Spitzer IRAC selection is a powerful tool for identifying luminous AGN .
For deep IRAC data , however , the AGN selection wedges currently in use are heavily contaminated by star-forming galaxies , especially at high redshift .
Using the large samples of luminous AGN and high-redshift star-forming galaxies in COSMOS , we redefine the AGN selection criteria for use in deep IRAC surveys .
The new IRAC criteria are designed to be both highly complete and reliable , and incorporate the best aspects of the current AGN selection wedges and of infrared power-law selection while excluding high redshift star-forming galaxies selected via the BzK , DRG , LBG , and SMG criteria .
At QSO-luminosities of log L _ { 2 - 10 keV } ( ergs s ^ { -1 } ) \geq 44 , the new IRAC criteria recover 75 \% of the hard X-ray and IRAC-detected XMM -COSMOS sample , yet only 38 % of the IRAC AGN candidates have X-ray counterparts , a fraction that rises to 52 % in regions with Chandra exposures of 50-160 ks .
X-ray stacking of the individually X-ray non-detected AGN candidates leads to a hard X-ray signal indicative of heavily obscured to mildly Compton-thick obscuration ( log N _ { H } ( cm ^ { -2 } ) = 23.5 \pm 0.4 ) .
While IRAC selection recovers a substantial fraction of luminous unobscured and obscured AGN , it is incomplete to low-luminosity and host-dominated AGN .