Astron. Astrophys. 326, 143-154 (1997)

2. Observations and data reduction

2.1. Sub-arcsecond imaging and deconvolution photometry

R 84 was observed using the ESO New Technology Telescope (NTT) during two runs. The best images were taken on 1991 December 26 using the SUperb Seeing Imager (SUSI) which functions with an active optics system (see ESO Web site for more information). The observing conditions were excellent with the seeing varying between 0 .50 and 0 .80 (FWHM). The detector was a Tektronix CCD (#25) with 1024² pixels of 24 µm. The filters used (their ESO numbers, central wavelengths, bandwidths), the exposure times, the dates, and the pixel size on the sky are summarized in Table 1.

Table 1. Journal of the imaging observations

Previously, NTT was used on 1990 January 10, during the commissioning period of the telescope, when it was equipped with EFOSC2. The detector was a Tektronix CCD (#16) with 512² pixels of 27 µm. The seeing varied between 0 .60 and 0 .75 (FWHM). Table 1 gives more information on the images.

Additional observations were carried out on 1988 August 31 at the ESO 2.2 m telescope using the adapter for direct imaging. The detector was an RCA CCD chip (#8) with 1024 640 pixels of 15 µm size. The seeing conditions were poor, 1 .3 (FWHM). However, the comparison of these 2.2 m observations with those obtained at the NTT telescope was very useful for checking the deconvolution code.

The data were all bias subtracted and flat-fielded. Only the image in the U band could not be flat-fielded, because of the too low S/N ratio of the flat-fields. On the other hand, the SUSI CCD produced some non-Gaussian noise in the images but at a very low level, negligible at the S/N ratio of the stars studied here. Since R 84 is a very bright object, it was not always possible to avoid saturation, especially on our good seeing observations. These are the SUSI R image which had 20 saturated pixels over R 84, and the EFOSC2 V image with 10 saturated pixels. We, therefore, used the observations of the other runs to check the results.

The photometry of the objects in the field of R 84 was carried out with a new deconvolution algorithm allowing not only to improve the spatial resolution of the images, but also to obtain reliable astrometric and photometric measurements of the stars. A full description of the method is given in Magain et al. (1997). The principle of that method is to avoid deconvolving with the total Point Spread Function (PSF), which would aim at obtaining infinite resolution. Rather, the new deconvolution allows to obtain an image with a better (but not infinitely narrow) PSF, basically chosen by the user.

In the case of R84, the final PSF is chosen to be a Gaussian with a FWHM of 3 pixels, the final pixel size being two times smaller than the original data pixels. The flux calibration was performed on the basis of the UBVR photometry carried out by Stahl et al (1984 ) in August 1983 using a diaphragm of 15 in diameter. This is basically the size of the field we use for the deconvolution. The integrated magnitudes are therefore re-distributed over all the components found.

Starting with a SUSI R image of seeing 0 .50 (FWHM) presented in Fig. 1, we get a restored image of R 84 with a final resolution of 0 .19 (FWHM), which is displayed in Fig. 2. We detect 31 components around R 84 over a 16 16 area. Owing to the high resolution of the images, for the first time we bring out stars #2, #4, and more especially #21 and #7 in the immediate vicinity of R 84 as well as the brighter components #34 and #35 lying further away to the south. Among the stars for which we have color indices, there are three red stars, R 84, #7, and #34. We will discuss about R 84, and #7 in Sect. 6. A prominent feature of R 84 is that it turns up to be the reddest star of the field. The photometric and astrometric results are summarized in Table 2. Note that the magnitudes of stars # 34 to #38 were obtained by aperture photometry.

Fig. 1. An R image of R 84 obtained using NTT+SUSI. Raw image with a resolution of 0 .50 (FWHM). Field 27 27 . Exposure time 1 sec. North is at the top and east to the left. Only the brighter components are labelled.

Fig. 2. A part of the SUSI R image focused on R 84 (left), and the corresponding deconvolution result (right). Star #1 is the transition object R 84. The pixel in the restored image is two times smaller than in the original one and the "seeing" is 0 .19 (FWHM). The faint diffuse background in this image comes from residual light due to diffraction by the spikes of the telescope. However, these residuals are negligible compared with the intensity of the stars themselves (see the text). The intensity cuts of this figure are chosen to display the full dynamics of the image, even at low light levels. Field 16 .4 16 .4. North is at the top and east to the left.

