---
title: AGN Flare Coarse Catalog (AGNFCC)
url: https://www.emergentmind.com/topics/agn-flare-coarse-catalog-agnfcc
type: topic
---

# AGN Flare Coarse Catalog (AGNFCC)

Searching arXiv for the AGNFCC paper and closely related catalog papers to ground the article in current literature.
First, I’ll look up the AGNFCC-defining paper.
The AGN Flare Coarse Catalog (AGNFCC) is an inclusive event catalog of flare-like optical outbursts in active galactic nuclei and AGN candidates, constructed from six years of Zwicky Transient Facility Data Release 23 light curves. In its published AGNFCC-V1.0 form, it contains 28,504 flare events selected from a parent sample of 1,966,641 sufficiently sampled AGNs and AGN candidates, and it is explicitly paired with the smaller high-purity AGN Flare Refined Catalog (AGNFRC) of 1,984 events [2507.20232]. AGNFCC is designed as a broad discovery layer rather than a final adjudication of physical origin: it captures statistically significant brightenings that are unlikely under a fitted stochastic baseline model, while preserving many events that may later prove to be blazar activity, supernovae in AGN hosts, tidal disruption events, microlensing episodes, changing-look behavior, or rare manifestations of intrinsic AGN variability [2507.20232].

## 1. Survey foundation and parent-sample construction

AGNFCC is built on ZTF DR23, using the regular-cadence \(g\) and \(r\) bands over an effective six-year baseline. DR23 includes all public-survey observations through 2024 October 31 and private-survey observations through 2023 June 30; the analysis uses only \(g\) and \(r\), excluding \(i\) because of insufficient sampling and limited coverage [2507.20232]. For declinations \(>-31^\circ\), ZTF’s public programs are described as the Galactic Plane Survey for \(|b|<7^\circ\) at about 1-day cadence and the Northern Sky Survey for \(|b|>7^\circ\) at about 3-day cadence [2507.20232].

The AGN parent sample is assembled by merging five external AGN or AGN-candidate catalogs: Milliquas v8 with 1,021,800 sources, the DESI DR1 AGN/QSO subset with 1,772,694, the LAMOST AGN catalog with 55,636, Quaia with 1,295,502, and the WISE AGN R90 catalog with 4,543,530 [2507.20232]. A \(1^{\prime\prime}\) deduplication yields 6,602,925 unique AGNs and AGN candidates, and a subsequent \(3^{\prime\prime}\) positional cross-match to ZTF DR23 produces about 3.57 million matched light curves [2507.20232]. Requiring more than 30 good observations per band after removing epochs with `catflags > 0` leaves 1,966,641 sources with adequate sampling in both \(g\) and \(r\) [2507.20232].

This parent-sample design is one of AGNFCC’s defining properties. It is broader than a pure spectroscopic AGN catalog, because it deliberately includes AGN candidates from heterogeneous selection channels, but it is narrower than an all-nuclear-transient search, because the search space is explicitly limited to previously cataloged AGNs or AGN candidates [2507.20232]. A plausible implication is that AGNFCC prioritizes contextualized nuclear transients over completely unconstrained discovery.

## 2. Detection logic: Bayesian blocks, Gaussian processes, and dual-band significance

The flare-finding pipeline operates in flux space. Each raw light curve contains heliocentric MJD, PSF magnitude, \(1\sigma\) photometric uncertainty, and `catflags`; after filtering, the light curves are cleaned with a hybrid sigma-clipping procedure intended to suppress isolated noise while preserving candidate bursts [2507.20232]. Points deviating by more than \(3\sigma\) from the median flux are flagged as outliers, but single isolated outliers are removed whereas two or more consecutive outliers, or isolated outliers surrounded by neighboring points at \(>2.5\sigma\), are retained as possible flare structure [2507.20232]. The cleaned light curves are then rebinned into 3-day intervals with inverse-variance weighting [2507.20232].

Candidate intervals are first identified with Bayesian blocks. The block decomposition is searched with a hill-climbing procedure in which a peak is a block higher than both neighboring blocks, and the peak is extended in both directions as long as subsequent blocks remain lower [2507.20232]. A flare interval is then defined as the portion of that elevated block sequence where the flux exceeds 0.2 times the difference between the peak flux and the median flux of the full light curve; intervals containing only one data point are discarded [2507.20232].

Ordinary stochastic AGN variability is modeled with a Gaussian process whose mean function is the light-curve median and whose covariance uses the Matérn-1/2, or exponential, kernel
\[
k(t,t') = \rho^{2}\exp\left(-\frac{|t-t'|}{\tau}\right),
\]
with covariance matrix
\[
K_{ij}=k(t_{i},t_{j})+\delta_{ij}\sigma_{i}^{2},
\]
and marginal log-likelihood
\[
\log \mathcal{L} = -\frac12(\boldsymbol{y}-\boldsymbol{m})^{T}\boldsymbol{K}^{-1}(\boldsymbol{y}-\boldsymbol{m})-\frac12 \log |\boldsymbol{K}| - \frac N2\log 2\pi
\]
[2507.20232]. The hyperparameters \(\rho\) and \(\tau\) are optimized by maximum likelihood using `celerite` [2507.20232].

