The 22-Month Swift-BAT All-Sky Hard X-ray Survey

We present the catalog of sources detected in the first 22 months of data from the hard X-ray survey (14--195 keV) conducted with the BAT coded mask imager on the \swift satellite. The catalog contains 461 sources detected above the 4.8 sigma level with BAT. High angular resolution X-ray data for every source from Swift XRT or archival data have allowed associations to be made with known counterparts in other wavelength bands for over 97% of the detections, including the discovery of ~30 galaxies previously unknown as AGN and several new Galactic sources. A total of 266 of the sources are associated with Seyfert galaxies (median redshift z ~ 0.03) or blazars, with the majority of the remaining sources associated with X-ray binaries in our Galaxy. This ongoing survey is the first uniform all sky hard X-ray survey since HEAO-1 in 1977. Since the publication of the 9-month BAT survey we have increased the number of energy channels from 4 to 8 and have substantially increased the number of sources with accurate average spectra. The BAT 22-month catalog is the product of the most sensitive all-sky survey in the hard X-ray band, with a detection sensitivity (4.8 sigma) of 2.2e-11 erg/cm2/s (1 mCrab) over most of the sky in the 14--195 keV band.


INTRODUCTION
Surveys of the whole sky which are complete to a welldefined threshold not only provide a basis for statistical population studies but are also a vehicle for the discovery of new phenomena. Compared with lower X-ray energies, where various missions from Uhuru (Forman et al. 1978) to ROSAT have systematically surveyed the sky and where slew surveys of later missions have added detail, our knowledge of the sky at hard X-rays (> 10 keV) has been rather patchy and insensitive. The sensitivity of the HEAO-A4 13-180 keV survey (Levine et al. 1984) was such that only 77 sources were detected.
Recently INTEGRAL-IBIS has provided some observations (Bird et al. 2007;Beckmann et al. 2006Beckmann et al. , 2009Krivonos et al. 2007) that are much more sensitive but have concentrated on certain regions of the sky; the exposure in the latest IBIS 'all-sky' catalog varies from one part of the sky to another by a factor of a thousand, some regions of the sky having only a few thousand seconds of observation. The RXTE all-sky slew survey (Revnivtsev et al. 2004) covers much of the sky in the 3-20 keV band and detects 294 sources, but the coverage is not uniform or complete and the sensitivity is weighted to lower energies such that the BAT and RXTE sources are not the same.
A survey in the hard X-ray band is important for several reasons. Observations below 15 keV can be drastically affected by photoelectric absorption in certain sources, giving a false indication of their luminosity. Populations of heavily absorbed or Compton-thick Active Galactic Nuclei (AGN) have been hypothesized in order to explain the portion of the spectrum of the diffuse hard X-ray background ascribed to unresolved sources (Gilli et al. 2007), but such objects have not been found in the necessary numbers, prompting questions as to the composition and evolution of a population of AGN that could explain its form (Treister et al. 2009). Hard X-ray emission is also being discovered from an unexpectedly large number of previously unknown Galactic sources, notably from certain cataclysmic variables, symbiotic stars and heavily obscured high mass X-ray binaries (Bird et al. 2007).
The Burst Alert Telescope (BAT) on Swift (Gehrels et al. 2004) has a large field of view and is pointed at a large number of different directions which are well distributed over the sky. The resultant survey provides the most uniform hard Xray survey to date and achieves a sensitivity sufficient to detect very large numbers of sources, both Galactic and extragalactic. Markwardt et al. (2005) have published the results from the first three months of BAT data, and Tueller et al. (2008) have published a survey of sources seen in the first 9 months of Swift observations, concentrating on the 103 AGN seen at Galactic latitudes greater than 15 • . We present here a catalog of all sources detected in the first 22 months of operations, Dec 15 -2006 increasing the number of AGN to 266 and including all other sources seen across the entire sky.
2. Swift-BAT Swift is primarily a mission for the study of gamma-ray bursts. Swift combines a wide field instrument, BAT, to detect and locate gamma-ray bursts (GRBs) with two narrow field instruments to study the afterglows (the X-ray Telescope (XRT) (Burrows et al. 2005) and the Ultra-Violet/Optical Telescope (UVOT) (Roming et al. 2005)). Swift-BAT is a wide field (∼2 sr) coded aperture instrument with the largest CdZnTe detector array ever fabricated (5243 cm 2 consisting of 32,768 4mm detectors on a 4.2mm pitch) . BAT uses a mask constructed of 52,000 5x5x1 mm lead tiles distributed in a half-filled random pattern and mounted in a plane 1m above the detector array.
This configuration results in a large field of view and a point-spread-function (PSF) that varies between 22 in the center of the field of view (FOV) and ∼14 in the corners of the FOV (50 • off axis). When many snapshot images (a snapshot is the image constructed from a single survey observation of ∼ 5minutes) are mosaicked together the effective PSF is ∼19.5 . Point sources are found using a fast Fourier Transform convolution of the mask pattern with the array of detector rates; this effectively uses the shadow of the mask cast by a source onto the detector array to create a sky image.
Over much of the BAT field of view, the mask shadow does not cover the whole array. The partial coding fraction is defined as the fraction of the array that is used to make the image in a particular direction and varies across the FOV. The BAT field of view is 0.34, 1.18 and 2.29 sr for areas on the sky with greater than 95%, 50% and 5% partial coding fractions.
Swift is in low Earth orbit, but because it can slew rapidly it can avoid looking at the Earth. The narrow field instruments cannot be pointed within 45 • of the Sun, within 30 • of the Earth limb, or within 20 • of the Moon.
The pointing plan for Swift is optimized to observe GRBs. This strategy produces observations spread out over a few days and at nearly random positions in the sky. The BAT FOV is so large that most of the sky is accessible to BAT on any given day, but the pointing is deliberately biased toward the anti-Sun direction in order to facilitate ground based optical follow-up observations of gamma-ray bursts. Even though the Swift pointing plan is optimized for GRB observations, BAT's large FOV and Swift's random observing strategy result in very good sky coverage (50-80% at >20 mCrab in one day) for transients (Krimm et al. (2006) 18 ). Over a longer term, this observing strategy produces an even more uniform sky coverage (see Figure 1), with an enhanced exposure at the ecliptic poles caused by avoiding the Sun and Moon. This high coverage factor means that the BAT survey can provide reasonably well sampled light curves and average fluxes compiled from data taken throughout the period covered by the survey.
The effective exposure time of each point in the field of view is the equivalent on-axis time (partial coding fraction times observing time); therefore the observing time for a place on the sky is generally much larger than is displayed in the effective exposure map. All of the source count rates from the BAT survey are normalized to this effective exposure.
The observing efficiency of Swift-BAT is high for a satellite in low Earth orbit, but observing inefficiencies (passages through the South Atlantic Anomaly (SAA) 16%, slewing 16%, down time < 1%) still result in a loss of 33% of the total observing time.
The BAT 22-month survey includes data taken between 15 Dec 2004 and 27 Oct 2006; there are 39.6 Ms of usable data in the 22 month survey. A typical point on the sky is within the BAT FWHM FOV ∼ 10% of the time, and the survey data screening rejects 0.5% of the data (see §3.1). These two effects result in an effective exposure of 3.9 Ms for a typical point on the sky. The histogram of the effective exposure (Figure 2) shows that most of the sky has an effective exposure time between 1.7 and 4.3 Ms, with a few regions receiving as much as 5.6 Ms.
The BAT PSF is determined by the mask tile cell and detector cell sizes. For an on-axis source, the PSF is approximately Gaussian with a FWHM of 22.5 arcminutes. In the native Cartesian tangent plane coordinate system of BAT images, the PSF has nearly a constant shape and size throughout the field of view. However, because tangent plane units are not spaced at equal celestial angles, the true PSF shape is compressed for off-axis sources, varying approximately as σ PSF = 22.5 /(1 + tan 2 θ) where θ is the angle of the source from the pointing axis. When averaged over many pointings, and weighted by partial coding and solid angle, the mean PSF is ∼19.5 arcminutes FWHM. We use this 19.5 arcminute PSF when analyzing the BAT survey mosaicked skymaps since they are composed of many contributing snapshot observations.
3. BAT SURVEY PROCESSING The following sections describe the BAT survey analysis techniques as implemented in the batsurvey software tool. General information on coded mask imaging can be found in Skinner (1995Skinner ( , 2008, Fenimore & Cannon (1981), and Caroli et al. (1986).

