Table 1 (click here) summarises the characteristics of the three different image cubes made; Figure 1 (click here) depicts an overview of the sky area surveyed, together with the sensitivity contours thereof.
Survey | Image cube center | (epoch) | Size | Resolution | Channel width | RMS | Velocity range | ||
![]() | ![]() | ![]() | kHz | mJy |
![]() | ||||
aVLA | 17![]() ![]() ![]() ![]() ![]() ![]() | (B1950) | 37 | 2.2 | 1.14 | 6.1 | 7.2 | ![]() ![]() | |
bVLA | 17![]() ![]() ![]() ![]() ![]() ![]() | (B1950) | 17 | 0.5 | 2.27 | 12.2 | 3.7 |
![]() ![]() | |
cATCA | 17![]() ![]() ![]() ![]() ![]() ![]() | (J2000) | 37 | 2.2 | 1.45 | 7.8 | 4.5 | -600![]() | |
Figure 1: Composite survey area. The filled circle represents the
position of Sgr A*, as well as the center for both the high resolution
VLA and the ATCA image cubes. The triangle represents the image cube
center for the concatenated VLA monitor data. Long-dashed squares are
the 37 VLA and ATCA surveys; we filtered out the corners for
which the distance to the field (pointing) center was larger than
24
. The short-dashed square encloses the high resolution,
17
VLA image. Drawn in solid lines are contours of equal
detection probability; lines connecting points with equal sensitivity,
according to the most sensitive image cube, after correcting for the
primary beam attenuation (thus acting as a "primary beam response" for
this combined survey). Contours drawn are, from inside out, 1.005, 1.1,
1.25, 1.5, 2, 4, 8 and 16 times 25 mJy. Note that the VLA beam degrades
faster because of a larger dish size; the VLA has 25 meter, the ATCA has
22 meter dishes. Approximate J2000 corner coordinates are: upper right
17
44
05
-28
40
31
to
lower left 17
47
06
-29
19
02
The monitor data was in 1992 the most extended, uniform 1612 MHz spectral line data set of the GC known by the authors. We did not consider extending the concatenation with observations by others (Habing et al. 1983; LWHM), because those observations were taken in a period when the VLA was limited in its spectral line capabilities. The observations and data reduction of the monitor data used are described in detail in vLJGHW. We chose to start from their calibrated data sets, mainly because bad visibilities and interference from the Russian GLONASS satellite positioning system had already been removed.
In summary, the GByte of calibrated visibility data sets
consist of 20 epochs of two-hour VLA observations, taken in different
array configurations in the period from January 1988 to January 1991.
Unfortunately for the current project, the sky position is not exactly
centered on Sgr A*, and the velocity coverage is limited to only -110
to
. The data have been reduced, calibrated and
analysed using the NRAO AIPS reduction package (versions from 15JAN88,
up to 15JUL94), which we continued to use on the concatenated data set.
Being forced to use different AIPS versions over the years introduced
some problems with the data tables during the concatenation process. We
will comment on that below. We are confident that the final results are
not different from what could have been obtained if we had started with
the raw visibility data and performed the concatenation in one AIPS
version only.
The following steps were taken for each individual data set to ensure the homogeneity of the sets, before concatenating them to one visibility file for final processing.
Each of the calibrated monitor data sets was checked for consistency by
fully imaging a couple of known strong OH sources. Because of different
problems, three epochs had to be regarded lost for our project. We did
not process the raw data, as a few missing epochs would not make a
significant difference in the noise statistics or detection
probabilities. Also, due to inconsistent removal of interference in
several epochs, we did not use any of the visibility data which could
have been affected. Therefore, all data for which the baselines were
shorter than 3 k was disregarded. If necessary, the u,v,w
vectors were recalculated to have the coordinates (B1950) RA 17
42
12
600 and DEC -28
58
18
00 as
phase center.
