2 The observations

Observations were performed with the IRAM-30 m-telescope and the SEST each on three different occasions and with the JCMT on one occasion. In Table 1 we list the observations of the different runs, the observed coordinates, chemical type (C or M) and pulsation period. In addition, references are given to previous detections in CO and in the case of non-detections (previously and by us) we refer to radial velocities obtained by other means, if avaialable. The stellar coordinates are in most cases taken from the SIMBAD data base (when listed to $\stackrel{<}{\sim}$0.2 $^{\prime\prime}$ accuracy), the Hipparcos input catalog, the Stephenson (1989) catalog of carbon stars, or derived from the Digitised Sky Survey. In few cases the IRAS coordinates are used, mostly for the less known carbon stars and the stars listed under their IRAS name. The pulsation periods are taken from the 4th edition of the General Catalog of Variable Stars (GCVS, Kholopov et al. 1985), Jones et al. (1990), Le Bertre (1992) and Joyce et al. in preparation.

The data obtained with the IRAM-30 m-telescope were taken in the period between 23 and 29 December 1994, between 26 and 28 January 1995 and on the 3rd and 4th of August 1995. For the December 1994 run (observers MG and JB) the ¹²CO J = 1-0 and J = 2-1 lines, and for some stars also the HCN(1-0) and SO(6₅-5₄) lines, were observed simultaneously using the 230 GHz and 3 mm SIS receivers. Both the two 1 MHz filter banks and the autocorrelator were used as backends. In cases where both data sets are available with good S/N we only present the one with the highest frequency resolution. The same procedure is adopted for the other observations discussed below. Table 3 gives the channel spacing which we finally decided to use.

The targets were observed using wobbler switching during all runs with the IRAM-30 m-telescope, mostly with a throw of 120 $^{\prime\prime}$ in azimuth.

Baselines were removed and all temperatures are on a main-beam brightness scale (the same applies for all temperatures presented in this paper). Due to different definitions used at different telescopes the observed antenna temperatures ($T_{\rm A}$) are converted to main-beam brightness temperatures ( $T_{\rm mb}$) by using $T_{\rm mb} = T_{\rm A}/B_{\rm eff}$ for the SEST and JCMT data (discussed below), and $T_{\rm mb} = T_{\rm A} \times F_{\rm eff}/B_{\rm eff}$ for the IRAM data, where $F_{\rm eff}$ and $B_{\rm eff}$ are respectively the forward and main-beam efficiencies (listed in Table 2 together with the FWHM beam widths).

The comparison with the calibration sources (supplemented by consistency checks from stars observed during more days, or even observed both in the December and January runs, as well as published results obtained with the IRAM-30 m-telescope) indicated that the CO J = 1 - 0 and the HCN observations during the December 1994 run are consistent with the calibration observations. The situation is different for the CO J = 2-1 and SO data. The observed CO J = 2-1 data had to be multiplied by factors 1.1 - 1.5 to obtain agreement with previous results. For the SO transitions there are no calibration sources and only very few published data. The 230 GHz receiver was tuned to the SO line during two consecutive days during the December 1994 run. We observed IRC+10 011 on both days and compared our data to that of Bujarrabal et al. (1994). On the first day our brightness temperature scale was too low by a factor of 2.2; on the second day we found agreement between the two. By scaling the result for VY CMa obtained with the NRAO 12 m telescope to the expected value for the IRAM-30 m-telescope, assuming a point source, we again found an underestimate of the temperature scale by a factor of 2.2 for the first day of SO observations. Hence, for the first day of SO observations the observed brightness temperatures were multiplied by 2.2; for the second day no correction was applied. We estimate the final calibration for all observed lines to be accurate by about 10% (1$\sigma$). This is true for all runs with the IRAM-30 m-telescope.

For the January 1995 run (Observer MG) the ¹²CO J = 1-0, 2-1 and 3-2 transitions were observed simultaneously with three SIS receivers. We used as backends the 1 MHz filterbank and the autocorrelator, which was split into three parts. The data reduction is similar as described above. For the J = 3-2 line there were at that time no calibration observations available for the IRAM-30 m-telescope.

For the August 1995 run (observer EJ) the ¹²CO J = 1-0, 2-1transitions were observed simultaneously, in a set-up identical to the December 1994 run.

The JCMT data were taken on August 28 and September 1 & 7, 1995 (observers FB and RT). The ¹²CO J = 2-1, J = 3-2 and, for a subsample, ¹³CO J = 2-1 and J = 3-2 were observed using two SIS receivers. The backend was the digital autocorrelation spectrometer. A chopping secondary was used with a throw of 2$^\prime$. The data reduction is similar as described above. We estimate the calibration to be accurate to about 10% (1$\sigma$).

The first set of SEST data was taken between 11 and 16 September 1996 (observer MG). Either ¹²CO J = 1-0 and SiO (3-2, v=0), or ¹²CO J = 1-0 and CS (3-2) were observed simultaneously using the 2 mm and 3 mm SIS receivers. The former setting was used to observe O-rich sources, and the latter was used for the C-rich sources. This was not our preferred set-up but the 1.3 mm receiver was unavailable during our run to observe the CO (2-1) line. The dual beam switching mode was used with a throw of 2$^\prime$26 $^{\prime\prime}$ in azimuth. One of two available acousto-optical spectrometers was split and connected to the two receivers with a channel separation of 0.7 MHz.

During the second (20-21 September 1997; observer RS) and third (30 September 1998; observer MG) SEST run the ¹²CO J = 1-0 and 2-1lines were observed simultaneously with the 1.3 mm and 3 mm SIS receivers using the dual beam switching mode as outlined above. We targeted only those stars which had already been detected in the 1996 September run. The data reduction for the SEST runs is similar as described above. Based on a comparison with calibration sources we estimate the brightness temperature scale to be accurate to about 10% (1$\sigma$).