BAT Survey Data Collection and Initial Filtering
The BAT instrument monitors the sky in "survey" mode when not within a few minutes of responding to a gammaray burst. In this mode detected events are binned into histograms by the instrument flight software and the histogram counts are periodically telemetered to the ground (typically on a 5 minute interval). These histograms contain detector (spatial) and pulse height (energy) information. On the ground the histograms are further adjusted to place all detectors on the same energy scale, and then for the standard survey analysis are re-binned into the eight survey energy bands: 14-20 keV, 20-24 keV, 24-35 keV, 35-50 keV, 50-75 keV, 75-100 keV, 100-150 keV, and 150-195 keV. . The bottom panel shows the measured 5σ sensitivity across the sky in units of mCrab in the 14-195 keV band. The bright spots at l=344.1, b=-44.0 (GRB060614), l=271.8, b=-27.2 (GRB060729), and l=254.7, b=-1.4 (GRB060428a) are areas of high systematic noise due to very long exposures (> 800 ks) performed early in the mission before dithering in roll angle was instituted.
Several quality filters are applied to the BAT survey data. First, the spacecraft must be in stable pointing mode, which means that the attitude control "10 arcmin settled" flag must be set. The spacecraft star tracker must be reporting "OK" status, and the boresight direction must be at least 30 • above the Earth's limb. Second, BAT must be producing good quality data, which means that the overall array event rate must not be too high or low (3000 cts s −1 < rate < 12,000 cts s −1 ); a count rate lower than 3000 cts s −1 means that the detector is not operating correctly, and a rate higher than 12,000 cts s −1 only occurs during passages through the SAA. A minimum number of detectors must be enabled (> 18,000 detectors out of 32,768), and no histogram bins can be reported as missing data because of bad telemetry. In addition, histogram time intervals that cross the UTC midnight boundary are discarded since the spacecraft has at times been commanded to make small maneuvers during that time. These temporal filters produce a set of good time intervals over which the histograms are summed. The finest time sampling of this survey analysis is approximately a single pointed snapshot (which have durations of ∼150-2000 s). The good time intervals are further checked so that the spacecraft pointing does not change appreciably during the interval (1.5 arcmin in pointing, 5 arcmin in roll), and data are excluded if the pointing has varied. Short intervals of 150 seconds or less are discarded in order to ensure enough counts across the detector for the balancing stage (see §3.2) of the processing to work correctly.
After temporal filtering each pointed snapshot is reduced to a set of eight detector count maps, one for each energy band. Since the systematic noise in the sky images depends on the quality of individual detectors, significant effort is made in optimizing the spatial filtering of the data (i.e, the masking of undesirable detectors). All detectors disabled by the BAT flight software are masked. In addition the detector counts maps are searched for noisy ("hot") detectors using the bathotpix algorithm; any detectors found to be noisy are masked. Finally, detectors with known noisy properties (i.e. high variance compared to Poisson statistics) are discarded. The "fixed pattern" noise (see §3.3) is also subtracted from each map.

Removal of Bright Sources
Bright point sources and the diffuse sky background contribute systematic pattern noise to the entire sky image, at approximately 1% of the source amplitude, due to the coded mask deconvolution technique. By subtracting the contributions of these sources from the detector images, the systematic noise can be significantly reduced. We used the batclean algorithm to remove bright sources and diffuse background at the snapshot level. The diffuse background is represented as a smooth polynomial in detector coordinates. Bright sources are represented by the point source response in the detector plane. The source responses are generated using ray tracing to determine the shadow patterns. Bright sources are identified by making a trial sky map, and any point source detected above 9σ in any energy band is marked for cleaning. In our experience-and based on the properties of the BAT mask-no new bright sources become detectable after the batclean calculation, so it is not necessary to iterate the process again. In order to preserve the original bright source intensities, we insert the fluxes from the uncleaned maps into the cleaned maps around the locations of these sources.
At the batclean stage the maps are also "balanced" so that systematic count rate offsets between large scale spatial regions on the detector are removed. Sources shining through the mask do not produce this kind of coherent structure, therefore this balancing stage helps to remove systematic noise. This process involves dividing the array into detector module sides (128 detectors, [= 16 × 8]), which are separated by gaps of 8.4-12.6 mm from neighboring detector module sides. The mean counts in both the outer edge detectors (44 detectors), and the inner detectors (84 detectors), are subtracted for each module separately, so that the mean rate is as close as possible to zero. Count rate variations from module to module are believed to occur because of variations in the quality of CZT detector material and because of dead time variations in the module electronics caused by noisy pixels. Variations between outer edge and inner detectors in each module are due to cosmic ray scattering and X-ray illumination of detector sides. The BAT coded mask modulates the count rate of cosmic sources on essentially detector-to-detector spatial scales, so the subtraction of the mean count rates averaged over many tens of detectors does not affect the coded signal.
Very bright sources which are partially coded will cast shadows of the mask support structures on the edges of the mask. These shadows are not coded by the mask, are highly energy dependent, and thus must be treated carefully. This is done by masking detectors in regions of the detector plane affected by mask-edge regions for bright sources (∼0.3 Crab or brighter) determined via ray tracing.
After subtracting bright sources and background, detectors whose counts are more than 4σ from the mean are discarded in order to further remove contributions from noisy detectors.

