpart. Their names are listed in Table 4 and they are displayed in Fig. 7. Most are detected just above the threshold and 11 were indeed present in the 3EG complementary list, just below the significance threshold.

\par\resizebox{7cm}{5cm}{\includegraphics{9685fig9.eps}}\end{figure} Figure 10: Histogram of the relative angular separation between the positions found for the EGR and 3EG counterparts in units of the 95% confidence angle for each source.
Open with DEXTER

7 EGR source distributions and potential counterparts

Because of the new gas data that we used at intermediate latitude, the comparison between the EGR and 3EG source characteristics enables us to judge, to some extent, the impact of our limited knowledge of gas mass tracers. The comparison between the flux and positions obtained with the Ring and Galprop models enables us to estimate the systematic uncertainties due to our limited knowledge of the true cosmic-ray distribution across the Galaxy. Figures 11 and 12 indicate that, in most cases, the differences are smaller than the statistical uncertainties. The distribution of 95% confidence radii peaks between $\sim $ $0.2^{\circ}$ and $\sim $ $0.7^{\circ}$. The uncertainty in the background creates an additional systematic error of $\sim $ $0.2^{\circ}$ for most sources, which should be kept in mind while searching for counterpart sources.

We searched the EGR error circles for potential counterparts of interest such as pulsars from the ATNF catalogue (Manchester et al. 2005), blazar candidates from the ASDC list (Massaro et al. 2005) and the CGRaBS list (Healey et al. 2008), other flat radiosources from the CRATES compilation (Healey et al. 2007), supernova remnants from the Green catalogue (Green 2006), OB associations (Mel'Nik & Efremov 1995), and X-ray and TeV pulsar wind nebulae (Li et al. 2008 and Grenier 2008). The results are displayed in Fig. 13. We have found 13 radio pulsar associations in addition to the 6 objects firmly identified by EGRET. Thirteen EGR sources coincide with supernova remnants, 9 with pulsar wind nebulae, 7 with OB associations, 53 with blazar candidates, and 19 with other flat radiosources. These associations should not be considered as source identifications, but as spatial coincidences worthy of further investigation, in particular with the improved angular resolution of GLAST. Yet, they reveal that as many as 87 sources have no obvious counterpart among the well-known $\gamma$-ray emitters, despite the large number of pulsars (1775) and radiosources (11 000) that were cross-correlated with sources and spread across the entire sky and along the Galactic plane. The lack of blazar counterparts is all the more surprising because the spatial distribution of sources off the plane is reminiscent of an isotropic, therefore extragalactic, distribution. The latitude distribution, shown in Fig. 14, is consistent above $30^{\circ}$ with a sample drawn from a uniform population, according to the exposure map, as shown by the black curve. The distribution flattens at lower latitude because of the increased background that drastically limits the survey sensitivity. Studying the consistency with an extragalactic population at medium latitudes and the implication of the lack of flat radio sources is beyond the scope of this paper and will be addressed in a future work. The sharp peak below $3^{\circ}$ in latitude indicates young emitters. Their clustering in the inner Galaxy ( $l \leq 30^{\circ}$), toward the direction tangent to the Carina arm, and toward the Cygnus region outlines their close relationship to large molecular complexes and star forming regions at a distance of a few kpc.

\par\resizebox{7cm}{5cm}{ \includegraphics{9685fi10.eps}}\end{figure} Figure 11: Histogram of the relative flux differences $\vert F_{\rm EGR}-F_{\rm sys}\vert/ \sigma _{F_{\rm EGR}}$ measured with the Ring and Galprop models in units of the statistical error in flux for each source. All fluxes are measured above 100 MeV.
Open with DEXTER

8 Discussion on specific sources

There is considerable interest in the physical processes occurring in the Galactic centre region. The 3EG catalogue lists one source located toward the Galactic centre, 3EG J1744-3011. We find two point sources in this region, EGR J1740-2851 at $l = -0.55^{\circ}$, $b = 1.05^{\circ}$ and EGR J1747-2852 at $l = 0.21^{\circ}$, $b = -0.24^{\circ}$. Figure 15 displays the TS-map for photons of energies that are higher than 1 GeV above the 3EG and the Ring background models. The $\theta_{95}$ error radius around EGR J1740-2851 and EGR J1747-2852 formerly excludes the Galactic Centre, but source locations and fluxes in this direction should be interpreted with extreme caution since the high gas optical depth around the Galactic centre and the velocity pile-up toward the centre induce large uncertainties in the total gas column densities.

