3. Optical observations
Optical images while the X-ray source was still on were taken for us in Service Time in the night of 1998 August 26 to 27 at the 3.5-m New Techology Telescope (NTT) at La Silla, using the Superb Seeing Imager SUSI2, and with the 8-m Unit Telescope #1 (Antu) of the Very Large Telescope at Paranal, using the VLT Test Camera. We will only discuss the NTT observations here, as these had better seeing. With SUSI2, one 10-s and two 100-s exposures were taken through a Bessell R filter, as well as two 900-s exposures through a Bessell B filter. During the observations, the seeing varied between from the first R-band image to in last B-band image. The night was not photometric. The detector was a mosaic of two EEV CCDs, each composed of square pixels of m on the side. They were read out binned by 2 in each direction, as the plate scale of would substantially oversample the seeing. For all but the first, 10-s R-band image, the telescope was offset such that the core of the cluster was not too close to the gap between the two CCDs. The data reduction was done using standard procedures, determining the bias from the overscan regions (after verification on bias frames) and correcting for pixel-to-pixel sensitivity variations using dome flats taken in the morning following the observations.
On 1999 July 15, when the X-ray source was off, we took images with the Focal Reducer/Low Dispersion Spectrograph FORS1 on Antu through Bessell R, B, and U filters. The night was not photometric, and the seeing varied from to . Two 10-s and one 100-s exposures were taken in R, two 30-s and one 300-s in B, and one 100-s and one 600-s in U. The detector was a Tektronix CCD with pixels of m. The standard resolution collimator was used, for which the plate scale is . The detector was read out through all four amplifiers, using the low-gain setting, of about .
The data reduction was done using standard procedures. From bias frames taken before and after the night, it was found that the level was somewhat variable, both in time and in position on the detector, but that the offsets remained constant relative to the levels found from the overscan pixels (the latter determined separately for the four quadrants). For bias subtraction, therefore, we subtracted both the levels from the overscan regions in individual frames, as well as an average of the overscan-corrected bias frames. The frames were corrected for sensitivity variations using flat fields constructed from images of the sky taken at dusk and dawn.
Since the conditions were not photometric during either of our two runs, we cannot reliably calibrate our data. We obtained a rough calibration using B and R magnitudes of Hamuy (1986) of the integrated flux of the cluster. Magnitudes are listed for two apertures, with diameters of 80 and 100"; these give consistent results. The calibration is consistent within 0.1 mag with one photoelectric B magnitude from Martins & Harvel (1979; star 5), within 0.3 mag with B and R magnitudes inferred from V and I magnitudes measured by Ortolani et al. (1994), but differs at the 0.5 magnitude level from what one would infer using photographic B magnitudes from Martins et al. (1980). We therefore estimate that the zero-point uncertainty on our quoted magnitudes is about 0.5 mag.
Astrometry of our frames was done relative to the USNO-A2.0 catalogue (Monet et al. 1998). For all USNO-A2.0 stars overlapping with the best-seeing FORS 10-s R-band image, centroids were determined and the pixel coordinates corrected for instrumental distortion using a cubic radial distortion function provided to us by T. Szeifert and W. Seifert (1999, private communication). From these, the zero point position, the plate scale, and the position angle on the sky were determined. Unexpectedly, the root-mean-square residuals are rather large, about in each coordinate. This is substantially larger than what we generally found in other fields; for instance, for a field at and , we find residuals of in each coordinate.