Table 2. Photometry of the core of LH 39 on the basis of deconvolved images. Star #1 is R 84

Despite the impression of perfection first felt when looking at the deconvolved images, one has to remember that it is a model of the reality constructed from imperfect data. If the PSF used for the deconvolution is derived from stars as bright as the object to deconvolve, Magain et al. (1997 ) have shown that the photometry of the point sources is basically photon noise limited even in the case of rather strong blends (e.g. two stars as close as one FWHM). However, in the data of R 84 there are some additional error sources in the astrometry and photometry: 1) the PSF is constructed on stars at least five times fainter than R 84; 2) R 84 itself is often saturated, sometimes heavily. Even if for most of the objects in the field of R 84 the only limitation to the photometric accuracy is the photon noise, the effect of an imperfect representation of the PSF is not negligible within a radius of 1 of R 84.

The PSF was constructed from 2 to 4 stars closer than , from R 84, in order to avoid any possible PSF variation across the field. In this small area, no star as bright as R 84 is available, especially in the red. In particular, the far wings of the PSF, as well as the diffraction spikes, are not modelled accurately enough for a perfect deconvolution of R 84 itself, and this affects the photometry of the closest neighbors, i.e. stars # 7 and # 21 (Fig. 2). Numerical simulations suggest that the uncertainty on the magnitudes of star # 21 is of the order of 0.3 mag, while it amounts to 0.2 for star # 7, in all the bands where we give a magnitude for these two objects. Note also that another consequence of the bad representation of the spikes of the PSF is to produce a diffuse background around the bright objects, especially R 84. This halo (Fig. 2) is not real, but is neither an artefact due to the deconvolution algorithm. It is simply due to the difference between the PSF used for the deconvolution and R 84 itself. However, the relative intensity between its highest values and the faintest stars is of the order of , negligible at the precision we need for our purpose.

Anyhow, the PSF was accurate enough to allow the photometry of R 84 itself even from the frames where the star central pixels are saturated. This was realized by giving an arbitrarily low weight to the saturated pixels, so that the image of R 84 was modelled from the wings of its PSF. Thanks to the good sampling of the original images, this procedure gives an accurate estimate of the star's magnitude and position. This is confirmed by comparing the results with those obtained from the unsaturated but much lower resolution images taken with the ESO 2.2 m telescope. Table 2 lists the magnitudes obtained for all the point sources with a S/N 10 in the central pixel. The typical error for a point source with this S/N ratio is of the order of 0.1 magnitude.

2.2. Adaptive optics imaging and near IR photometry

R 84 was observed in August and December 1995 with the ESO ADONIS adaptive optics system on the 3.6 m telescope. Images were taken in the H and K bands with a pixel size of 0 .05. For more details see Table 1. During the August run, four photometric standards were also observed: HD 115394, HD 193901, HD 207158 and HD 218814, with the following exposure times : 8 15 s, 20 5 s, 20 5 s and 4 45 s in both H and K. The reference star SAO 249234 was also observed, for later deconvolution, with an exposure time of 100 3 s in both H and K.

The images taken in August 1995 were affected by a strong noise due to the poor quality of the detector during that run. Moreover, a very bad seeing (2 to 3 ) throughout the night was responsible for a very poor adaptive optics correction. For instance, the Strehl ratio varied between 0.008 and 0.1 in K and the FWHM between 0 .6 and 1 .2. For these reasons, the observations of August 1995 only showed the two brightest stars of the field (R 84 and #11). These observations were nevertheless vital to perform a photometric calibration of the main star, thanks to the four photometric standards. This was done using an aperture of diameter 5 . The transformation from the instrumental system to the standard photometric system was carried out using the IRAF/NOAO PHOTCAL package. Note that the transformation was only possible for the August data as no photometric standard had been observed in December.

The photometric calibration enabled us to calculate the magnitude of star R 84: H = 8.56 0.04, K = 8.13 0.03. These results agree very well with those of Stahl et al. (1984 ). The errors include an uncertainty due to variations of the PSF with time estimated to be about 0.01 mag for our integration time (Esslinger & Edmunds 1997 ). The results were also checked by performing the same operation with some other aperture diameters. Once we had the magnitude of R 84, we could use it to calibrate the images taken in December 1995. This was carried out by measuring the flux of the star on these images with an aperture of the same size as before.

