BASS XLV: Quantifying AGN Selection Effects in the Chandra COSMOS-Legacy Survey with BASS

Yarone M. Tokayer, Michael J. Koss, C. Megan Urry, Priyamvada Natarajan, Richard Mushotzky, Mislav Balokovic, Franz E. Bauer, Peter Boorman, Alessandro Peca, Claudio Ricci, Federica Ricci, Daniel Stern, Ezequiel Treister, Benny Trakhtenbrot

First listed 2025-01-28 | Last updated 2025-03-10

Abstract

Deep extragalactic X-ray surveys, such as the Chandra COSMOS-Legacy field (CCLS), are prone to be biased against active galactic nuclei (AGN) with high column densities due to their lower count rates at a given luminosity. To quantify this selection effect, we forward model nearby ($z\sim0.05$) AGN from the BAT AGN Spectroscopic Survey (BASS) with well-characterized ($\gtrsim$1000 cts) broadband X-ray spectra (0.5-195 keV) to simulate the CCLS absorption distribution. We utilize the BASS low-redshift analogs with similar luminosities to the CCLS ($L_\mathrm{2-10\ keV}^\mathrm{int}\sim10^{42-45}\ \mathrm{erg}\ \mathrm{s}^{-1}$), which are much less affected by obscuration and low-count statistics, as the seed for our simulations, and follow the spectral fitting of the CCLS. Our simulations reveal that Chandra would fail to detect the majority (53.3%; 563/1056) of obscured ($N_\mathrm{H}>10^{22}\ \mathrm{cm}^{-2}$) simulated BASS AGN given the observed redshift and luminosity distribution of the CCLS. Even for detected sources with sufficient counts ($\geq30$) for spectral modeling, the level of obscuration is significantly overestimated. This bias is most extreme for objects whose best fit indicates a high-column density AGN ($N_\mathrm{H}\geq10^{24}\ \mathrm{cm}^{-2}$), since the majority (66.7%; 18/27) of these are actually unobscured sources ($N_\mathrm{H}<10^{22}\ \mathrm{cm}^{-2}$). This implies that previous studies may have significantly overestimated the increase in the obscured fraction with redshift and the fraction of luminous obscured AGN. Our findings highlight the importance of directly considering obscuration biases and forward modeling in X-ray surveys, as well as the need for higher-sensitivity X-ray missions such as the Advanced X-ray Imaging Satellite (AXIS), and the importance of multi-wavelength indicators to estimate obscuration in distant supermassive black holes.

Short digest

Uses local BASS analogs with well-characterized 0.5–195 keV spectra, forward-modeled into the CCLS redshift–luminosity–exposure space and re-fit with the CCLS pipeline to quantify obscuration biases. Finds Chandra would miss 53.3% (563/1056) of simulated obscured (NH>10^22 cm^-2) sources, and even detected ≥30-count spectra have NH systematically overestimated. The strongest failure mode is for CT best-fits (NH≥10^24 cm^-2), where 66.7% (18/27) are actually unobscured (NH<10^22 cm^-2). Results imply deep <10 keV surveys overstate the rise of the obscured fraction and the fraction of luminous obscured AGN, motivating forward modeling, multi-wavelength obscuration tracers, and higher-sensitivity missions like AXIS.

Key figures to inspect

• Figure 1: Check the L2–10 keV vs. redshift overlap of BASS and CCLS to validate that BASS provides luminosity-matched local analogs for the forward modeling.
• Figure 2: Inspect the luminosity, redshift, and exposure matching (panels a–c) to see that the simulations reproduce the CCLS selection function; this underpins the demographic comparisons.
• Figure 3: Follow the count-rate vs. redshift tracks for unobscured vs. CT templates at CCLS quartile luminosities; note where sources drop below 3/30/70-count thresholds and the reflection-driven bump for high-L CT near z≈2 entering Chandra’s band.
• Figure 4: Compare count distributions for simulated detections vs. CCLS and the NH-split histograms; observe that obscured/CT spectra pile up at low counts and that most undetected zero-count cases are heavily obscured/CT, illustrating the 53% miss rate.