Unlocking AGN Variability with Custom ZTF Photometry for High-Fidelity Light Curves and Robust Selection

TL;DR

This work addresses robust identification of active galactic nuclei via optical variability, including systems where host galaxy light dilutes the nuclear signal. It advances methodology by constructing high-fidelity light curves from aperture photometry on ZTF difference images (DI-Ap) with careful epoch calibration, enabling reliable variability metrics such as the standard deviation, Pvar, and qualitative DRW timescales. A balanced hierarchical random forest is trained on these DI-Ap features to classify nearly 40 million sources over >8,000 deg^2 into 17 classes, yielding 341,938 AGN candidates across four redshift bins with high AGN recovery and reduced contamination in host-dominated systems. Cross-validation with Gaia-unWISE colour selections and eROSITA X-ray data shows that the variability-selected AGN sample is complementary to colour and X-ray selections, including a substantial population of X-ray faint or undetected AGN, demonstrating the method's broad applicability for SMBH demographics and multiwavelength AGN studies.

Abstract

(Abridged)We explore the potential of optical variability selection methods to identify AGN, including those challenging to detect with conventional techniques. Using the unprecedented combination of depth, sky coverage, and cadence of the ZTF survey, we target even starlight-dominated AGN, known for their redder colours, weaker variability signals, and difficult nuclear photometry due to their resolved hosts. We perform aperture photometry on ZTF reference-subtracted images for 40 million sources across 8,000 deg^2, assemble light curves and classify objects employing an RF algorithm into 14 classes, including 341,938 candidate AGN. We compare variability metrics derived from our photometry to those obtained from ZTF Data Release light curves (DR11-psf), to assess the impact of our analysis. We find that the fraction of low-z quiescent galaxies exhibiting significant variability drops dramatically (from 98\% of the sample to 7\%) when replacing the DR11-psf light curves with our difference image, aperture photometry (DI-Ap) version. The overall number of variable low-z AGN remains high (99\% when using DR11-psf lightcurves, 83\% when using DI-Ap), however, implying that our photometry can detect the fainter variability in host dominated AGN. The classifier effectively distinguishes between AGN and other sources, demonstrating high recovery rates even for AGN in resolved nearby galaxies. AGN candidates in eROSITA's eFEDS field, detected in X-rays and bright enough for ZTF optical observations, were classified as AGN (79\%) and non-variable galaxies (20\%). These groups show a 2 dex difference in X-ray luminosity but not in X-ray flux. A significant fraction of X-ray AGN are optically too faint for ZTF, and conversely, a quarter of ZTF AGN in the eFEDS area lack X-ray detections, highlighting a wide range of X-ray-to-optical flux ratios in AGN.

Figures (16)

• Figure 1: Distribution of the standard deviation of the light curves obtained from our custom-made photometry (blue, solid line) and directly from the ZTF data release DR11-psf light curves (pink, dashed line) for four types of objects in the labelled set: non-variable stars, non-variable galaxies, and low redshift ($z<0.5$) AGN and mid-redshift AGN ($0.5<z<3$). The standard deviation is calculated from the magnitudes.
• Figure 2: Distribution of DRW $\tau$ values associated with our DI-Ap light curves (blue, solid line) and DR11-psf light curves (pink, dashed line), respectively, for non variable stars, non-variable galaxies, low redshift ($z<0.5$) AGN and mid-redshift AGN ($0.5<z<3$) in the labelled set.
• Figure 3: Confusion matrices of the node_init (top-left), node_variable (bottom-left), and third level (right) obtained by using the HBRF in the testing set. The confusion matrices show the results as percentages, rounded to integer values, which are computed by dividing each row by the total number of objects with True labels.
• Figure 4: Number of candidates per class for all the sources in the ZTF-ID list (top; 39,772,280 sources in total), and for the sources with a probability in the node_init $P_{init}\geq0.7$ (bottom; 37,988,187). The number of sources per class is shown on top of each bar.
• Figure 5: Normalized probabilities of the node_init (left), node_variable (center), and node_stochastic (right), for the four AGN classes. The red lines show the median probability for each class. The black lines show the 5th and 95th percentiles of the probabilities.