We report Chandra X-ray constraints for 20 of the 52 high-redshift ultraluminous infrared galaxies ( ULIRGs ) identified in the Spitzer Extragalactic First Look Survey with f _ { \nu } ( 24 \mu { m } ) > 0.9 mJy , \log ( { \nu f _ { \nu } ( 24 \mu { m } ) \over \nu f _ { \nu } ( R ) } ) > 1 , and \log ( { \nu f _ { \nu } ( 24 \mu { m } ) \over \nu f _ { \nu } ( 8 \mu { m } ) } ) > 0.5 .
Notably , decomposition of Spitzer mid-infrared IRS spectra for the entire sample indicates that they are comprised predominantly of weak-PAH ULIRGs dominated by hot-dust continua , characteristic of AGN activity .
Given their redshifts , they have AGN bolometric luminosities of \approx 10 ^ { 45 } – 10 ^ { 47 } erg s ^ { -1 } comparable to powerful Quasi-Stellar Objects ( QSOs ) .
This , coupled with their high IR-to-optical ratios and often significant silicate absorption , strongly argues in favor of these mid-IR objects being heavily obscured QSOs .
Here we use Chandra observations to further constrain their obscuration .
At X-ray energies , we marginally detect two ULIRGs , while the rest have only upper limits .
Using the IRS-derived 5.8 \mu m AGN continuum luminosity as a proxy for the expected X-ray luminosities , we find that all of the observed sources must individually be highly obscured , while X-ray stacking limits on the undetected sources suggest that the majority , if not all , are likely to be at least mildly Compton-thick ( N _ { H } \gtrsim 10 ^ { 24 } cm ^ { -2 } ) .
With a space density of \approx ( 1.4 ) \times 10 ^ { -7 } Mpc ^ { -3 } at z \sim 2 , such objects imply an obscured AGN fraction ( i.e. , the ratio of AGN above and below N _ { H } = 10 ^ { 22 } cm ^ { -2 } ) of \gtrsim 1.7 :1 even among luminous QSOs .
Given that we do not correct for mid-IR extinction effects and that our ULIRG selection is by no means complete for obscured AGN , we regard our constraints as a lower limit to the true obscured fraction among QSOs at this epoch .
Our findings , which are based on extensive multi-wavelength constraints including Spitzer IRS spectra , should aid in the interpretation of similar objects from larger or deeper mid-IR surveys , where considerable uncertainty about the source properties remains and comparable follow-up is not yet feasible .