The largest effect of using different AIPS versions could be seen in some tables containing additional information to the data. We mainly had to deal with the flagging tables as the table format had changed. It was impossible to restore the original flagging, so for each combination of baseline and time in a flagging table, all visibility information was marked as bad data. Whenever applying the task SPLIT, the sky frequency (different for each epoch) and velocity information in each header and antenna file got reset and had to be repaired by hand. As we will not use frequencies (but velocities instead) and the velocity consistency had already been checked with spectra and maps, we took an arbitrary file header and used it to set all the relevant values to match. The same was done for the frequency information in the antenna file. By doing this, one introduces an error less than 0.05%, in appearance comparable to radial bandwidth smearing, but in net effect negligible. At this time the data sets were also converted from circular polarisations to Stokes "I" to reduce the file sizes with a factor of two immediately.
As the absolute flux calibration of each data set had been done carefully by vLJGHW for the flux monitoring program, we have not performed any further bandpass or amplitude correction, visibility phase recalculation or flagging operation on the individual monitor data sets.
All data sets were added in a similar manner to create one concatenated visibility data set of about 35 observing hours. Each epoch was put in a different sub-array without rescaling the visibility weights. Due to different calibration paths, the individual weights of the visibility points differed per monitor data set. To get each epoch contribute equally, we treated all visibility weights to be unity. Although formally one has to account for the epochs that are taken in one polarisation only (two data sets before August 1988), we have not looked into this matter after converting each data set to Stokes "I".
To subtract confusing continuum emission, a baseline was fitted to each visibility spectrum, using three regions in the total spectral band which seemed to be void of line sources (Van Langevelde & Cotton 1990; Cornwell et al. 1992). Several channels of each region were averaged, and interpolated to represent the continuum emission. This visibility model for the continuum emission was subtracted from all channels of the concatenated VLA data set. On the fly, all visibility points exceeding twice the expected flux in the channel with the strongest source, were clipped. For epoch alignment we used a single self-cal iteration on the visibility phases. It corrects for relative systematic (atmospheric) effects in the different monitor data sets. We selected the red shifted peak of OH359.938-0.077; a single channel with only one strong, 6 Jy peak, and close to the field center.
From this calibrated visibility data set we made two naturally weighted
image cubes and a number of "clean boxes"; strong sources outside the
main image cube were mapped in small fields to limit their side-lobe
interference. The first image cube is a full primary beam, low spatial,
but full frequency resolution image cube (survey "a" in Table 1 (click here)). The
next is a small field, high spatial resolution image cube ("b") of the
same VLA data set, for which we averaged two channels. The latter was
centered on Sgr A* and the - not fully removed - extended continuum
emission of the Sgr A complex. The low resolution image was chosen to
match the spatial resolution of the ATCA. The large pixel size resulted
in having only the shorter baselines (about one half of the 550000
visibilities) contribute to the image, whereas for the high resolution
image cube, about 90% of the visibilities could be used. Using an even
higher resolution, to allow all visibilities to be used, would result in
huge maps or alternatively too many "clean boxes" (more than 15). It
would thus require an extra pass of subtraction of sources, without a
substantial decrease in noise level. The numbers used were a trade-off
between resolution, noise level, sky coverage, execution time and disk
usage. For reference, to map the entire region at full spatial and
spectral resolution would require at least 34 Gb of disk space (with our
choices less than 4), and because the execution time is strongly
dependent on the disk I/O speed, even with modern workstations it would
take months to execute (compared to a handful of days). Effectively,
only the baselines between 4 and 32.2 k and 4 and 143.8
k
were used for the low and high spatial resolution images,
respectively. Fitted RMS noise levels per channel were measured to be
7.2 and 3.7 mJy on average, respectively.
The ATCA was used on 8, 10, 11 and 13 July 1994. To reduce broadband interference by GLONASS and to filter out as much as possible of the extended continuum emission in the GC region, the longest baseline configuration was selected for all the observations. The GC was observed at a central frequency of 1612.0 MHz, covering 8 MHz bandwidth over 1024 frequency channels. The integration time was about 12 hours for each of the observing days. To circumvent GLONASS interference in the bandpass calibration, the primary flux calibrator B1934-638 was observed three times per observing run; typically every 6 hours. The phase calibrator B1748-253 was observed roughly every 40 minutes. Once per observing run we also observed the VLA primary flux calibrator B1328+307 to check the consistency of the flux density measurements between the ATCA and VLA. Actually, for July 10th, we used B1328+307 for bandpass calibration, because interference affected all B1934-638 observations of that day.
After removing bad visibility data points and applying the bandpass
calibration, the sky-frequencies were converted to LSR velocities.