Fixed Pattern Noise
Non-uniform detector properties cause variations between the background count rates measured in different detectors. These spatial differences form a relatively stable pattern over timescales much longer than a day and are not addressed by the batclean algorithm. These spatial differences also comprise a fixed noise in detector coordinates which is transformed by the survey processing into unstructured noise in the sky image. This fixed pattern is determined by constructing long term averages of the residual BAT count rates of each individual detector, after subtracting the contributions of bright sources as described above. In this construction, variable terms average to zero and only the stable pattern remains. This pattern maps are then subtracted from each snapshot detector image. The benefit of removing this fixed pattern noise is that each individual detector is addressed, thereby removing systematic noise on a finer scale than the balancing stage mentioned previously.
The contributions of some detectors to this fixed pattern is time dependent as a result of temporal variations in detector performance. We address this time dependence in each detector by fitting a polynomial to the daily average value. The fits are done on data spanning weeks to many months, and the polynomial used has ∼ 1 order per 30 days fit.
This approach (subtracting the long-term average fixed pattern noise from the data) avoids removing any legitimate signal from sources since Swift changes its pointing direction on much shorter time scales.
In practice, the entire survey processing must be run once initially for all of the data, with the pattern contribution set to zero, in order to determine the residual rates mentioned above. Once the pattern maps have been computed, the processing is run a second time using those values.

