Why line surveys?

Hot molecular cloud cores are the formation sites of massive stars. in these cores, the complex chemistry gives rise to a set of abundant molecular species. Each molecule generates a set of spectral lines, resulting in a ``line forest''. As the sensitivity of telescopes and receivers has improved, it has become possible to reach the line confusion limit. To determine the presence and measure the abundance of any given molecule requires the observation of many lines. The best way to do this is by means of a line survey. In fact, unbiased spectral line surveys are necessary to obtain an unbiased view of the chemistry of a specific region. Unfortunately, due to atmospheric constraints, ground based telescopes are only able to observe limited frequency ranges. The most prominent and prototypical hot core is the Orion hot core. It is one of the richest known source of molecular lines, in part due to its proximity (about 500 pc). There have been a number of line surveys targeting this source at millimeter and submillimeter wavelengths. Examples are a 325 to 360 GHz line survey by Schilke et al. (1997), and a survey of the 607 to 725 GHz range by Schilke et al. (1998) [Fig. 1], both with the Caltech Submillimeter Observatory (CSO). A survey of Orion-KL from 780 to 900~GHz has been started by Schilke, Mehringer and Phillips at the CSO.

Fig 1.- Line survey of Orion-KL from 607 to 725 GHz, with the CSO. The data were taken in double sideband mode and subsequently deconvolved using an MEM type technique. The strongest lines and the (tentatively) new molecule SiH are marked.

Line surveys of star forming regions contain a wealth of data. With sensitive receivers, one can scan large spectral bands in a few days observing time, both from current ground based telescopes and with future airborne or spaceborne platforms. Also data reduction, in the sense of producing clean, calibrated spectra with all baselines etc. removed, is not very time consuming. The time consuming part is the identification of spectral features, determining source parameters and the interpretation. The identification part suffers potentially from incomplete molecular line catalogs. Although the catalogs, in particular the JPL catalog , have been augmented and improved considerably in recent years, there is still a lack of data for lines from molecules like vibrationally excited vinyl cyanide or ethyl cyanide, which will display a multitude of lines contributing to the low intensity ``chaff'' in high quality spectra. Apart from possible interest in these lines themselves, they make identification of other low intensity lines difficult if not impossible.

Interpreting line surveys

The next problem is determining molecular parameters, such as rotation temperatures, column densities etc. Traditionally, the approach here has been by-hand fitting of lines, possibly with multiple components, and producing rotation diagrams. For large surveys, by-hand fitting is impractical. Furthermore, one often suffers from line blending problems, which make determining line parameters difficult. Yet another drawback of rotation diagram methods is that they give erroneous results if the lines fitted happen to be optically thick. This can be amended by using isotopomeric lines to determine optical depths and correcting for that, but again that is a time consuming process, impractical for large amounts of data. In view of this, we have decided to approach the problem in a novel way. Using a molecular line catalog, the complete spectrum of a line survey is fitted at once, yielding directly the molecular parameters of interest. We assume that the line emission of each molecule is a superposition of several radiatively non-interacting components, each of which can be described by a source size, column density, velocity width, velocity offset and a single rotation temperature. These components can partly be identified as spatially distinct sources (hot core, compact ridge), partly as clump/interclump medium in one source. Another assumption is a Gaussian line shape for each component. Deviations from this behavior are certainly possible and will be present. They are partly compensated by assigning multiple components. The largest drawback of the method perhaps is the assumption of LTE, although in the case of the large densities of hot cores, this assumption is fairly accurate for most molecules.

However, compared to the classical rotation diagram method, this technique has several advantages: simultaneous fitting of multiple molecules and components, thus minimizing blending problems, and intrinsic ability to take optical depth effects into account. Even failure to fit the spectra using the assumptions here will provide the useful information of non-LTE conditions, which would otherwise not easily be identifiable. Unidentified lines are more easily found in the residuals of the data minus the fit. This software can also be used to predict spectra of frequency ranges inaccessible from the ground but observable with FIRST or SOFIA

Results from the CSO 607-720 GHz line survey of Orion-KL

Using the Caltech Submillimeter Observatory (CSO), we performed a line survey of the Orion hot core in the frequency range 607-725 GHz (Fig. 1). This spectral range covers almost the entire 450 micrometer atmospheric window bracketed by very broad H2O absorption lines at 556.9 GHz and 752.0 GHz. Two gaps are present in the band: one at 620.7 GHz, caused by atmospheric water, and one, less deep, at 715.4 GHz, caused by molecular oxygen. Additionally, ozone lines around 656 GHz worsen the transmission in this range. The data were taken between Nov '94 and Jan '95, using the 650 GHz facility SIS receiver (Kooi et al. 1994). Since the weather conditions were excellent, eight nights were sufficient to cover the whole spectrum twice. The single sideband system temperatures were in the range 1500--3000~K and the spectral resolution obtained is 1~MHz. The beam efficiency was measured to be 0.55 on Jupiter and the lines were corrected for that. The pointing was checked by observing planets and by observing lines of molecules which are known to peak in the vicinity of the hot core. We estimate an overall accuracy of 3''. The calibration was affected by occasional sideband imbalances. However, since each line was observed four times with different tunings, this was not a big problem. We estimate the overall calibration accuracy to be 30\%, except close to the deep atmospheric absorptions, where it is worse. The observations were performed using symmetric position switching to positions 5\arcmin\ away in Azimuth. Since the OMC-1 ridge runs north-south and the data were not taken at very low elevations, this setup excludes off-source contamination for all species with the possible exception of CO(6-5).

