On the compared accuracy and reliability of spectroscopic and photometric redshift measurements

We present a comparison between the catalog of spectroscopic redshifts in the Hubble Deep Field (HDF) recently published by Cohen and collaborators and the redshifts that our group has measured for the same objects using photometric techniques. This comparison is performed in order to fully characterize the errors associated with the photometric redshift technique. The compilation of spectroscopic redshifts incorporates previously published results, corrections to previously published wrong values, and new data, and it includes over 140 objects in the HDF proper. It represents the deepest, cleanest, most complete spectroscopic catalog ever compiled. We particularly study each and every object for which our redshift and the one measured by Cohen and collaborators seem to disagree. In most of those cases, the photometric evidence we put forth is strong enough to call for a careful review of the spectroscopic values, since the spectroscopic values seem to be in error. We show that it is possible to characterize the systematic errors associated with our technique, which when combined with the well-measured photometric errors allow us to obtain complete information on the redshift of each galaxy and its associated confidence interval, regardless of its apparent magnitude. One of the main conclusions of this study is that, to date, all the redshifts from our published catalog that have been checked have been shown to be correct (within the stated confidence limits). This implies that our set of spectrophotometric galaxy templates is a fair representation of the galaxy population at all redshifts (0 < z < 6) and magnitudes (R<24) explored thus far. On the other hand, spectroscopy of faint sources is subject to unknown and uncharacterized systematic errors. These errors will in turn be transmitted to any photometric redshift technique that uses spectroscopic samples in its calibration. Our analysis proves that photometric redshift techniques can and must be used to extend the range of applicability (in redshift, signal-to-noise, and apparent magnitude) of the spectroscopic redshift measurements.