Sky Maps
Sky maps are produced for each snapshot using the batfftimage algorithm which cross-correlates the detector count maps with the mask aperture pattern. Sky maps are sampled at 8.6 arcmin on-axis, which corresponds to half the natural element spacing for the coded mask. The natural sky projection for these maps is tangent-plane; thus, the sky-projected grid spacing becomes finer by a factor of ∼2 at the extreme edges of the field of view. The angular extent of a sky map from one snapshot covers the region in the sky where the BAT has some non-zero response. This field of view is approximately 120 • × 60 • , although the sensitivity is much reduced at the edges of the field of view due to projection effects through the mask (foreshortening of the mask and shadowing due to the mask thickness at large off-axis angles) and partial coding.
The snapshot maps are corrected for partial coding, geometric projection effects, and the number of active detectors. Thus, they represent the BAT count rate per fully illuminated detector, corrected approximately to the on-axis response. An examination of the measured count rates of the Crab nebula (considered to be a stable point source for the BAT) shows some systematic residual trends as a function of off-axis angle and energy. These effects are primarily due to absorption by passive materials in the field of view, whose absorption lengths scale approximately as sec θ, where θ is the off-axis angle. The absorptions can be as high as 50% at the lowest energies and largest angles, but are typically smaller. After correction for these effects, the count rate estimates are accurate to within a few percent.
Partial coding and noise maps which represent the partial exposure of each pixel in the sky map are created for each pointed snapshot. The partial coding maps are further adjusted to correct for the fact that some parts of the sky are occulted by the Earth during the observation. For each observation, a map of the average Earth occultation is computed showing the fraction of the observation time each pixel is occulted, and the partial coding map is multiplied by this occultation map to account for the reduced effective observing time. The noise maps are generated by computing the local r.m.s. of the pixel values in an annulus around each position (see §3.6).
Individual pointed snapshot sky maps are discarded if the differences between the model used in the bright source removal and the cleaned, binned detector plane data lead to a reduced chi-square value greater than 1.25. This filtering excluded primarily data around the bright X-ray source Sco X-1, which produces such strong count rate modulations that they aren't reduced to zero at the Poisson statistical level by batclean. This does produce a significant exposure deficit around Sco X-1 and the Galactic center region.
3.5. Mosaicking The sky images from each snapshot are weighted by inverse variance (i.e. noise −2 ) and combined into all-sky maps. Each snapshot sky map contributing to the mosaic is trimmed such that all areas of the snapshot have greater than 15% partial coding. The sky is divided into six facets in Galactic coordinates, with grid spacing of the pixels 5 arcmin at the center of each facet. The Zenithal Equal Area projection was used in order to minimize distortion far from the center of projection. Each individual sky image is projected and resampled onto the all-sky grids by bilinear interpolation, as are the partial coding and noise maps. The final result is a set of weighted flux maps, propagated noise maps and effective exposure maps for each energy band and facet combination, plus an additional one for the total energy band of 14-195 keV.
This analysis procedure produces a sky image where each pixel represents the best estimate of the flux for a point source at the corresponding position in the sky (see Fenimore & Cannon (1981) for more information on coded mask image reconstruction).
3.6. Source Detection A "blind" source detection algorithm was used to search for sources in the mosaicked significance maps using the full survey bandpass of 14-195 keV. The significance map is the ratio of the counts map to the local noise map.
The RMS noise map is calculated from the mosaicked sky map using an annulus of radius 30 pixels (2.5 • ) with an inner exclusion radius of 8 pixels (40 ). An 8 pixel radius around the position of all known BAT sources is also excluded from the regions used for background calculation. We do not attempt to fit the PSF of the source for the noise calculation and hence the positions of sources must be eliminated from this calculation to get an accurate measure of the underlying noise in the image.
The noise is assumed to be a smooth function of image position and so the value at the center of the annulus is well approximated by the average value in the annulus. This calculated noise includes both statistical and systematic noise and is therefore a better estimate of the total noise in the image than the noise calculated from a PSF fit. The noise from every source is distributed over the whole image, just as the signal from the source is distributed over the detector array, so no local enhancement of noise at the position of the source is expected.
The blind search algorithm first finds all peaks in the map by searching for pixels that are higher than each of the surrounding 8 pixels. If the significance in a peak pixel is greater than our detection threshold of 4.8σ (see §4.4), the excess is considered to be a detection in the blind search for new sources.  CVs / Stars 36 7 Pulsars / SNR 15 8 X-ray Binaries 121 a Sources listed as unidentified have an object with unknown physical type as a counterpart. Some of these objects are associated with a source detected at another wavelength. b Sources classified as galactic are so assigned because of observed transient behavior in the X-ray band along with insufficient evidence to place them in another class. c Sources in the extragalactic class are seen as extended in optical or near-IR imagery, but do not have firm evidence (such as an optical spectrum) from other wavebands confirming whether they harbor an AGN. d Sources classified as "beamed AGN" include blazars, BL Lacs, FSRQs, quasars, and other high redshift AGN.
The 22-month catalog and associated data in electronic form can be found online at the Swift website. 19 Figure 3 shows the distribution of sources on the sky color coded by source type, with the symbol size proportional to the source flux in the 14-195 keV band. Table 1 gives the distribution of objects according to their source type. Sources classified as "unidentified" are those where the physical type of the underlying object (e.g., AGN, CV, XRB, etc) is unknown. These sources have a primary name derived from the BAT position. Some unidentified BAT sources are associated with sources in the X-ray or gamma-ray bands (with positions unable to sufficiently determine an optical counterpart or physical source type), and these sources can be distinguished by having a name in the catalog derived from the observation in the other waveband. The few sources classified only as "Galactic" generally lie in the plane and have shown some transient behavior which indicates a Galactic source, but no other information is available that would allow further classification. "Extragalactic" sources are detected as extended sources in optical or near-IR imaging, but do not have other indications of being an AGN. The "Beamed AGN" category includes BL Lacs, blazars, and FSRQs. Table 5 is the listing of all the sources detected above the 4.8σ level in a blind search of the 22-month Swift-BAT survey maps. The first column is the source number in the 22-month catalog. The second column of the table is the BAT name, constructed from the BAT source position given in columns three and four. In cases where the source has been previously published with a BAT name corresponding to a slightly different location (e.g., a source position from a previous BAT catalog with less data), we have used the first published name but have given the correct 22-month BAT coordinates in columns two and three. The fifth column is the significance of the blind BAT source detection in sigma units. Instances where more than one possible counterpart to a single BAT source is likely are indicated with ditto marks in columns 2-5.
The sixth column gives the name of the identified counterpart to the BAT hard X-ray source with the most precisely known position. These are often optical galaxies, or 2MASS sources, and are associated with a source detected in the medium-energy X-ray band (3-10 keV) in Chandra, XMM-Newton, or XRT images. Counterpart determination is discussed in §4.2. The seventh column gives an alternate name for the counterpart. We have preferred to list a well known name (e.g., Sco X-1) or a name from a hard X-ray instrument or high energy detection. The best available coordinates of the counterpart (J2000) are given in the table in columns 8 and 9.
The 10th and 11th columns give the 14-195 keV flux of the BAT source (in units of 10 −11 ergs sec −1 cm −2 ) and its 1σ error. The BAT flux for each counterpart is extracted from the hard X-ray map at the location of the counterpart given in columns 8 and 9. The flux determination method is described in §4.5.
The 12th column indicates whether there is source confusion: there is source confusion either if there is more than one possible XRT counterpart or if two likely hard X-ray sources lie close enough together to make a proper extraction of the flux not possible with the standard method. The treatment of confused sources is discussed in more detail in §4.3. We define two classes of source confusion: "confused" sources, and "confusing" sources. A source is "confusing" for the purposes of this column if a fit to the map indicates that the source contributes to the hard X-ray flux of a neighboring source. A "confused" source has received more than 2% of its flux from a neighboring source. A confused source is labeled with an "A" in this column, and a confusing source with a "B" (the case of a very bright source next to a weak one would result in the bright source labeled with a "B" and the weak source with an "A"). A source that is both confused and confusing (e.g., the case where there are two similar strength sources close to each other, such as when there are two possible XRT counterparts to a single BAT source) is labeled with an "AB".
When a source has an entry in column 12, a best estimate of the counterpart flux is listed in column 10 from a simultaneous fit of all the counterparts in the region to the BAT map. When the entry is "A" or "AB" in column 12 (indicating a confused source), the error on the flux is not well defined, and column 11 is left blank. (See §4.5).
The 13th and 14th columns list the source hard X-ray hardness ratio and its error computed as described in §4.7. The hardness ratio is defined here as the ratio of the count rate in the 35-150 keV band divided by the count rate in the 14-150 keV band.
The 15th and 16th columns give the redshift and BAT luminosity of the counterpart if it is associated with a galaxy or AGN. The source luminosity (with units log[ergs s −1 ] in the 14-195 keV band) is computed using the redshift and flux listed in the table and a cosmology where H 0 = 70 km s −1 Mpc −1 , Ω m = 0.30, and Ω Λ = 0.70.
The 17th column lists a source type with a short verbal description of the counterpart.