Coincidences with supernova remnants were noted (Sturner & Dermer 1995) and are confirmed in the present analysis (see Table 5), but several also host a pulsar wind nebula, as in CTA 1 and IC 443; we therefore require far higher resolution $\gamma$-ray images to identify the origin of the emission, especially in these crowded regions. EGRET detections are confirmed toward two TeV-emitting wind nebulae around PSR J1420-6048 (in Kookaburra, EGRJ1418-6040) and PSR J1826-1334 (EGRJ1825-1325). Another interesting candidate is the wind nebula of the 11 kyr old and very energetic pulsar PSR J2229+6114 toward EGRJ2227+6114.

We note, as shown in Fig. 16, the positional coincidence within 0.5$^{\circ }$ between the new EGR J0028+0457 source and the millisecond X-ray pulsar PSR J0030+0451. This 300 pc distant pulsar, discovered in 2000 (Somer 2000, D'Amico 2000), has an X-ray counterpart exhibiting a double peaked pulse profile as measured by ROSAT (Becker et al. 2000). Millisecond pulsars have low magnetic fields and produce relatively few electron-positron pairs; the electric field is not therefore screened and the spectral cutoff due to pair production attenuation occurs at high energy. They are good candidates for accelerating particles to high energies. Harding et al. (2005) predicted a $\gamma$-ray flux for PSR J0030+0451 well above that of the $\gamma$-ray millisecond pulsar PSR J0218+4232 for which a pulsed emission was marginally detected (Kuiper et al. 2000).

\resizebox{7cm}{5cm}{\includegraphics{9685fi11.eps}} \end{figure} Figure 12: Histogram of the relative angular separation between the positions found with the Ring and Galprop models in units of the 95% confidence angle for each source.
Open with DEXTER

\includegraphics[width=14cm,clip]{9685fi12.eps} \end{figure} Figure 13: The revised EGRET source catalog, shown in Galactic coordinates. The symbols indicate the counterpart types found in the error box: identified pulsars as black squares; other ATNF pulsars as open squares; LSI +61 303, LMC, and solar flare as black triangles; ASDC and CGRaBS blazar candidates as black diamonds; other flat-spectrum radiosources from CRATES as open diamonds; supernova remnants from the Green catalogue as stars; no counterpart as crosses.
Open with DEXTER

Four massive binaries were detected at TeV energies, namely PSR B1259-63 (Aharonian et al. 2005), LSI +61$^{\circ }$ 303 (Albert et al. 2006), LS 5039 (Aharonian et al. 2006), and Cyg X-1 (Albert et al. 2007), which illustrates the very efficient particle acceleration in compressed or shocked pulsar winds, as well as in microquasar jets. Inverse Compton scattering of the bright stellar radiation would dominate at GeV energies. We find no interesting EGRET counterpart to these high-energy objects, besides the LSI +61$^{\circ }$ 303 radiosource. The latter has long been associated with the COS-B source 2CG 135+01 and the EGRET source 2EG J0241+6119 (Kniffen et al. 1997), yet has moved out of the 3EG error box and the marginal $\gamma$-ray variability could not be associated with radio flux variations. In the present analysis, we find the radiosource very close to the centre of the EGR J0240+6112 source. On the other hand, we find no source toward the dust enshrouded microquasar candidate, AX J1639.0-4642, or the Be/X-ray binary, AO 0535+26, both proposed as 3EG counterparts (Combi et al. 2003; Romero et al. 2001).