The images taken in December 1995 were very good (Strehl ratios 0.13 and 0.25 in H and K, FWHM resolutions 0 .12 and 0 .15 respectively) and showed 17 stars in a field of 12 .8 12 .8. To perform photometry we used both aperture photometry and PSF fitting in the IRAF/NOAO APPHOT and DAOPHOT packages. We used an aperture of diameter 0 .5. This size was chosen to include at least two dark rings of the diffraction-limited image, which limited the errors due to anisoplanatism to less than 0.01 mag (Esslinger & Edmunds 1997 ). PSF fitting, which is more sensitive to anisoplanatism, was only used to check the results of aperture photometry. Table 3 shows the results for each star, i.e. the magnitudes in H and K and the H - K colors.

Table 3. H and K photometry of the cluster

The accuracies for stars other than R 84 were generally better than 0.1 mag in both H and K bands. For star #15, with H = 16.74, it was about 0.3, and the worst was for star #19 of H = 18.94 which amounted to 0.8 mag. These were mainly the errors given by the photometry packages. For stars #7 and #15, which were in the halo of #1, the packages did not give accurate errors. We had to estimate them by changing the size of the aperture and checking the variations in the fluxes. We found that the magnitudes of stars #7 and #15 were respectively inaccurate within 0.2 and 0.3 mag in both bands. Stars #12, #17, and #19 were very close to the edge of the field of view, especially in K. This forced us to take a smaller aperture and also estimate the error ourselves. The magnitude of star #34 was not measured, since only a part of its halo was visible.

In an attempt to deconvolve our images from the run of December 1995, we used the PSF calibration star. Due to temporal variation of the ADONIS PSF, the result of the process is very disappointing. Both simple algorithms such as the Lucy-Richardson method and our new algorithm leave very significant residuals, due to the fact that the PSF used for the deconvolution is not the actual PSF of the image. As a result, we cannot detect faint objects very close to R 84. For display purposes, and to enhance contrast of the faint objects, we subtracted the PSF from R 84. In Fig. 3 we show the result of the operation, where the strongest residuals are masked.

Fig. 3. An undeconvolved adaptive optics image of R 84 obtained using ADONIS at the ESO 3.6 m telescope through an H filter. Resolution 0 .12 (FWHM) without deconvolution. R 84 being comparably very bright, we subtracted a properly placed and scaled reference star from it and masked the residual in order to bring out the fainter stars. Field 12 .8 12 .8. North is at the top and east to the left.

2.3. Spectroscopy: CASPEC echelle and EMMI long slit

R 84 was observed with the CASPEC spectrograph attached to the 3.6 m telescope on 1989 September 14. The 31.6 lines mm^-1 grating was used with a 300 lines mm^-1 cross dispersion grating and an f/1.5 camera. The detector was CCD #8, a high resolution chip of type RCA SID 006 EX with 1024 640 pixels and a pixel size of 15 µm. The central wavelength was 4250 Å and the useful wavelength range 3850 to 4820Å corresponding to orders 118 to 148 of the Thorium-Argon calibration arc. The resulting FWHM resolution as measured on the calibration lines is 0.2 Å. All the reductions were performed using the ECHELLE context of the MIDAS package. No flat-field correction was applied since the echelle orders in the flat-field frame appeared not aligned with the orders in the object frame. Orders 136 and 137 ( 4160 to 4190) are affected by a bad column of the detector and due to the lack of an appropriate flat-field correction, they could not be used. The individual orders were normalized by fitting fourth order polynomials and the accuracy of the normalization was checked by comparing overlapping regions of adjacent orders.

Several moderate resolution long-slit spectra were taken of two stars (#35 and #36) in the direction of R 84 using NTT+EMMI with grating # 12 on 1993 September 22. The CCD detector was Tektronix # 31 with 1024² pixels of size 24 µm. The range was 3810-4740 Å and the dispersion 38 Å mm^-1, giving FWHM resolutions of pixels or Å for a 1 .0 slit. Although the angular separation between stars #35 and #34 is only 1 , we expect no significant contamination of the spectrum of star #35 by #34 since the latter star is very faint in the blue.

© European Southern Observatory (ESO) 1997

Online publication: April 20, 1998
helpdesk.link@springer.de