Source Positions and Uncertainties
The BAT position is determined by using the BAT public software tool batcelldetect to fit the peak in the map to the BAT PSF (a two dimensional Gaussian with a FWHM of 19.5 arcminutes). The batcelldetect program performs a least-square fit using the local rms noise to weight the pixels in the input map. These fit positions were used to generate the BAT positions in the catalog and the names of newly detected BAT sources.
The PSF fit using batcelldetect also reports a formal position uncertainty based on the least-square covariance ma- trix. However, because neighboring pixels in the coded mask images are inherently correlated, the formal uncertainty reported by this technique will not be representative of the true uncertainty. Therefore we choose to use the offset between the fit position and the counterpart position as an indicator of the BAT position error.
The batcelldetect program also has the option of fitting source locations using an input catalog of starting positions. We have used this capability to test the stability of the source positions found by batcelldetect by using an input catalog where all the starting positions have been offset by 8 arcminutes in a random direction from the source position found in the blind search. We have performed this test with several different offsets and find that the fit converges to within 1 arcminute of the counterpart position for BAT sources that are not confused. For a few sources, the fit sometimes converges onto a side peak instead of the primary peak, but this error is not repeated in additional tests starting from other randomized positions. This type of systematic error in the position determination does not occur in the blind search ( §3.6) since we use the maximum pixel to start the fit instead of a randomized spot 8 arcminutes from the blind position. Anomalous offsets in the source position are identified by examination of the image and refitting.
In order to judge the accuracy of the BAT positions, we plot in Figure 4 the angular separation between the BAT position and the counterpart position against the significance of the BAT source detection. The accuracy of the BAT position improves as the significance of the detection becomes stronger. There are 461 BAT sources in Table 5 with detection significances greater than 4.8σ; there are 479 possible counterparts, and of these 25 are located grater than 5 arcminutes from the BAT position. Therefore, there is only a ∼ 5% chance of a BAT-detected source (> 4.8σ) having a counter- FIG. 4.-The BAT position error as a function of the BAT detection significance. The angular separation between the counterpart position and the fitted BAT position is used to determine a measured position error for each source. This measured position error is plotted as a function of BAT detection significance. The dashed line in the plot shows the 96% error radius as a function of BAT source detection significance. Sources with large position errors are almost always low galactic latitude sources falling in regions of high source density and locally higher noise. part farther away than 5 arcminutes.
In Figure 4 we also plot a line showing our estimate of the BAT position error for a given source significance. This estimate for the error radius (in arcminutes) can be represented with the function BAT error radius = 30 where S/N is the BAT detection significance. This empirical function includes a systematic error of 0.25 arcmin deduced from the position errors of very significant sources. This error radius includes 96% of the sources that are greater than 5 • from the Galactic plane and 15 • from the Galactic center. The error radius encloses 85% of sources in the Galactic plane. Sources known to be confused are not included in the plot.

Counterparts
Counterparts to the BAT sources were primarily discovered by examining X-ray images taken with instruments with good angular resolution. Chandra resolution is sometimes required on the plane, otherwise XMM-Newton, Suzaku or ASCA images were examined. ROSAT images and source catalogs were of relatively low importance for counterpart identification because of ROSAT's lack of effective area in the hard X-ray band, because of the poor correlation between ROSAT flux and the BAT hard X-ray flux (see Tueller et al. (2008), Figure 7), and because of the high chance probability of finding a ROSAT source in the BAT error circle.
If no archival X-ray images existed for the location of a BAT source, we requested Swift-XRT followup observations of the field containing the BAT source. A 10 ks observation with XRT is deep enough to detect almost all BAT sources. BAT extragalactic sources are usually AGN contained in bright (J ∼ 13), nearby galaxies at redshift z < 0.1 and are easily identified in an XRT observation.
The X-ray counterpart to an unabsorbed BAT source is a very bright XRT source, which is easily detected with a 2 ks XRT observation. However, most of the new BAT sources are heavily absorbed in the X-ray band and were not detected by ROSAT. We have found empirically that XRT can detect essentially all of the BAT sources (including the absorbed ones) in a 10 ks observation.
We require consistency of the BAT and the X-ray spectrum (> 3 keV) when simultaneously fit with an absorbed power law allowing only a renormalization between BAT and XRT to account for variability. This consistency of the spectra is required for all sources not previously known to be hard X-ray emitters, except transients and sources known to have highly variable spectra where the BAT spectrum averaged over years cannot be directly compared to the XRT measurement from a single observation.
A small fraction of XRT follow-up observations in the 5-10 ks range detected multiple sources consistent with the BAT position. In these cases the counterpart to the BAT source was almost always identified by limiting the bandpass of the X-ray image to the higher energy 3-10 keV band. This bandpass filtering usually reduced the number of sources in the field to a single hard source.
In the few cases where two or more hard sources still remain after bandpass filtering, all are considered possible counterparts to the BAT source and listed in the catalog with a flag indicating that the counterpart identification suffers from source confusion. There are 18 more possible counterparts in Table 5 than there are blind BAT sources (461) because of the 15 cases where there are one or more possible counterparts to a single BAT source.
Because the counterpart identification requires an X-ray point source with a small error radius (∼ 4 arcsec), a positional coincidence with a known source or bright galaxy, and an X-ray spectrum consistent with the BAT flux, we believe that the counterpart misidentification rate is extremely small. All of the identified counterparts listed in Table 5 are hard X-ray sources.

Confused Sources
Sources are labeled as confused in our table when the highest pixel associated with the BAT source in the mosaicked maps (the "central pixel" value) has a significant contribution from adjacent sources. This includes the cases when two possible X-ray counterparts lie within a single BAT pixel and when two BAT sources are close enough that each contributes flux to the location of the adjacent source.
Using the positions of the X-ray counterparts as an input catalog, we calculated the fractional contribution of each BAT source to its neighbors. We used the significance measured in the blind search and the 19.5 arcminute FWHM Gaussian BAT PSF to calculate the intensity of each source at the position of its neighbors. The central pixel value for each BAT source was assumed to be the sum of the source plus all the contributions from its neighbors. This creates a set of linear equations that can be solved for the true significance of each source. We solved these equations with the constraint that sources were not allowed to have negative significance. This procedure was devised to determine cases where the BAT significance is altered because of the presence of a very strong nearby source.
If we found that the resulting fit S/N from the technique that accounted for contributions from neighbors was lower than the central pixel S/N from the blind search by 2%, we labeled it as confused.

Detection Significances and Limits
The detection significance for the BAT sources in the catalog is extracted from the mosaicked significance map at the BAT position (see §3.6). The significance is taken from the highest pixel value in the blind search. Figure 5 shows the distribution of individual pixel significances from the mosaicked map of the entire sky. As is usual for a coded mask imager, the noise distribution is a Gaussian function centered at zero significance, with a width of σ = 1 and a total integrated area equal to the number of pixels in the map. The large tail at positive significance is due to real astrophysical sources present in the map.
The distribution of the pixel significances in Figure 5 closely follows a Gaussian distribution for the negative significances. The positive side of the distribution also follows a Gaussian, but with the addition of pixels with enhanced significances because of the presence of real sources in the map.
Examination of the negative fluctuations provides a good measure of the underlying noise distribution. There is only 1 negative pixel in the entire map with a magnitude greater than 5σ. We therefore choose our detection limit to be 4.8σ. This detection limit is also the same as used in the 9-month version of the Swift-BAT catalog. While it is clear that there are several real sources with significances somewhat smaller FIG. 6.-The integral distribution of sky coverage versus sensitivity achieved in the survey. The 1 mCrab sensitivity limit (for 50% sky coverage) corresponds to a flux of 2.3 × 10 −11 ergs cm −2 s −1 in the 14-195 keV band. than 4.8σ, we choose this value in order to minimize false sources caused by random fluctuations. We expect random fluctuations to account for 1.54 sources at the 4.8σ level in our sky map of 1.99 × 10 6 independent pixels. Figure 6 shows the integral distribution of sky coverage versus sensitivity achieved in the survey. We achieve a sensitivity of better than 1 mCrab for half the sky, which corresponds to a flux of < 2.3 × 10 −11 ergs cm −2 s −1 in the 14-195 keV band.