Another noticeable new source is EGR J1642+3940 detected at 5.8$ \sigma$ rather close to 3C345. 3C345 is one of the most prominent flat spectrum ( $\alpha=-0.1$) radio-loud, superluminal sources and is therefore an excellent candidate to be a $\gamma$-ray blazar. EGRET surveyed this region 12 times, in particular during period 5190 when a flare was found. We analyzed this particular period using the Ring model since it had not been used in the overall detection search. Figure 17 shows the resulting TS contour, for photons of energies above 100 MeV, that is well centered on 3C345. The cross corresponds to the EGR position (period 5190), the plus sign to the position with maximum likelihood, and the black dots to both the position of 3C345 and a nearby AGN. A marginal detection was also achieved for period 3034 at a level of 2.1$ \sigma$. It should be noted, however, that the small photon excess above 500 MeV was attributed to a flare from Mrk 501 by Kataoka et al. (1999) because the centroid was closer to the famous TeV source, so the association of EGR J1642+3940 with 3C345 is unclear. GLAST should easily confirm or disprove the association.

Several radiogalaxies (Cen A, NGC 6251, J1737-15) and a Seyfert 1 galaxy (GRS 1734-292) have been proposed as possible counterparts to 3EG sources (Hartman et al. 1999; Di Cocco et al. 2004; Combi et al. 2003; Foschini et al. 2005). These objects triggered some interest because their identification would raise important questions about the origin of the $\gamma$ rays at large angles from the strongly beamed emission from the jet. We do not, however, confirm the spatial coincidence with EGR sources in the present work. All these galaxies are located well beyond the 95% confidence region of EGR sources.

\resizebox{6.8cm}{4.8cm}{\includegraphics{9685fi13.eps}} \end{figure} Figure 14: Latitude distribution of the EGR sources with young Galactic sources at $\vert b\vert <3^{\circ }$, nearly isotropically distributed sources far from the plane, as expected from the black curve, and a flattening at mid-latitude because of the rapid increase in the interstellar background flux.
Open with DEXTER

\resizebox{7.8cm}{5.8cm}{\includegraphics{9685fi14.eps}} \end{figure} Figure 15: TS-map obtained at energies above 1 GeV toward the Galactic centre above the 3EG a) and Ring b) interstellar model
Open with DEXTER

Table 5: Names of the sources and supernova remnants found in spatial coincidence.

\end{figure} Figure 16: Likelihood TS contours for energies above 100 MeV and periods incompassing PSR J0030+0451. The cross, the plus sign and the black dot respectively mark the EGR catalog position, the position with maximum likelihood and the pulsar location.
Open with DEXTER

\resizebox{7.2cm}{5.2cm}{\includegraphics{3c345.eps}} \end{figure} Figure 17: Likelihood TS contours (50%, 68%, 95% and 99% confidence) for energies above 100 MeV and period 5190. The cross is the EGR catalog position, the plus sign the position with maximum likelihood and the black dots mark the radio positions of 3C345, Mrk 501, and 4C+38.41.
Open with DEXTER

9 Conclusions

We have searched for point-like sources in the reprocessed EGRET data from cycle 1 to 9 using new interstellar background models based on the most recent HI, CO, and dark gas data, as well as two different assumptions for the cosmic-ray distribution (the GALPROP diffusion model or a radial emissivity gradient fitted to the diffuse EGRET data). We have used the 3EG tools, likelihood method, procedure, and significance threshold to detect sources, but have expanded the search to 3 different energy bands (above 100 MeV, 0.3-1 GeV, and above 1 GeV). The resulting number of detected sources has decreased by more than a third. Many unidentified sources, in particular among those spatially associated with the Gould Belt, are not confirmed as significant excesses. Their emission can be explained by the additional interstellar emission and its structure. Several interesting counterparts to 3EG sources, such as radiogalaxies, massive binaries, and microquasars, are now found outside the 95% confidence region. We have cross-correlated the new source positions with large pulsar, supernova remant, pulsar wind nebulae, OB associations, and radiosource catalogues, yet half the sample has no attractive counterpart among the potential $\gamma$-ray emitters. Thirty new possible $\gamma$-ray sources have also been found.

This EGR catalog will be available in FITS format at the Strasbourg astronomical Data Center (CDS) and in ASCII format at http://www.aim.univ-paris7.fr/EGRET_catalogue/home.html

We are deeply grateful to Bob Hartman for his helpful explanations about the construction of the 3EG catalogue, and to Seth Digel and Andy Strong for their help with the gas and Galprop maps.



Table A.1: The EGR catalogue.

Table B.1: The EGR confused sources catalogue

Copyright ESO 2008