Each Bayesian-block candidate is then tested against the fitted GP baseline through a deviation statistic \(\lambda\), and its significance is estimated by generating many mock light curves with the same cadence and errors as the real data from the fitted GP null model [2507.20232]. The resulting significance, \(p_{\rm flare}\), is defined as the probability that the observed flare is not attributable to intrinsic AGN variability, and AGNFCC adopts the threshold
\[
p_{\rm flare} > 0.998
\]
in both \(g\) and \(r\), using the weaker of the two bandwise significances as the catalog significance [2507.20232]. This dual-band requirement is a central quality-control feature: AGNFCC is not a single-band excursion catalog.

## 3. Catalog structure, event parameters, and the relation to AGNFRC

AGNFCC is intentionally permissive. AGNFRC is the stricter subset obtained by additional cuts on long observational gaps, peak significance, non-flare excesses, BIC improvement for an explicit flare model, flare coverage, cadence within the modeled flare window, amplitude contrast, and source confusion [2507.20232]. The conceptual distinction is simple.

| Catalog | Selection posture | Size |
|---|---|---:|
| AGNFCC | Inclusive discovery catalog | 28,504 flares |
| AGNFRC | High-confidence refined subset | 1,984 flares |

The per-flare schema includes `AGN_name`, `RA`, `DEC`, redshift `z`, parent-catalog code `source`, source `type`, `ALeRCE` classification where available, flare peak time `t0`, exponential decay timescale `te`, Gaussian rise timescale `tg`, baseline flux densities `r0_g` and `r0_r`, flare amplitudes `A_g` and `A_r`, `max_gap`, `peak_sigma_g`, `peak_sigma_r`, `excess_num_g`, `excess_num_r`, `delta_bic`, `flare_point_g`, `flare_point_r`, and eight boolean flags tied to the refined-catalog veto logic [2507.20232].

The rise and decay parameters are tied to an explicit asymmetric flare model,
\[
y(t) = \left\{ \begin{aligned}
&r_{0} + A\exp\left(-\frac{(t-t_{0})^{2}}{2t_{g}^{2}}\right), \quad t \leq t_{0} \\
&r_{0} + A\exp\left(-\frac{(t-t_{0})}{t_{e}}\right), \quad t \ge t_{0},
\end{aligned} \right.
\]
where \(r_0\) is the baseline flux, \(A\) the amplitude, \(t_0\) the peak time, \(t_g\) the Gaussian rise timescale, and \(t_e\) the exponential decay timescale [2507.20232]. These are characterization parameters rather than the primary detection statistic.

The eight refinement flags encode the major AGNFRC rejection criteria: `flag_1` marks `max_gap > 500 d`; `flag_2` minimum peak significance \(<3\); `flag_3` too many non-flare \(>3\sigma\) points; `flag_4` \(\Delta \mathrm{BIC}<10\); `flag_5` flare window not fully covered; `flag_6` too few points inside the flare window; `flag_7` amplitude ratio \(<0.2\); and `flag_8` source confusion [2507.20232]. Because AGNFCC retains these flags, it can be re-filtered into custom purity-completeness regimes without re-running the full search.

A notable feature is that AGNFCC allows multiple flares per AGN. This matters because AGNFRC’s stricter logic was applied assuming at most one significant flare per AGN, so recurrent systems can survive in AGNFCC while being excluded from AGNFRC [2507.20232].

## 4. Event demographics and astrophysical content

AGNFCC is explicitly heterogeneous. The paper identifies associations with known supernovae, tidal disruption events, and blazars, and notes that a few events may be linked to binary black hole mergers or microlensing [2507.20232]. Cross-matching to Roma-BZCAT plus Milliquas blazar flags yields 830 blazars in AGNFCC, about 3% of the coarse catalog, while cross-matching to the Transient Name Server identifies 71 supernovae and 2 TDE matches [2507.20232]. This composition is a reminder that AGNFCC is not a single-phenomenology catalog.

The spatial distribution follows the survey footprint and the AGN parent distribution, being roughly uniform above declination \(-30^\circ\) and outside the Galactic plane [2507.20232]. In redshift, AGNFCC is more concentrated toward lower redshift than the full well-sampled AGN sample, which the paper interprets as a contrast-selection effect: nearby and lower-luminosity AGNs make flares easier to detect against the baseline, whereas high-redshift luminous AGNs suppress fixed-luminosity flare contrast [2507.20232].

The flare-characterization fits yield rise and decay timescales spanning from a few days to several hundred days, with most flares having \(t_g\) and \(t_e < 50\) days [2507.20232]. The paper cautions that the shortest timescales are uncertain because of the 3-day binning [2507.20232]. Morphologically, both fast-rise/slow-decay and slower-rise/faster-decay events occur [2507.20232].