Counterpart Fluxes
Fluxes of the counterparts to BAT sources were extracted from the mosaicked maps using the pixel containing the position of the identified counterpart. For sources where a counterpart is not known, we use the fitted BAT position to determine the flux.
We have chosen to normalize source fluxes in the 8 survey bands to the Crab because the systematic uncertainties in the survey averaged Crab spectrum are smaller than the uncertainties in the BAT survey response matrix. The source fluxes in each band were computed by comparing the source count rate to the measured rate of the Crab Nebula in each band: BAT source flux = BAT source count rate Crab count rate Crab flux, (2) where the Crab flux in each band is given by where a and b are the lower and upper BAT band edges and E the energy in keV. We take the Crab counts spectrum to be determined by fitting a power-law model to BAT on-axis calibration observations taken early in the Swift mission. These values are consistent with characterizations of the Crab spectrum using data from Integral/SPI , Integral/IBIS , HETE/FREGATE (Olive et al. 2003), SAX/PDS (Fiore et al. 1999), and GRIS (Bartlett 1994 (5) Sources with a spectral index very different from the Crab can have a small but significant residual systematic error in the fluxes determined with this method.
In order to gauge this error we generated counts spectra for different models in the eight survey bands using the BAT onaxis spectral response matrix. The Crab comparison flux determination method described above was used to obtain the model fluxes in each of the eight survey bands. The flux errors between the computed fluxes and the model fluxes in the individual bands were always < 10% for a range of model spectral indices between 1 and 3 and so we deem this technique acceptable for producing the source fluxes in each of the survey bands.
We fit the 8-channel spectra with a power law model in order to produce an overall hard X-ray flux for each BAT source. We used XSPEC and a diagonal matrix to fit the 8channel spectra with the pegpwrlw model over the entire 14-195 keV BAT survey energy range in order to extract the source flux in this band. This approach was selected because it weights the energy bands by their individual uncertainties; a simple sum of the bands would produce a very large error due to the high weight it assigns to the noisiest bands at the highest energies.
The 1σ error in the overall flux was determined by using the error function in XSPEC and is given in Table 5. For the highest significance BAT sources (> 100 sigma), this procedure does not produce a good fit (reduced χ 2 1), but this is to be expected from the very high significances of each point and the coarse energy binning. To evaluate the systematic error in the fitting we performed fits to our model spectra generated from the response matrix. For power law spectra, the systematic error in the flux is dominated by the error in the individual data points as calculated above. Sources with hardness ratios less than 0.1 are not well fit with a power law, and the systematic uncertainty in the flux can be significantly larger.
The fluxes for sources marked as confused were calculated in a slightly different way. Instead of using the count rate extracted from the map at the counterpart position, we performed a simultaneous fit to find the fluxes of all the sources in the confused region as described in §4.3. For these sources we do not quote an error on the flux estimate because the behavior of the errors with this fitting technique is not well known. Any source with a confused flag should be considered as detected by BAT but the flux should be considered as an upper limit.

Sensitivity and Systematic Errors
In this section, we compare the expected statistical errors with the actual measured statistical noise in the final mosaic maps. From the perspective of pure Poisson counting statistics, the uncertainties are governed primarily by the properties of the coded mask and the background (see Skinner (2008) for details). The expected 5σ noise level can be expressed as (adapting from Skinner (2008) Eqn. 23 and 25): where b is the per-detector rate, including background and point sources in the field of view; N det is the number of active detectors (N det ≤ 32768); T is the effective on-axis exposure time; and α is a coefficient dependent on the mask pattern and detector pixel size (α = 0.733 for BAT). The partial coding, p, enters the expression through the "effective on-axis exposure" time, T = pT o , where T o is the actual exposure time. Using nominal values (b = 0.262 cts s −1 detector −1 ; N det = 23500 (the exposure-weighted mean number of enabled detectors); and Crab rate = 4.59 × 10 −2 cts s −1 detector −1 ), we find the estimated Poisson 5σ noise flux level to be We consider this to be a lower limit to the expected Poisson noise level for a given effective exposure. In reality, the background rate b may be higher than the nominal value by up to 50% depending on the particle environment of the spacecraft. Also, along the Galactic plane the contributions of bright sources such as the Crab, Sco X-1 and Cyg X-1, are not strictly negligible, and will raise the overall level of b by up to ∼10%. All of these adjustments would cause a Poisson noise level larger than given by equation 7, by an amount that depends on the specific satellite conditions during the survey. We estimate that, averaged over the entire survey duration, the true Poisson noise level may be 5-15% higher than the lower limit quoted above. Figure 7 compares the measured noise and expected noise versus effective on-axis exposure. We see that both noise measures are decreasing approximately as T −1/2 , which suggests that the dominant errors are uncorrelated over time. It also suggests that pointing strategies such as roll-angle dither-FIG. 7.-Measured 5σ BAT sensitivity limit for pixels in the all-sky map, as a function of effective exposure time, T , for the 3-month (red; Markwardt et al. (2005)), 9-month (green; Tueller et al. (2008)) and 22-month (blue; this work) survey analyses. The contours indicate the number of pixels with a given sensitivity and effective exposure. The contour levels are linearly spaced. The red dashed line represents the original T −1/2 sensitivity curve quoted in Markwardt et al. (2005). The black dashed line represents a lower limit to the expected Poisson noise level (see §4.6). The measured noise is approximately 30-45% higher than the expected Poisson noise.
ing have been successful in reducing pointing-related systematic errors. However, the measured noise is still higher than the expected Poisson noise by ∼30-45%, and we take this to be a measure of the unmodeled systematic variations on the detector plane.
The largest likely contributors to systematic variations are improper subtractions of diffuse background, and of bright sources. BAT count rates are background-dominated -the background rate is equivalent to ∼6-9 Crab units -so the coded mask analysis is particularly sensitive to imperfect subtraction of spatial background variations. While the pattern map method and the functions fitted during the cleaning stage produce a good model of the detector background, some imperfections remain. One effect is that detector-to-detector sensitivity differences, coupled with varying exposures to the Xray background, can lead to excess residuals.
For similar reasons, bright sources may also contribute systematic noise. The brightest sources are clustered along the Galactic plane, and thus contribute noise in those preferred locations. Indeed, we note that the measured noise is ∼50% higher in the Galactic center region, where there is a concentration of bright point sources. This is a larger factor than can be accounted for by a larger count rate. Modeling of point sources may be imperfect for the same reasons as for the background. Also, there may be other effects such as side illumination of detectors that may contribute additional noise. Here, "side illumination" refers to the facets of the individual CdZnTe detectors which do not face the pointing direction, but are still sensitive to X-rays. Off-axis sources will shine through the mask and illuminate the sides, producing an additional (although fainter) coded signal. At the moment, illumi-nation of the sides of detectors is not modeled at the imaging or cleaning steps, and thus there will be an additional noise due to the effect.
Proper modeling of these systematic error contributors will be the subject of future work. At this stage we do not have an in-depth analysis of the quantitative contributions of each effect to the systematic noise, and in some cases the analysis may be prohibitively difficult. Still, at the current exposure levels, the noise level seems to be decreasing with exposure, and we do not appear to be reaching an ultimate systematic limit in this analysis.

