Photometric redshifts of faint X-ray sources

Active Galactic Nuclei (AGN) are the accreting super massive black
hole (SMBH) at the center of massive galaxies. The tight M-σ and
M_BH-M_bulge correlations reveal that the host galaxies are
affected by the effects of the SMBHs. In addition, many works
studying on the rest-frame color-magnitude relation have shown that
AGN host galaxies have intermediate colors, which are considered as
a transition from the blue cloud to red sequence in host-galaxy
color evolution. Some works interpreted this result as an evidence
for the AGN feedback, in the sense that the accretion process not
only builds up the massive black holes, but also has a powerful
influence on the surrounding environment, triggering or suppressing
the star-forming activity in the host galaxy. These correlations
make obvious the need to investigate AGN-host coevolution. One way
to understand this coevolution is to study the AGN duty cycle
(i.e., the time scale that the SMBH is active), which can be
obtained by estimating AGN population among all the galaxies
through cosmic times. Such demographic studies require a
well-sampled census and accurate redshift information. In
particular at high redshift, objects are extremely faint and sample
numbers are very small. This could cause large statistical errors.
For example, Aird et al. (2010) argued that luminosity-dependent
density evolution with a flattening faint-end slope of the X-ray
luminosity function (XLF) at z > 1.2 may result from
catastrophic photo-z failures caused by observational limitations
and improper templates used for photo-z computation. However,
spectroscopic redshifts are time-consuming and difficult to be
obtained for faint sources at high redshifts. Therefore we have to
rely on photometric redshifts (photo-z) techniques which need to be
tuned specifically to be reliable for AGNs (i.e., proper magnitude
priors, appropriate AGN-galaxy hybrid template for SED-fitting, and
correct multi-wavelength counterparts). In recent years, many deep
and high-resolution observations become available in multiple
wavebands, specially at near/mid-infrared. This allows us to reach
higher redshift, and make more accurate analysis on the
multi-wavelength properties of AGNs. In this thesis, we focus on
the ECDFS area which comprise also the GOODS-S and CDFS regions.
This is the portion of the sky with the deepest and most complete
photometric information from X-ray to radio, including intermediate
bands from the Subaru telescope, and optical/near-infrared data
from the Hubble space telescope. To compute accurate photo-z using
these data, first we combined multi-wavelength catalogs from UV to
infrared after the astrometric calibration and correction for the
different methods of flux extraction (e.g., total fluxes, flux
apertures and PSF- fitted photometry). Second we identified the
best multi-wavelength counterparts for X-ray sources from the
4Ms-CDFS and 250ks-ECDFS surveys, taking into account the
positional errors and multiple magnitude distributions as priors
simultaneously. We found that more then 96% of X-ray sources have
multi-wavelength counterparts. Thirdly we built a new library of
active galactic nuclei/galaxy hybrid templates appropriate for the
faint X-ray population in the CDFS to simulate the AGN spectral
energy distribution from low to high redshift. For X-ray-selected
AGNs, we achieved a photo-z accuracy of 0.013 with an outliers
fraction of 5.3%, while for non-X-ray galaxies, the photo-z
accuracy is 0.010 with an outlier fraction of 4.6%. With the
SED-fitting results of our well-trained AGN-galaxy hybrids, we
further studied the galaxy and AGN host properties via the
rest-frame color-magnitude diagram (CMD) which is an useful probe
to trace the stellar populations. We made corrections for dust
extinction and/or AGN contamination for the galaxy/AGN host colors
in the CANDELS/GOODS-S region. We found that the AGN host colors
also present bimodality in the CMD up to z~2.5 as found in normal
galaxies, and the position of the blue peaks in the AGN samples are
almost constant with cosmic time. This implies a weak connection
between AGN activity and star formation in the host galaxy. For the
X-ray sources in the 4Ms-CDFS survey, we found that for most of the
sources, the correction for dust extinction is larger than the
correction for the AGN contribution. This is because the AGN
population in this field is dominated by low-luminosity AGNs which
have host-dominated SEDs. However for few bright sources, their
host colors are strongly effected by AGN contribution rather than
by the dust extinction. For these sources, the correction for AGN
contribution is about two times larger than the correction for dust
extinction in general. Therefore AGN/galaxy decomposition becomes
more important in a shallower and wider X-ray surveys, e.g.,
XMM-COSMOS and eROSITA, which contains a larger fraction of bright
AGNs. Furthermore, with our accurate redshifts for galaxies and
AGNs, we defined a high-redshift (high-z) sample using the redshift
probability distribution function P(z) rather than relying on the
best-fit value of photo-z. We integrated P(z) within a given
redshift range to obtain the photo-z probability in that range and
selected high-z sources above a given threshold. When computing the
number of sources in a given redshift range, each source will not
be counted as "1" but as the proportion of it. We compared this
P(z) technique with traditional color techniques adopted for galaxy
evolutionary stages, like the Lyman break galaxy and the BzK
color-color selection via sample completeness and purity. We found
that the P(z) technique is the most efficient and reliable method
for selecting high-z sources. This is not surprising as it makes
use of photometric information from the entire SED rather than
using only three photometric points. Lastly, we built a high-z (z
> 3) sources list for X-ray sources in ECDFS region, and
compared our list with previous work. In our work, we made better
X-ray-to-optical/NIR associations considering the positional errors
and magnitude distribution. In addition, we obtained accurate
photo-z using well-established AGN-galaxy hybrids for X-ray
selected AGNs and applied P(z) for each source. These procedures
help improving on our high-z sample selection.