Fig 2.- Detail of Fig. 1: sulphur dioxide band at 660 GHz (lower panel). In the upper panel, the fit to the data is shown.

The double sideband data were subsequently deconvolved using a Maximum Entropy deconvolution method (similar to the one used by Sutton et al. 1995) to produce a single sideband spectrum. One advantage compared to CLEAN type deconvolution methods, which had been used for earlier Caltech surveys (Sutton et al. 1985, Blake et al. 1986, Schilke et al. 1997) is that it permits fitting the sideband gain ratios, while these ratios had to be provided for CLEAN. The achieved sensitivity is sufficiently high to securely identify 1~K lines over most of the band.

The disadvantage of all deconvolution methods is that pointing shifts result in line shape and strength changes for the various measurements of the same line (due to the structured source), which results in spurious features in the synthesized SSB spectrum. The MEM method seems to produce less of those than the CLEAN deconvolution and there are ways to suppress them by using model spectra. In questionable cases, inspections of the original DSB spectra were performed.

Line blending also makes secure identifications difficult in some cases, this time independent of the observing method. Here, line strength predictions based on other lines of the same species provide one way to disentangle features. Obviously, some iteration is required to derive a self-consistent result. Another problem, connected to blending, is the inability to determine a reliable baseline, which particularly affects the estimation of the line parameters of weak lines.

The analysis of the survey reveals 1064 spectral features consisting of 2128 lines, partially blended (see Fig. 1). The number of U-lines is 202 or about 19%. This number is presumably largely due to the sparseness of laboratory data in this wavelength range and will hopefully be reduced in the future. The spectrum is dominated by CO, CS, SO, SiO, HCN, HCO+, H2CO, SO2 and CH3OH, with many lines from very highly excited states. The sensitivity to these highly excited lines is partly due to the small beam size (about 12\arcsec), well matched to the hot core. Contrary to the findings of Harris \etal\ (\cite{harris}), we have no lack of weak or intermediate strength features in our spectrum (see e.g. Fig. 3).

Fig 3.- Detail of Fig. 1: methanol band and other lines at 673 GHz (lower panel). In the upper panel, the fit to the data is shown.

It has been predicted that the number of lines per frequency interval drops if one goes toward higher frequencies. Indeed, some molecules familiar from lower frequency work do not show up any more, notably the heavier linear rotors like HC3N, OCS and symmetric tops like CH3CCH, even CH3CN gets weaker, although many lines are still detectable. What is left are, as predicted, the simple di- and triatomic rotors, but also asymmetric molecules such as H2CO, CH3OH and SO2, and even heavy asymmetric rotors such as C2H5CN, HCOOCH3 and CH3OCH3. The complex spectra of these species results in a line density per frequency interval which is not much different from that at lower frequencies. New molecules, which do not contribute significantly to lower frequency surveys are the light hydrides, such as SiH, HCl and H2S, which begin to become important.

Using the continuum offset of the chopped observations in a line free part of the spectrum, we obtain an offset of 7.3 K = 237 Jy/beam at the CSO, taking the DSB calibration for the continuum offset into account. The average line emission over the range of the survey for identified lines is 1.3 K or 41 Jy/beam giving a total flux of 278 Jy/beam. Wright et al. (1992) measure a peak flux of 375 Jy/beam with the JCMT. Considering the different beam sizes, these two values are comparable to each other for a source size of about 10''. The integrated line emission contributes at least 15% to the total emission at this frequency. This value is in agreement with a prediction by Groesbeck (1994) who estimates that the line contribution to the total flux in Orion drops from 50=% at 800 micron to 10% at 450 micron. Our estimate takes only the identifiable lines into account. Approximately 81% of the easily discriminated lines are currently identified. However, a forest of low level lines and line wings could also contribute a pseudo-continuum of unknown strength. Models using Gaussian line shapes and the information available in the JPL catalog (Groesbeck 1994) predict that the contribution of this pseudo-continuum is low, but the existence of non-Gaussian wings could alter the contribution significantly. However, it seems implausible that the relative strength of the pseudo-continuum to the identifiable line emission changes very much with frequency. Thus, although the absolute values may be altered, it seems safe to conclude that the relative importance of line contribution to the continuum drops with frequency.

In Fig. 2 and 3, we show two selected frequency ranges, in the lower panel the measured data including the identification, in the upper panel the fitted data. In Fig. 2, a Q-Branch of SO2 is the dominant feature. The (preliminary) fit manages to reproduce the relative intensities reasonably well, although it fails to reproduce the complex spectral line shape. Figure 3 shows a somewhat wider band, which is dominated by CH3OH lines. Again, the preliminary fit produces most of the features, but there is room for improvement.