Spectral Analysis
In §4.5 we use a simple power-law fit to the data to estimate the source flux. However, because the catalog contains sources with various different physical natures and spectral shapes, we choose to use the more robust characterization given by a hardness ratio to describe the BAT spectra.
Hardness ratios for the Swift-BAT sources were calculated by taking the sum of the count rate in the 35-150 keV band and dividing by the count rate in the 14-150 keV band. Errors on the hardness ratio were calculated by propagating the errors on the count rates in the individual 8 bands except when the source is listed as confused. Figure 8 shows a map of the source positions on the sky, with the source flux represented by the size of the point and the source hardness by the color (red is softer and blue is harder). Figure 9 shows the hardness ratios of the 22-month BAT sources by source class.
A mapping can be made between hardness ratio and powerlaw index for sources that have spectral shapes well described by a simple power-law model (e.g., the majority of AGN in our catalog). This mapping is well represented by Γ = 3.73 − 4.52 HR, where Γ is the power-law index and HR is the hardness ratio as defined above. Figure 10 shows the correlation between power-law index and hardness ratio for the BAT survey sources. The correlation holds well for sources with hardnesses above about 0.15, but begins to break down for softer sources. An illustration of this problem is the soft BAT spectra of clusters of galaxies, which have thermal spectra with temperatures ∼ 10 keV and are usually detected only in the lowest energy BAT band and are not well fit by a power-law model. We leave to a later paper a more careful spectral analysis using models appropriate to the physical nature of the sources.
5. Swift-BAT SURVEY SOURCES Although it is for extragalactic astronomy that the present survey represents the greatest step forward, a comparison of the results for Galactic sources with earlier work also has interesting implications.
Of the 479 sources in Table 5, 97% have reasonably firm associations either with objects known in other wavebands or with previously known X-ray or gamma-ray sources. More than 60% of the associations are with extragalactic objects. At high Galactic latitudes (|b| > 10 • ) the density of identified extragalactic sources is 22.6 sr −1 , and it is only only slightly reduced at low latitudes to 19.2 sr −1 . This suggests that only ∼ 7 extragalactic sources in the plane are missed through reduced sensitivity, lack of information in other wavebands, or confusion, and illustrates the uniformity of the survey.
153 BAT AGN were previously reported in the BAT 9month AGN survey ). Winter et al. (2009) provide X-ray spectral fits for these sources and provide measures of the luminosity, n H , etc. for the sources in the 9-month catalog. Most of the sources in the 9-month catalog also appear here in the 22-month catalog; however, variability in the sources has caused 11 sources to drop out of the 22-month list that were in the 9-month catalog.

New sources
During the survey BAT has detected a number of new sources that are transients or other Galactic objects not previously reported as hard X-ray sources. Some of these have been reported in Astronomer's Telegrams or elsewhere, others appear for the first time in this compilation. For convenience these are summarized in Table 2.
In Table 3 we note other sources that are detected in this survey and where XRT follow-up has provided additional information, but where a unique optical, IR, or radio counterpart is still lacking, or where there is only a BAT detection. Some of these are almost certainly Galactic objects as may be judged from their proximity to the Galactic plane. Table 4 list the new AGN discovered in the 22-month Swift-BAT survey. Table 4 lists those objects discovered with BAT whose AGN nature could be confirmed with an optical spectrum. In column 3 of the table we list the source of the optical data, and columns 4 and 5 list our own typing of the spectrum and the redshift. The optical spectra are mostly obtained from data in the public domain such as SDSS or 6df, but in a few cases we have obtained data from our own observations taken at the 2.1m telescope on Kitt Peak.

