THE MID-IR- AND X-RAY-SELECTED QSO LUMINOSITY FUNCTION

We present the J-band luminosity function (LF) of 1838 mid-infrared and X-ray-selected active galactic nuclei (AGNs) in the redshift range 0 < z < 5.85. These LFs are constructed by combining the deep multi-wavelength broadband observations from the UV to the mid-IR of the NDWFS Bootes field with the X-ray observations of the XBootes survey and the spectroscopic observations of the same field by AGES. Our sample is primarily composed of IRAC-selected AGNs, targeted using modifications of the Stern et al. criteria, complemented by MIPS 24 mu m and X-ray-selected AGNs to alleviate the biases of IRAC mid-IR selection against z similar to 4.5 quasars and AGNs faint with respect to their hosts. This sample provides an accurate link between low-and high-redshift AGN LFs and does not suffer from the usual incompleteness of optical samples at z similar to 3. We use a set of low-resolution spectral energy distribution templates for AGNs and galaxies presented in a previous paper by Assef et al. to model the selection function of these sources and apply host and reddening corrections. We find that the space density of the brightest quasars strongly decreases from z = 3 to z = 0, while the space density of faint quasars is at least flat, and possibly increasing, over the same redshift range. At z > 3, we observe a decrease in the space density of quasars of all brightnesses. We model the LF by a double power law and find that its evolution cannot be described by either pure luminosity or pure density evolution, but must be a combination of both. We used the bright-end slope determined by Croom et al. (2QZ) as a prior to fit the data in order to minimize the effects of our small survey area. The bright-end power-law index of our best-fit model remains consistent with the prior, while the best-fit faint-end index is consistent with the low-redshift measurements based on the 2QZ and 2SLAQ surveys. Our best-fit model generally agrees with the number of bright quasars predicted by other LFs at all redshifts. If we construct the QSO luminosity function using only the IRAC-selected AGNs, we find that the biases inherent to this selection method significantly modify the behavior of the characteristic density phi(*)(z) only for z < 1 and have no significant impact upon the characteristic magnitude M-*,J(z).