Hunting for Compact Object Binaries from eRASS1 Optical Counterparts through ZTF Time-domain Photometry and Multi-wavelength Surveys

A systematic census of compact object binary (COB) candidates, primarily X-ray binaries (XRBs), is conducted by integrating eROSITA X-ray data with ZTF time-domain photometry and multi-wavelength observations to uncover hidden populations. The study establishes two complementary pipelines that yield distinct source samples, demonstrating that coupling eROSITA with wide-field time-domain photometric and multi-wavelength surveys is a highly efficient strategy for discovering COBs.

The gist

Coupling eROSITA with wide-field time-domain photometric and multi-wavelength surveys offers a highly efficient strategy for uncovering the hidden population of COBs.

Sample Creation and Selection Processes

The research establishes two distinct samples based on different selection criteria to reduce contamination from extragalactic sources like AGNs. The first sample consists of 151 periodically variable sources selected based on optical periodic variations, while the second sample comprises 1958 distance-constrained sources chosen based on elevated X-ray luminosities or high log(FX/Fopt).

The selection for the first sample involves several steps:

- Identifying initial candidates using the X-ray main sequence defined by A. C. Rodriguez (2024) and retaining only those with a negative class gal exgal value and NWAY match flag = 1, yielding a sample of 13,631 sources initially.

- Searching for periodic modulation in the g and r band light curves using the Lomb-Scargle periodogram with a searched period range of 0.01-10 days.

- Applying a modified Line 2 selection criterion: log(FX/Fopt) = (GBP − GRP) − 3, which selects sources with more likelihood to be COBs.

The second sample is constructed using astrometric information from Gaia, requiring significant and reliable Gaia astrometric measurements, specifically retaining only sources with "Gaia PM SNR > 5 and Gaia PARALLAX SNR > 5. Distances are estimated from parallaxes, and the sample is further refined by requiring either LX > 1031 erg s−1 or log(FX/Fopt)−(GBP−GRP) > −3.0," resulting in a total of 1958 sources.

Multi-Wavelength Characterization

The samples are characterized using various multi-wavelength diagnostics to distinguish between CVs and LMXBs, and to identify radio counterparts.

- Distinguishing CVs from LMXBs involves exploring the use of log(FUV/FX) in combination with the X-ray hardness ratio (HR). A threshold at log(FUV/FX) = 0 serves as a reference line, where about 84% of non-magnetic CVs lie above this line and all confirmed LMXBs fall below it.

- The X-ray HR is defined differently for XMM-Newton and Chandra data. In the comparison sample, the X-ray HR derived from these datasets does not show a strong separation between CVs and LMXBs.

- Cross-matching with radio catalogs (VLASS and RACS) restricts the sample to sources flagged with the morphological classification “S” (single-component sources) to ensure they are point-like.

Final Candidate Identification

The final identification of compact object binaries is achieved through cross-matching both samples with radio catalogs. This process yields a final sample of seven compact radio point sources. Among these seven radio-detected sources, one is classified as a pulsar in SIMBAD, and the remaining four lack firm classifications and represent potential XRB candidates, which merit future follow-up observations. The entire selection process relies on the optical counterpart catalog built by M. Salvato et al. (2025), which provides reliable associations based on positional matching, photometric properties, colors, morphology, and parallax information.

Future Directions

The study suggests that further constraints are needed to fully classify the candidates. The reliability of isolating XRB candidates using log(FUV/FX) is limited due to potential contamination from magnetic CVs. Future work will require improved radio measurements and optical spectroscopic characterization to clarify the nature of these four systems. Additionally, future higher-cadence time-domain observations, together with complementary surveys such as LSST, will help better constrain orbital periods. The framework is noted as being readily extendable to other X-ray surveys and will become increasingly powerful with future data releases from eROSITA.

How it works

The framework operates by establishing two complementary pipelines: one based on optical periodic variations (Sample 1) and another based on distance constraints derived from Gaia astrometry (Sample 2).

Here are specific improvements for AI systems, derived from the methodologies and findings presented in this scientific paper:

The following improvements focus on leveraging the multi-wavelength data integration, statistical filtering techniques, and empirical diagnostic tools developed in this study to enhance astrophysical discovery and classification capabilities within AI systems.

1. Replace generic object classification models with a hierarchical pipeline that incorporates spectral energy distribution (SED) diagnostics:

2. Implement a two-stage candidate selection filter using both time-domain variability analysis (e.g., Lomb-Scargle periodogram on ZTF data) and distance/luminosity constraints derived from astrometry (Gaia parallaxes):

3. Develop an automated cross-matching algorithm that prioritizes optical counterparts based on probabilistic classification metrics (e.g., the gal exgal class or NWAY match flags) rather than relying solely on positional proximity:

4. Integrate a learned diagnostic tool, such as the empirical threshold for log(FUV/FX) = 0, to distinguish between Cataclysmic Variables (CVs) and Low-Mass X-ray Binaries (LMXBs), acknowledging the inherent limitations (e.g., potential magnetic CV contamination).

5. Create a specialized module for validating uncertain or ambiguously classified candidates by querying auxiliary data sources against positional error regions, specifically checking the photometric variability of neighboring sources in X-ray uncertainty contours:

6. Establish a radio-X-ray correlation verification layer that cross-references identified compact objects with radio catalogs (VLASS/RACS) and uses the derived LX–LR plane to validate interpretations (e.g., distinguishing between XRB, MSP, or AB candidates).

The improved AI system can perform the following specific functions:

1. Identify and prioritize potential Compact Object Binaries (COBs), specifically X-ray binaries (XRBs), by efficiently filtering the massive eROSITA catalog using a combination of periodic optical behavior and high X-ray luminosity/distance constraints.

2. Differentiate between stellar populations (CVs) and accreting compact objects (LMXBs) with higher confidence than current methods by analyzing the UV-to-X-ray flux ratio, while flagging systems that require further manual verification due to spectral overlap (magnetic CV contamination).

3. Automate the identification of periodic phenomena in optical light curves using time-domain photometry, significantly reducing the false positive rate associated with long or stochastic variability.

4. Generate a ranked list of high-priority COB candidates by iteratively applying selection criteria derived from both X-ray properties (e.g., Line 2 above the main sequence) and geometric constraints (Gaia distance measurements).

5. Perform robust, probabilistic optical counterpart association for X-ray sources by utilizing learned positional matching algorithms trained on multi-band photometric and morphological data from surveys like LS10, minimizing contamination from background AGN.

6. Validate the physical nature of radio-detected compact objects by mapping them onto the LX–LR diagram and assigning confidence scores based on alignment with known theoretical correlations (e.g., BH/NS correlation), thereby reducing ambiguity in identifying true XRB candidates versus other sources like RS CVn binaries or pulsars.