Extragalactic Sources
Most of the extragalactic identifications are with relatively nearby Seyfert galaxies and many of the remainder are with beamed AGN (blazars, BL Lac, FSRQ, etc) sources at much higher z. Figure 11 shows some typical BAT source host galaxy images from the Palomar digital sky survey. The field of view is 2 arcmin across for each subimage. The figure was produced by dividing the hardness-luminosity plane into 70 bins and randomly choosing a BAT source from that category to display. Of note are the high fraction of spiral galaxy hosts (as opposed to ellipticals), and the high number of interacting galaxies.
The 234 sources that have an identification with a well established Seyfert galaxy more than double the number in any previous hard X-ray survey. The distribution of column densities, spectral indices, and luminosities for the survey sources will be presented in a separate paper. As elsewhere in this paper, we emphasize that this catalog is based on mean flux levels over the entire 22 month period. The detections of sources with significant temporal variability over the survey period and the implications of such variability will be discussed elsewhere (Skinner et al., in preparation). Figure 12 shows a histogram of the redshifts of all the AGN found in the 22-month catalog. The distribution of the Seyfert galaxy redshifts from the 22-month survey (left panel of Figure 12) is highly biased towards low redshifts (z ∼ 0.03) with a tail extending out to z ∼ 0.1 and a few more distant objects out to z ∼ 0.3. The right panel of Figure 12 shows the redshift distribution of the beamed AGN. This distribution is quite different from the Seyfert galaxies in the left panel, with redshifts that extend to z ∼ 4 and with no objects at z < 0.033. Since we have no selection biases with respect to these beamed AGN  NOTE. -These sources have new information but no firm identification with an optical/IR/Radio object. a Galactic coordinates are given as an indication of whether the sources is likely to be Galactic.   (as opposed to optical searches for blazars) these different redshift distributions are a fundamental property of these classes and are directly related to their luminosity functions and evolution ). Figure 13 is a histogram of the luminosities of the Seyfert galaxies detected in the BAT 22-month survey. The luminosity distribution of the Seyfert galaxies continues to show a difference between the type Is and type IIs as noted in . This indicates that the true luminosity distribution is indeed different for these two classes, which is inconsistent with the unified model of AGN. A K-S test of these two luminosity distributions shows that they are drawn from the same parent distribution with a probability of only 0.30.
As in earlier hard X-ray surveys, the second most common category of extragalactic sources are beamed AGN, which include types such as blazars, BL Lacs, FSRQs, etc. There are 32 objects in this category. The highest redshift is for QSO J0539−2839 at z = 3.104.
We have detected 10 clusters of galaxies at > 4.8σ; Perseus, Coma, Ophiuchus, Cygnus A, Abell 2319, Abell 754, Abell 3266, Abell 2142, Abell 3571, and Triangulum Australis. The BAT spectra of Cygnus A and Abell 2142 in the 14-100 keV band are dominated by the AGN component in or around the clusters. The other 8 clusters are all hot (kT ∼10 keV); their BAT spectra are consistent with an extension of the thermal emission modeled with ASCA/XMM-Newton/Chandra archival data in the 2-10 keV band and do not require any additional component for a good fit. In other words, there is no evidence of a non-thermal diffuse component in these clusters. We estimated the upper limit of the nonthermal emission by adding a power-law model to the spectral fit for the 10 detected clusters. The upper limit is ∼ 6×10 −12 ergs cm −2 s −1 on average and the lower limit for the magnetic field B ranges from ∼ 0. Some 20 sources are clearly identified, largely through follow-up Swift-XRT observations, with galaxies from which no sign of nuclear activity has yet been reported in other wavebands. Their mean luminosity is only slightly lower than that of those classed as Seyferts on the basis of optical spectra (10 43.53 compared with 10 43.75 erg s −1 ). These are probably low-z counterparts to the X-ray bright, optically normal (XBONG) sources discovered in the Chandra and XMM-Newton deep fields (Barger et al. 2005;Comastri et al. 2002).

X-ray Binaries
As can be seen from Table 5, approximately two-thirds of the Galactic sources are X-ray binaries. Of those whose nature is known, about 40% are high mass X-ray binaries, which reflects the BAT's sensitivity to hard-spectrum X-ray sources. 60% are low mass X-ray binaries, which typically have softer spectra but can have a higher total flux. The low mass Xray binary population is concentrated near the Galactic plane and bulge, whereas the high mass X-ray binary population is more distributed, including significant contributions from the Magellanic clouds.
The sensitivity of the survey is such that high luminosity sources (> L x ∼ 10 36 erg s −1 ) are detectable anywhere in the Galaxy and the catalog is complete for sources that emit continuously at this level. However, since many X-ray binaries are transient it is likely that there are a significant number of additional X-ray binaries that are not seen in the present analysis (which is based on fluxes averaged over 22 months), but that can be detected in specific shorter intervals. This will be the subject of further work. The detection of outbursts from transients that do not repeat, or that repeat only on timescales of several years, should scale approximately linearly with the length of the survey.

CVs and Other Accreting White Dwarf Systems
Accreting white dwarf binaries constitute the second most common category of Galactic sources. Of these, 31 have been identified as cataclysmic variables (CVs) or CV candidates. Of the 31 CVs detected with BAT, 14 are in the Barlow et al. (2006) (INTEGRAL) list, while 17 are new detections in hard X-ray surveys. Because the INTEGRAL catalog goes deeper near the Galactic plane but the BAT catalog is more sensitive at higher latitude, the lists of CVs in the two catalogs are complementary. With the expanded list, we confirm that the hard X-ray selected CVs are dominated by magnetic CVs of the intermediate polar (IP) subtype, also known as DQ Her type stars (see also Brunschweiger et al. (2009)).
In addition, 4 hard X-ray bright symbiotic stars have been detected by BAT, as summarized by Kennea et al. (2007), and there is now a candidate to be the fifth member of the class. Finally, BAT has also detected one Be star, gam Cas, for which an accreting white dwarf companion is one of three possibilities proposed as the origin of the X-ray emission (Kubo et al. 1998).

SNRs and Non-Accretion Powered Pulsars
We detect hard X-ray emission from 8 pulsars and/or their associated Pulsar Wind Nebula (PWN) or supernova remnants (SNRs). Our upper limit on PSR J1846−0258 is consistent with the flux level at which it was detected in a long INTEGRAL-IBIS observation and reported in Bird et al. (2007). In the case of HESS J1813−178 it appears that we are detecting emission that is from the point source seen at lower energies (Funk et al. (2007)) rather than directly related to the slightly extended VHE gamma-ray emission. The only SNR-related source that is not associated with a PWN is Cas A.

The Galactic Center
Because of the limited resolution of the BAT instrument the emission reported as from Sgr A* should be regarded as the net emission from a region of ∼ 6 radius centered on the Galactic center. It is possible that a number of sources contribute.
6. CONCLUSIONS The 22 month BAT catalog reinforces and enhances the results from the 3 month (Markwardt et al. 2005) and 9 month  catalogs and shows that the BAT survey continues to increase in sensitivity roughly as the square root of time and is far from being confusion or systematics limited. Future papers will discuss the X-ray spectral and timing properties of these sources as well as the nature of the optical identifications, in a manner similar to that of Winter et al. (2008Winter et al. ( , 2009. We have also obtained extensive optical imaging and spectroscopy of the AGN population (Koss, in preparation; Winter et al. (2009)), more detailed X-ray observations as well as Spitzer IR spectroscopy (Melendez et al. 2009) and radio data which will be presented in future publications. These results will allow the first large scale categorization of the AGN phenomenon from a large uniform and unbiased sample as well as a detailed comparison with results obtained with other selection techniques.