4FGL-Xiang-DR2 Gamma-Ray Catalog

• 4FGL-Xiang-DR2 is a high-precision gamma-ray source catalog that aggregates 16 years of Fermi-LAT data with enhanced elliptical localization for improved counterpart identification.
• The catalog employs a systematic three-step cross-matching pipeline with 5BZCAT_err and multi-wavelength statistical modeling to identify 17 new GeV blazar candidates.
• It refines positional uncertainties using a bivariate normal likelihood approach with binned likelihood analysis, ensuring robust source characterization and variability assessment.

4FGL-Xiang-DR2, usually abbreviated DR2, is an updated high-precision $\gamma$-ray source catalog constructed from 16 years of Fermi-LAT data and supplemented with refined positional uncertainties derived through elliptical localization. It was built together with a supplementary version of the fifth edition of Roma-BZCAT, denoted 5BZCAT_err, and the two catalogs were used in a systematic three-step pipeline for spatial cross-identification and multi-wavelength statistical modeling. In the associated analysis, this framework yielded 17 new GeV blazar candidates, with Fermi-LAT characterization indicating point-source morphology for all candidates, no apparent spectral curvature, and significant variability only for J1043.7+5323 (Xiang et al., 29 Aug 2025).

1. Catalog identity and placement within the Fermi-LAT catalog landscape

The designation 4FGL-Xiang-DR2 refers to a catalog distinct from the Fermi-LAT Collaboration’s 4FGL-DR2. In the Xiang catalog, DR2 denotes an updated release following an earlier DR1 that had used circular $r_{95}$ positional errors; in the Fermi-LAT Collaboration catalog, 4FGL-DR2 is an incremental 10-year release of the fourth LAT source catalog spanning 50 MeV to 1 TeV and containing 5,788 entries, with sources formally detected if $\mathrm{TS}>25$ (Ballet et al., 2020).

Within the Xiang release, Figure 1 is described as showing the global sky distribution of DR2 with 1,082 sources at $\mathrm{Signif}\geq 4\sigma$, compared against 4FGL-DR4, which is summarized there as 7,195 high-latitude and 1,405 low-latitude sources (Xiang et al., 29 Aug 2025). Candidate extraction in the blazar-identification study was restricted to DR2 sources with $\mathrm{Signif\_Avg}\geq 4\sigma$, so the catalog is used both as a source list and as the starting point for subsequent association analysis.

The nomenclature can therefore be a source of confusion. A plausible implication is that “DR2” in this context should always be read together with “Xiang” when the subject is refined localization, cross-matching to 5BZCAT_err, and identification of new GeV blazar candidates, whereas 4FGL-DR2 without the Xiang qualifier refers to the Fermi-LAT Collaboration release based on 10 years of data (Ballet et al., 2020).

2. Data foundation and likelihood analysis configuration

DR2 is based on 16 years of Fermi-LAT Pass 8 data, specifically P8R3_V3 with $\mathrm{IRF}=\mathrm{P8R3\_SOURCE\_V3}$, spanning MET 239557417 (2008-08-04) through MET 745981270 (2024-08-29) (Xiang et al., 29 Aug 2025). The event selection used $\mathrm{evclass}=128$ (SOURCE), $\mathrm{evtype}=3$ (Front+Back), zenith angle $<90^\circ$, and an energy range of 100 MeV–1 TeV. The analysis adopted binned likelihood fitting with 10 bins per decade and spatial pixels of size $0^\circ.1$.

The region-of-interest design was target-centered. A $r_{95}$0-radius ROI was adopted, while all 4FGL-DR4 and DR1 sources within $r_{95}$1 were included in the model; sources within $r_{95}$2 had normalization and spectral index free in the fit. Both the Galactic diffuse component, gll_iem_v07.fits, and the isotropic background, iso_P8R3_SOURCE_V3_v1.txt, were renormalized (Xiang et al., 29 Aug 2025).

