PAUCam is a Visitor’s Instrument at the William Herschel Telescope. The instrument is operated by the PAUCam collaboration, formed by 5 of the 7 groups originally in the Consolider Project, namely CIEMAT and IFT (in Madrid), and IEEC, PIC and IFAE (in Barcelona). PAUCam is also offered for public use by interested members of the WHT community of users, when not dedicated to the PAUS survey. At present, external groups are assisted by the PAUCam collaboration in the operation of PAUCam.
PAUCam had first light on the night of June 3, 2015. Since then, we had observation nights in 2015B (10 days), 2016A (13 days), 2016B (20 days), 2017A (27 days) and 2017B (28 days). The operation of the camera was correct. However some detected problems led us to two interventions, during the summers of 2016 and again of 2017. A picture of the camera is shown in Fig. 1.
PAUS is a photometric (as opposed to spectroscopic) survey, with images taken with a large number of filters as to obtain what we could call a pseudo spectrum, or low-resolution spectrum. The goal was to obtain a resolution in redshift z better than 0.0035(1+z) for galaxies up to magnitude iAB = 22.5 (and a sizable sample up to iAB = 23.5). Based on simulations we concluded that the number of required filters would be about 40. Figure 2 shows the actual spectral responses of the 40 narrow-band filters. The wide-band filter set, in the “ugrizY” bands is intended for other usages of PAUCam, others than those proposed by the PAU Survey collaborators (and in fact they have been used by external observers already). The wide-band filter transmissions are illustrated in Figure 3. A detailed PAUCam description and operation will be published during the first half of 2018.
The work during 2017 has concentrated in the many issues of data processing required to bring the data to science quality status. There has been a lot of progress in several aspects of this work:
Data refinements. When reading the CCDs it was noticed that there was some noise around their periphery. This was traced to scattered light that entered a small gap between the filters and the frame holding them. New trays were installed during the summer of 2017, and the problem mainly disappeared. For earlier data the problem can be corrected by masking a few CCD columns.
A code solution was implemented to correct the charge effects induced by very large signals in some CCD pixels. This cross-talk is an effect of the way the charge is collected and read in the CCDs and affected images with very bright objects.
Photo-z precision. The basic goal of PAU is to obtain accurate pho-z for all the objects in a given area up to certain magnitude, namely 0.0035(1+z) precision in z, for all objects up to iAB = 22.5 (and a large fraction of those to iAB = 23.5). We are not far from that goal. The main reason for not having yet reached the goal is related to the fact that we find some objects in which the flux of a narrow band bin is substantially different from that of the neighbors. Some of this “outlayers” are possibly caused by the scattered light problem mentioned above. This is being investigated at present.
Calibration. Since the bands of the PAU filters have never been used before in any project, there are no standard sources to calibrate the individual responses of the PAU filters. Given the large number of filters it is not possible to calibrate them with their own data by for example measuring spectrophotometric standards in the right conditions. A method was therefore developed for PAU photometric calibration.
Star-Galaxy separation from PAU fluxes. As with the photo-z measurements one can expect that the large number of PAU filters would facilitate the separation of stars and galaxies just with their fluxes. After trying several machine-learning methods, preliminary information indicates that this is indeed the case. With a convolutional neural network and with a learning sample based in a matched sample of COSMOS objects, we can separate stars and galaxies with high efficiencies. This can be seen in Figure 4, which shows TPR-FPR curves for several sizes of the training sample. With a large enough training sample the classification is exceedingly good. Again this will be published in an incoming publication.