Most likely, the problem lies with the severe crowding in the field of NGC 6440, which causes many of the USNO-A2.0 stars to be blended. Furthermore, the USNO-A2.0 stars appear to have been derived from two different sets of plates, with epochs 1950.465 and 1980.488. The former will be plates taken from Palomar, at high airmass. For our solution, therefore, we selected only measurements for the more recent epoch, and from these, only those 246 stars which were well-exposed and appeared stellar on our FORS image. For these, the rms residuals were and in right ascension and declination, respectively. This is still much larger than usual, and therefore it is difficult to estimate the uncertainty. Fortunately, there is one Tycho star, TYC 6257-368-1, at the Southern edge of our VLT images. While this star is strongly overexposed, it was possible to determine a reasonably accurate centroid (verified using different exposures). The coordinates we infer using our astrometry are offset from the Tycho-2 coordinates (Hog et al. 2000) by and in right ascension and declination, respectively. We conclude that the systematic uncertainty in our astrometry is , i.e., substantially less than the uncertainty in the X-ray positions.
The individual frames were tied in to the astrometry using some 400 secondary reference stars in a square region 2´ on a side centered on NGC 6440, but excluding stars within 20" of the core. Typical residuals range from in each coordinate for the tie to the SUSI R-band images to for that to the FORS U-band images.
3.2. Variable sources
In Fig. 2, the reduced on and off-state B and R-band images are shown. The positions of the two X-ray sources fall close to the core of NGC 6440 and because of the crowding it is difficult to discern variable stars directly. To get an objective measure of variability in the core, we formed on-off difference images. For this purpose, we used the optimal image subtraction technique introduced by Alard & Lupton (1998). In this method, the image with better seeing is convolved with a kernel chosen such that the convolved point-spread function is as close as possible to that of the image with the worse seeing. In our case, this did not work perfectly, because for both bands the seeing is comparable in the on and off images, while the shapes of the point spread functions are somewhat different; the result is a convolution kernel which is negative in some points (in directions where "sharpening" of the better-seeing image is required). We proceeded by convolving the worst-seeing (FORS) images with a Gaussian with . Relative to this slightly smoothed image, it was possible to determine a kernel which is positive everywhere. After convolution with this kernel, the SUSI image and the slightly smoothed FORS image have very similar point-spread functions, and the subtraction works well.
The on minus off difference images found using the above method are shown in Fig. 2. It is clear that a number of stars brightened or dimmed considerably; a conspicuous one is star V0 on the Western end of the core. Most of these variables, including V0, are bright, red stars, most likely near the end of the asymptotic giant branch; indeed, the reddest, most luminous object in the sample of Ortolani et al. (1994) was found to be variable as well (it is out of the field shown in Fig. 2).
If the optical emission of the X-ray source is dominated by an accretion disk, as is generally the case for low-mass X-ray binaries, the optical counterpart is expected to be blue, and the brightess difference between on and off is expected to be relatively large. Given the problems with crowding, the resolution element that includes the source may not appear to be blue, but the difference between on and off should be blue. For this reason, we do not consider V0 to be a likely candidate; more likely, it is a Mira type variable. One could envisage a wide binary in which a Mira star transfers matter during its expansion. However, in that case the outburst would be expected to last rather longer than observed.
In each of the error circles, there is one source which was brighter when the X-ray source was on, and for which the difference flux is substantially bluer than that for other variables in the field. For X1, the source V1 is just outside the western edge of the error circle, while for X2, V2 is just inside the error circle. The coordinates of these variables are given in Table 3.V2 may be interesting in particular, since the relative brightening for that source is the second-largest in the field (about 10% in B, in R; the largest relative brightening - 50% and 100%, respectively - is shown by V0).
If V1 or V2 were the optical counterpart of the X-ray transient, we would expect that the source in quiescence had negligible contribution to the optical flux in the off image. The excess flux in the on image then would correspond to the the total flux of the source in outburst. We estimate the corresponding magnitudes using our approximate calibration. For V1 we find , , and for V2 , . Correcting for reddening, one infers and , respectively, again indicating that V2 is very blue.
From the above, it seems that V2 has all the characteristics expected for the optical counterpart. It is clear, however, that finding a variable source inside one of the two error circles is not unlikely, even one whose variation is relatively blue. Therefore, at present we consider this source as no more than an interesting candidate.
© European Southern Observatory (ESO) 2000
Online publication: July 13, 2000