These choices place DR2 within a standard LAT likelihood framework but with an explicitly localization-oriented configuration. The emphasis on free source parameters within the inner ROI and renormalized diffuse backgrounds is significant because the later cross-matching stage depends directly on the fidelity of source positions and positional covariance.

3. Positional uncertainty refinement and catalog content

A defining feature of DR2 is the replacement of DR1’s circular error radii by full elliptical localization. Each source was localized with Fermipy’s gta.localize() to derive a 95%-confidence uncertainty ellipse characterized by semi-major axis $r_{95}$3, semi-minor axis $r_{95}$4, and position angle $r_{95}$5 (Xiang et al., 29 Aug 2025). In practice,

$r_{95}$6

but the catalog retains the full five-parameter ellipse.

The localization likelihood around the best-fit position is approximated as a bivariate normal,

$r_{95}$7

where $r_{95}$8 is the covariance matrix encoding $r_{95}$9, $\mathrm{TS}>25$0, and $\mathrm{TS}>25$1 (Xiang et al., 29 Aug 2025). The final positions and ellipses were obtained through the iteration optimize() \rightarrow fit() \rightarrow localize() \rightarrow fit().

The catalog schema reflects this localization emphasis. The reported columns include the source designation (NAME), J2000 coordinates (RAJ2000, DEJ2000), 68% and 95% ellipse semi-axes, position angle, SpatialType, SpectrumType, Spectral_Index, Index_err, TS, Npred, band-limited photon fluxes (Flux100, Flux1000, Flux10000) and their errors, corresponding energy fluxes (Eflux100, Eflux1000, Eflux10000) and their errors, and TSvar, defined as the variability index computed from an annual-binned light curve (Xiang et al., 29 Aug 2025). The TS column is explicitly described as a detection Test Statistic, approximately significance squared.

This transition from circular to elliptical uncertainties is central to the catalog’s scientific role. It directly improves positional realism for source association, especially in workflows that propagate full covariance rather than a single radial containment value.

4. Cross-identification with 5BZCAT_err

The cross-matching component of the DR2 program used a companion catalog, 5BZCAT_err, produced by attaching NED-derived 95% error ellipses to the Roma-BZCAT entries (Xiang et al., 29 Aug 2025). The operational pipeline consisted of three steps: efficient retrieval of positional uncertainties and multi-wavelength fluxes from 5BZCAT_err, calculation of spatial association probabilities with DR2, and statistical modeling of multi-band fluxes.

Spatial cross-matching was performed with the Bayesian code nway, using an elliptical Gaussian position-uncertainty model plus an optical-magnitude prior (Xiang et al., 29 Aug 2025). For a candidate counterpart $\mathrm{TS}>25$2 at angular offset $\mathrm{TS}>25$3, the likelihood was written as

$\mathrm{TS}>25$4

where $\mathrm{TS}>25$5 is the prior from the $\mathrm{TS}>25$6-band magnitude. The posterior probability that 5BZCAT source $\mathrm{TS}>25$7 is the true match to DR2 source $\mathrm{TS}>25$8 is

$\mathrm{TS}>25$9

with $\mathrm{Signif}\geq 4\sigma$0 the “no-match” background term.

The adopted acceptance criteria were stringent: $\mathrm{Signif}\geq 4\sigma$1, $\mathrm{Signif}\geq 4\sigma$2, and $\mathrm{Signif}\geq 4\sigma$3 (Xiang et al., 29 Aug 2025). TOPCAT visual cross-checks confirmed all 17 matches. The combination of formal posterior thresholds and manual verification is important because the output of the pipeline is not merely a nearest-neighbor list, but a curated candidate set intended for astrophysical interpretation.