Host-type demographics are also informative. In the Million Quasars classification, QSOs dominate the full parent sample at more than 80%, but their fraction decreases in the flare catalogs, while sources classified simply as AGN and narrow-line AGN increase in AGNFCC; BL Lac objects become more prominent especially in AGNFRC [2507.20232]. A plausible implication is that the refined subset preferentially retains cleaner jet-dominated or otherwise morphologically distinctive events, while the coarse sample is more permissive of complex narrow-line systems with excess variability.

External classification overlap is consistent with this interpretation. About 40% of AGNFCC flares have an ALeRCE alert-stream classification match, rising to 70% in AGNFRC [2507.20232]. The paper treats this as empirical support for the higher reliability of the refined subset.

## 5. Position within the broader AGN-catalog ecosystem

AGNFCC is best understood as one layer in a larger AGN-catalog infrastructure. Its own parent sample already includes the WISE AGN R90 catalog, which contains 4,543,530 AGN candidates selected across 30,093 deg\(^2\) and was designed for 90% reliability, alongside the much larger C75 catalog of 20,907,127 candidates optimized for 75% completeness [1706.09901]. Mid-infrared AGN priors are especially valuable because they recover obscured systems that optical AGN selections can miss.

Hard X-ray AGN catalogs supply a complementary prior. BASS DR2 provides a nearly all-sky, spectroscopically complete census of 858 hard-X-ray-selected AGNs from the Swift/BAT 70-month survey, with vetted counterparts, redshifts, classifications, and derived black hole and accretion properties; for AGNFCC workflows its primary utility is as a robust “known AGN” backbone rather than as a flare source [2207.12432]. Because BAT selection is comparatively insensitive to obscuration up to Compton-thick levels, it is particularly useful when evaluating nuclear flares in hosts that look optically quiescent but may harbor hidden AGN activity [2207.12432].

Jet-dominated high-energy catalogs define another important comparison class. The Fourth LAT AGN Catalog contains 2,863 high-latitude AGNs and shows that FSRQs are on average softer and more variable than BL Lacs in the gamma-ray band, with variability indices derived from 1-year and 2-month light curves [1905.10771]. The 2026 preliminary Fermi-LAT multi-timescale variability catalogue extends this source-level variability description to 3-day, 7-day, and 30-day cadences for 1,429 AGN light curves, but it remains a variability catalogue rather than a flare-event catalogue [2602.01597]. For AGNFCC, such gamma-ray resources are best used as flare-proneness priors for blazar-dominated subsets.

Within optical time-domain work, the CAZ catalog provides a complementary blazar-centered framework: 7,918 blazar-selected AGN with nightly optical light curves from CRTS, ATLAS, and ZTF between 2007 and 2023, plus Bayesian-block-derived flaring periods and source-level flare statistics [2510.16584]. CAZ is more specialized toward jet-dominated AGN, whereas AGNFCC is broader in parent-sample construction but tied specifically to ZTF-based flare discovery [2510.16584].

## 6. Limitations, interpretive cautions, and common misconceptions

AGNFCC is not a pure physical-classification catalog. Its principal contamination source is precisely the phenomenon it is designed to transcend: rare but genuine intrinsic AGN variability [2507.20232]. The paper explicitly warns about look-elsewhere false positives in nearly two million light curves, source blending from the \(3^{\prime\prime}\) ZTF cross-match, photometric artifacts surviving preprocessing, and model misspecification arising from the use of a simple GP kernel for real AGN variability [2507.20232]. The absence of a flare in AGNFRC does not mean the corresponding AGNFCC event is unphysical; it means only that the event fails stricter morphological or coverage cuts.

A second misconception is to treat AGNFCC as complete. It is incomplete for flares during long gaps, events not clearly visible in both \(g\) and \(r\), very rapid outbursts blurred by 3-day binning, and high-redshift or high-luminosity AGNs where flare contrast is weak [2507.20232]. The paper also does not provide classical precision, recall, ROC curves, or injection-recovery completeness grids, and it does not explicitly state the number of unique host AGNs represented by the 28,504 flares [2507.20232]. These absences are material for rate inference.

A third misconception is to equate optical non-detection with nuclear inactivity. The mid-infrared flare in the Type 2 AGN SDSS J165726.81+234528.1 brightened by 3 mag in WISE \(W1\) and \(W2\) between 2015 and 2017 while showing no significant optical changes larger than about 0.2 mag, illustrating that dramatic nuclear transients can be optically hidden by dust geometry and reprocessing [1907.12721]. AGNFCC’s ZTF basis therefore makes it powerful for optical flares but not exhaustive for obscured accretion-state changes.

The most appropriate interpretation is consequently layered. AGNFCC is a broad discovery catalog of statistically unusual optical brightenings in known or candidate AGN hosts; AGNFRC is the high-confidence subset; and external AGN reference catalogs such as WISE AGN, BASS DR2, and 4LAC supply source-level context needed to decide whether a given event is best understood as a jet flare, a disk-state change, a supernova in an AGN host, a TDE-like phenomenon, microlensing, or an extreme realization of stochastic AGN variability [1706.09901] [2207.12432] [1905.10771].

Source: https://www.emergentmind.com/topics/agn-flare-coarse-catalog-agnfcc