Computer models suggest that searches for active supermassive binary black holes could produce hundreds of candidates from individual signals, but far fewer systems would show both signs at once. Across the reported scenarios, the model produced up to 318 candidates with a spectral energy distribution, or SED, deficit, while periodic-light-curve tests returned about 135 to 810 systems for Doppler-driven variability and 254 to 1,323 for hydrodynamic variability.
The authors generated a mock population by Monte Carlo sampling across redshift, mass ratio, accretion rate, black-hole mass and separation, matching the number of mock systems to their model-estimated cosmic counts. The runs combined two choices for the black-hole-host scaling relation with two accretion prescriptions, producing four model combinations. The analysis also included simulated single-AGN controls.
The signals do not pick out the same systems
For the SED test, a candidate had to meet thresholds for changes between adjacent and non-adjacent wavelength bands. The adjacent-band threshold was 1.0, the non-adjacent threshold was 1.2, and the continuity cutoff was above -0.1. The resulting candidates clustered in a narrow part of the modeled population, with separations generally between 100 and 1,000 gravitational radii, mass ratios at or below 0.1, and periods usually shorter than 30 years.
The periodic screen used a V-band variability threshold of at least 0.05 magnitude and an observer-frame period of about 5 years or less. The authors describe that test as idealized and not completeness corrected, so its totals do not represent the recovery performance of a particular survey. In the model, periodic selections were more numerous than SED selections, but the exact totals changed with the adopted scaling and accretion models.
False alarms were a major qualification. In the fiducial single-AGN noise control, the SED test's false-alarm rate was below 0.001%. Relaxing the adjacent-band threshold to 0.5 raised that estimate to 0.37%, while adding 5% asynchronous variability could push it to 13%. Applying the index-decline continuity control brought the rate below 0.05%. Dust reddening changed the false-positive rate by less than 5%, and adding X-ray reprocessing changed the selected counts by less than 1%.
Requiring both kinds of evidence produced a much smaller, more selective set. In the paper's labels, DopVar refers to Doppler-driven variability and HydroVar to hydrodynamic variability. Under the redshift-dependent accretion model, joint counts were 9 DopVar-plus-SED systems and 19 HydroVar-plus-SED systems in BBH-nonevol, rising to 42 and 75 in BBH-zevol. Only about 7% to 9% of periodic-variable systems also showed an SED deficit. Looking the other way, 37% to 53% of SED-deficit systems were HydroVar and 20% to 26% were DopVar.
The authors' strongest argument for combining the tests comes from contamination. Assuming the two false-alarm processes were statistically independent, they estimated a joint false-positive rate of about 10^-7 from individual rates of 0.05% for the SED test and at most 0.02% for the periodic test. That is an assumption-based forecast, not a measured rate, and the analysis notes that independence may fail for all physical contaminants.
The forecast moves with the assumptions
The choice of black-hole-host scaling relation had a large effect. Using the redshift-dependent relation increased the predicted number of individual SED-deficit or periodic detections by roughly 4.5 to 5.6 times, and increased joint detections by about 3.5 to 4.7 times. The reported counts therefore describe a range of model outcomes, not a single expected discovery total.
For pulsar-timing arrays, the model forecast about four SED-deficit systems meeting a signal-to-noise ratio above 3 for CPTA and about 15 for SKA-PTA in 20-year observations. None of the jointly selected systems with periods below 5 years was expected to be detected by either array. These figures are idealized forecasts tied to the assumed PTA configurations and simulated populations.
Wavelength coverage changed the modeled yield as well. In an alternative setup using JWST NIRCam-like bands from 0.70 to 2.77 micrometres, the model returned 193 candidates for BBH-nonevol and 969 for BBH-zevol, with the latter population peaking around redshift 3. The comparison suggests that results depend not only on the binary model but also on which parts of the spectrum a search can sample.
A forecast still waiting for real-world tests
The practical idea is to combine the two screens: a spectral deficit can nominate one set of systems, while periodic variability supplies a different and partly overlapping set. The authors interpret the combination as a more reliable candidate filter because its modeled contamination is very low under the independence assumption. Because the study is built from semianalytic mock populations, it is a forecast for choosing targets, not a report of confirmed binaries. Actual yields will depend on how these assumptions perform in survey data.
The manuscript is labeled a preprint dated 27 August 2026 and lists arXiv:2608.25507v1 dated 26 August; its acceptance and receipt fields remain placeholders. The article says the underlying data will be shared on reasonable request to the corresponding author.
Paper data and sources
Original title: A Population Study for Searching Supermassive Binary Black Holes in Active Galactic Nuclei: Continuum Spectral Features and Periodic Variabilities
Authors: Zekun Li, Changshuo Yan, Youjun Lu
Journal/Repository: Monthly Notices of the Royal Astronomical Society, Volume 551, Issue 1, September 2026
Status: Peer-reviewed
First online: 2026-08-26
DOI: 10.1093/mnras/stag1409
Original paper · Full text