Following the standard phase calibration, the extended continuum
emission from compact HII regions and the Sgr A complex was
subtracted with a simple two region baseline interpolation in the
visibility domain. Excessive amplitude visibilities were clipped before
applying one self-calibration iteration solution of the visibility
phases, on the same peak and in a similar manner as to the concatenated
VLA data set. Again we used "clean boxes" for removing side-lobes of
strong sources outside the main image cube. To avoid remaining effects
of interference by GLONASS and the subtracted continuum emission of the
Sgr A complex in the image cube, we only used baselines exceeding 4
k. Because of the large amount of disk space required (5 MB per
channel), full resolution images were made by the hundred for channels
69 to 964 (-600 to
). The RMS noise level is about
4.5 mJy on average.
Because we used self-cal to align different observing days after we had removed all continuum emission from the data, all positions of the OH/IR stars changed with respect to Sgr A*. Therefore, the channel with the maser line we used for self-cal, was mapped from the unsubtracted calibrated ATCA data set. In this map, the positions of Sgr A* and the line were measured, after which the maser positions with respect to Sgr A* could be determined. The same was done for one of the A-array VLA monitor data sets. The positional offsets with respect to Sgr A* in Table 2 link both VLA and ATCA observations together.
In order to search for discrete line sources, each of the three dimensional image cubes was projected into one two dimensional image in the following way: for each pixel in the sky plane, the maximum intensity over the whole frequency/velocity axis was stored in a new, two dimensional sky image. We shall refer to this image as the "maxmap". In that way one gets an overall view of the sky location of intensity maxima, although without directly knowing the velocity corresponding to the peaks. The "maxmap" resembles a continuum image (but recall that we already have filtered out the continuum emission), but with a much higher signal to noise ratio than if we would have averaged the pixel intensities over all channels. Alternatively, one can make a total intensity map (or "zeroth order map"); an image for which the flux densities that are higher than the threshold are integrated over the feature. However, as the total intensity map takes into account every pixel above a certain threshold - and thus indeed would be very useful when looking for double peaked features - it is also very sensitive to broad line sources anywhere in the spectrum. Because we have to deal with the remains of the extended continuum emission of the Sgr A complex, we observe a vast amount of broad line sources (see also Fig. 8 (click here)). These sources cause severe problems in searching the GC area; from non-linear spectral slopes to different noise and detection statistics. With this in mind, we prefer to use the "maxmap" instead of the total intensity map concept. The final results in both methods do not differ much, but the search is much more straightforward in the "maxmap".
Resulting were three "maxmaps" that all have the same noise level for
the detection of sources as the original image cubes, but now consisting
of pixels containing sources and the high-end part of the noise
distribution of the original map (i.e. only positive values around and
above the RMS noise level; comparable to a Rayleigh noise distribution
with a constant offset). Now, instead of searching the original image
cubes in each channel separately, and finding each source in several
channels, one obtains the same detections by searching the (one channel)
"maxmap" only. The "maxmaps" were searched for pixels with intensities
over 40 mJy (5.5), 25 mJy (6.7
) and 25 mJy (5.5
)
for the VLA 37
, VLA 17
and the ATCA 37
image
cubes, respectively. Pixels were then grouped in "islands" and for each
"island", the spectrum in the original image cube was taken at the
position with the highest pixel value.
The major advantage is that one would only get one spectrum for each
"island" (e.g. 7000 pixels make 200 islands, thus only 200 spectra),
risking that one could miss a second object that is spatially adjacent
to the object whose spectrum is drawn. Chances of missing a star are
higher close to stronger sources (having larger "islands", beam area)
and close to the remnants of improperly removed continuum features of
the Sgr A complex (non-linear spectral slopes). Therefore we checked
individual pixels of the local maxima in the confused area of the
central 5 by hand. We think we might have missed only a couple
of objects in our total analysis. The expectation value for two sources
to be positionally coincident is less than 1% over the data set, but on
the other hand we cannot correct for sources submerged in areas of
absorption, or the bowl of negative emission caused by the extended
background. The spectra were inspected for a second peak, which we
required to be at least 4.5
(respectively 32.4, 16.6 and 20.2
mJy observed flux density) and within the interval of
of the first peak.