5. Statistical modeling of multi-wavelength flux distributions

The multi-wavelength component of the DR2 analysis retrieved broadband fluxes from 74 MHz to 100 GeV for 3,442 5BZCAT_err sources via astroquery.ned (Xiang et al., 29 Aug 2025). Each positive band flux $\mathrm{Signif}\geq 4\sigma$4 was transformed with a Box–Cox map,

$\mathrm{Signif}\geq 4\sigma$5

with $\mathrm{Signif}\geq 4\sigma$6 chosen by maximum-likelihood estimation.

After transformation, the flux distribution in each band was fit by a truncated normal on $\mathrm{Signif}\geq 4\sigma$7,

$\mathrm{Signif}\geq 4\sigma$8

where $\mathrm{Signif}\geq 4\sigma$9 and $\mathrm{Signif\_Avg}\geq 4\sigma$0 are the standard normal PDF and CDF, $\mathrm{Signif\_Avg}\geq 4\sigma$1 and $\mathrm{Signif\_Avg}\geq 4\sigma$2 are MLE-fitted, and $\mathrm{Signif\_Avg}\geq 4\sigma$3 are the minimum and maximum of the transformed data (Xiang et al., 29 Aug 2025). Goodness of fit was verified by Kolmogorov–Smirnov tests, with all $\mathrm{Signif\_Avg}\geq 4\sigma$4 after removal of 7% outliers at 74 MHz via a Local Outlier Factor search.

This statistical layer was used to assess consistency between the 17 cross-matched candidates and known blazar populations. The reported result is that 15 of the 17 sources fall within the $\mathrm{Signif\_Avg}\geq 4\sigma$5 confidence interval of the distribution model, demonstrating strong statistical consistency with known blazar samples (Xiang et al., 29 Aug 2025). The paper further states that, as Fermi-LAT data continues to accumulate, the remaining two sources are expected to converge toward the high-confidence region, which it interprets as further support for the common-source hypothesis.

6. Source characterization, variability, and public data products

The 17 candidates were subjected to additional LAT characterization. Point-source status was checked with extension(), yielding $\mathrm{Signif\_Avg}\geq 4\sigma$6 for all candidates, and spectral curvature was tested with curvature(), yielding $\mathrm{Signif\_Avg}\geq 4\sigma$7; a Power-Law model was therefore adopted for the sample (Xiang et al., 29 Aug 2025). Their cross-match TS maps were visually inspected as part of candidate validation.

Variability was evaluated from light curves above 100 MeV, binned in 16 yearly intervals. The variability index was defined as

$\mathrm{Signif\_Avg}\geq 4\sigma$8

where $\mathrm{Signif\_Avg}\geq 4\sigma$9 is the likelihood in bin $\mathrm{IRF}=\mathrm{P8R3\_SOURCE\_V3}$0 (Xiang et al., 29 Aug 2025). A 99%-confidence variability threshold of $\mathrm{IRF}=\mathrm{P8R3\_SOURCE\_V3}$1 was adopted. Only J1043.7+5323 exceeded this threshold, with $\mathrm{IRF}=\mathrm{P8R3\_SOURCE\_V3}$2, and that variability was confirmed with 4-month bins, for which $\mathrm{IRF}=\mathrm{P8R3\_SOURCE\_V3}$3. The remaining 16 candidates were steady, with $\mathrm{IRF}=\mathrm{P8R3\_SOURCE\_V3}$4.

The associated abstract summarizes the sample in equivalent terms: except for J1043.7+5323, which exhibits significant flux variability above 100 MeV, the remaining sources display stable fluxes, without apparent spatial extension or spectral curvature (Xiang et al., 29 Aug 2025). This is relevant for classification because the candidate set is being advanced specifically as GeV blazar candidates rather than as a generic class of new LAT detections.

The DR2 project also provides substantial auxiliary material. The full DR2 FITS file, multi-wavelength flux tables, TS maps, SEDs, and LC products are reported as publicly available through the China-VO PaperData repository (Xiang et al., 29 Aug 2025). For research use, this makes DR2 not only a catalog but also a reproducible analysis package centered on localization, association, and multi-band statistical consistency.