2. Observations, reduction and additional data
The observations were made between 1995 March and 1997 May with the 14-m radome-enclosed millimeter telescope of the Five College Radio Astronomy Observatory located in New Salem, Massachusetts. In all, 86 full 12-hour transits (LST 19h- 7h) were allocated to the M 31 project at an average rate of 5 transits per month. About 40 of that time was lost to bad weather conditions. The telescope is of Cassegrain design with an azimuth-elevation mount. Since the pointing becomes less reliable at elevations greater than 75o, we used only that portion of the allocated LST range during which M 31 was in the elevation range 35o-75o. The beginning and end of the shifts (el 35o) were used for pointing and calibration measurements, while the LST range 23:00-02:00 (el 75o) was used to perform CO observations of M 33. The pointing and focus were optimized by observing strong SiO masers at least 2 or 3 times during each 12-hour run. The pointing was shown to be accurate to 7" rms.
The QUARRY receiver is a single-sideband multi element focal plane array (Fig. 3, Erickson et al. 1992), with 15 Schottky diode mixers, IF amplifiers and polarization interleaving optics within a single dewar that rotates to track equatorial (or Galactic) coordinates. The quasi optical system in front of the feed horns contains a chopper wheel for system calibration and a Michelson interferometer for image sideband termination (rejection level 18-20 dB). The 15 receivers are tuned automatically under computer control; the single side band (SSB) receiver noise temperature averaged over the array is typically 300 K at 115 GHz. The observations were calibrated using the standard chopper-wheel method (Penzias & Burrus 1973), switching between the sky and an ambient-temperature blackbody. The total system temperature during the survey varied between 600 K and 1200 K. The intensities in this paper are expressed in main-beam brightness temperatures: = /. The main beam efficiency of the telescope is 0.42 at 115 GHz (Ladd & Heyer 1996).
The spectrometers were a set of 15 autocorrelators, each providing a total bandwidth of 80 MHz, a spectral resolution of 378 kHz, and a sampling of 300 kHz. At 115 GHz, this yields a velocity resolution of 1.0 , a sampling of 0.8 , and a total bandwidth of 210 . The typical linewidths of the observed CO profiles were 10-30 , so the spectra were Hanning-smoothed to a resolution of 3.25 to improve the signal-to-noise ratio.
While the velocity coverage for each position of our CO survey is limited to 210 by the width of the correlator, it is known that the H I at certain positions - particularly within 5 kpc of the center - can be spread over as much as 300 . This wide H I coverage results from the fact that for most positions across the disk of M 31, two velocity components are present: one corresponding to the main disk itself and the other to the warped outer disk seen in projection against the main disk (Brinks & Burton 1984). The sensitive CO survey performed along the major axis by Loinard et al. (1995) with the IRAM 30-m telescope showed that at an angular resolution of less than 1´ the kinematics of the molecular gas in the main disk follows very closely that of the H I , but that the warped outer disk is not detected in CO. In the inner 5 kpc, where the H I emission associated with the main disk can become fainter than that associated with the warped outer disk, if any CO is detected, it is always at the velocity of the main disk . We used the Westerbork H I observations of Brinks & Shane (1984) to determine the velocity of the expected CO emission - always centering on the component associated with the main disk. While in the Population I ring the velocities of both the main and the outer disk could be included in the backends, close to the center, only the main disk could be covered.
With QUARRY, a 30 pixel 5´ 4´ map sampled every 50" (a footprint ) can be obtained in two pointings (Fig. 3). To cover most of the southwestern half of M 31 where emission was detected by the CfA survey, we observed 76 adjacent footprints, representing 1,500 square arcminutes and 2,280 individual spectra. The resulting map is regularly sampled every beamwidth (50"), i.e. undersampled by a factor of two in both directions relative to Nyquist sampling. This sampling was chosen because it provides roughly four times the spatial coverage of Nyquist sampling, while degrading the effective angular resolution only slightly. Numerical simulations with Galactic CO data smoothed to the linear resolution of the 14-m telescope at the distance of M 31 confirmed that such undersampling had a negligible affect on our maps and on the conclusions deduced from them. The effective resolution of our survey is 1´.
To achieve flat spectral baselines, the observations were made by position switching every 15 seconds between the source and an emission-free OFF position taken well outside the molecular disk of M 31 (at a displacement in azimuth of to depending on the observed position), and at the same elevation as the ON. Each scan consisted of 10 such ON-OFF sequences, with a total (ON+OFF) integration time of 5 minutes. On average, 20 scans (100 minutes of integration) were necessary to obtain the target sensitivity of 20-25 mK in at a resolution of 3.3 .
2.2. Data reduction
The Westerbork H I survey of Brinks & Shane (1984) was used to determine for each position the velocity range where CO emission was expected. To each individual spectrum, a horizontal line (0th order polynomial) was fitted to the baseline outside this "emission window" (generally about 50 wide) and removed. The spectra were then inspected individually, and those with obvious baseline distortions were discarded. Scans corresponding to the same position were averaged together, weighted according to their individual rms noise levels. Usually, a first order baseline was removed from the average spectra, but in some cases, second or third order baselines were removed. Given the large width of the backend compared to that of the emission lines, this could be done quite reliably. A sample of fully reduced CO spectra is shown and compared with H I spectra in Fig. 4.
Since the velocity range covered by the survey was larger than the bandwidth of the spectrometer, different footprints were centered at different velocities. To make data manipulation easier, the fully reduced spectra were resampled on a common spectral grid, and eventually stored in a FITS format data cube. Any further processing of the data set was made on this cube.
As will be shown, the CO emission in M 31 is widespread both spatially and in velocity, but most of the spectral lines detected are fairly weak and, as was mentioned earlier, much narrower than the bandwidth of the spectrometer. Many of these lines are lost in the noise of a spatial map integrated over all velocities; space-velocity maps integrated over all X or all Y suffer a similar degradation of signal-to-noise. In order to lower the noise in our integrated maps by a factor of a few, we have used the "masked moment" method originally developed to analyze 21-cm surveys of galaxies with limited signal-to-noise (Tilanus & Allen 1991), and later applied to extragalactic or Galactic CO data (see Adler et al. 1992; Digel et al. 1996). This method uses a heavily smoothed and therefore very low-noise version of the survey to define integration limits. In our case the survey was smoothed to a resolution of 20 in velocity and to 2´ spatially, resulting in an rms noise of 4 mK. Integrations of the original survey were then taken only over channels for which the intensity in the smoothed survey was above 12 mK (3).
There are a few drawbacks to the masked moment method. First, the smoothing will result in an enhanced signal-to-noise ratio only for emission features that are still at least marginally resolved (in space and velocity) after the smoothing. Small, weak features might therefore be lost in the process, but such small features would be lost in the noise in simply integrated maps as well. Second, the noise in integrated maps is not constant; one can however calculate the noise at each position based on the size of the integration window(s). In our case, the noise in the velocity-integrated map (Fig. 5) is 0.45 K in the molecular ring and 0.35 K elsewhere; the lowest contour in Fig. 5 thus corresponds to at least 3 over nearly the entire map, and each feature in it is therefore very likely real. Indeed, we have confirmed point by point that nearly every feature in Fig. 5 corresponds to a significant spectral line at a velocity coincident with an H I component.
2.3. Additional data
In addition to our CO survey, we will make use here of existing data sets in the literature or provided by colleagues. Massive H II regions as traced by H emission have been listed by Pellet et al. (1978); a list of OB associations has been compiled by Magnier et al. (1993). We will also make use of the UV map at 203-nm obtained by Milliard (1984) using the French balloon-born telescope SCAP 2000. The image used here is that produced by Koper (1993) who removed the Galactic foreground Galactic stars from it, re-calibrated it and smoothed it to an angular resolution of 2´.
The distribution and kinematics of the atomic hydrogen in M 31 are deduced from the complete 21-cm survey made by Brinks & Shane (1984) at Westerbork; its angular resolution of 24" 36" (in and ), is similar to ours, and its spectral resolution (8.2 ) is adequate for an accurate study of kinematics. The southwestern half of M 31 is approaching us at a velocity of -300 to -600 , and is therefore essentially free of Galactic foreground emission. Under the assumption that the 21-cm line remains optically thin, its integrated intensity map then provides directly the mass surface density of atomic hydrogen.
The distribution of dust can be obtained in various ways, none of which are free of biases and selection effects. We will use here the IRAS map at 100-µm (Xu & Helou 1996) produced at the Infrared Processing and Analysis Center (IPAC) with the IPAC high resolution (HiRes) processor (Aumann et al. 1990, Rice 1993). Its angular resolution is 100". Since only fairly warm dust emits significantly at 100-µm, this image gives little information on the distribution of cold dust. To provide some information on that, we will use an optical image obtained at the Palomar Schmidt telescope (POSS II) and processed by the ESO photographer Hans-Hermann Heyer, who removed, applying a process known as unsharp masking, an extended component from the original image in order to enhance the contrast. Dust absorption is more easily seen in such an image, but it is one which conveys little quantitative information, and suffers from a strong selection effect: only dust located in front of a strong stellar background will be detected. That located in front of the galactic bulge is readily detected for example, but that behind is not.
© European Southern Observatory (ESO) 1999
Online publication: November 16, 1999