Active galactic nuclei do not exhibit strictly sinusoidal brightness variations

Periodic variability in active galactic nuclei (AGN) light curves has been proposed as a signature of close supermassive black hole (SMBH) binaries. Recently, 181 candidate SMBH binaries were identified in Gaia DR3 based on apparently stable sinusoidal variability in their $\sim$1000-day light curves. By supplementing Gaia photometry with longer-baseline light curves from the Zwicky Transient Facility (ZTF) and the Catalina Real Time Transient Survey (CRTS), we test whether the reported periodic signals persist beyond the Gaia DR3 time window. We find that in all 116 cases with available ZTF data, the Gaia-inferred periodic model fails to predict subsequent variability, which appears stochastic rather than periodic. The periodic candidates thus overwhelmingly appear to be false positives; red noise contamination appears to be the primary source of false detections. We conclude that truly periodic and sinusoidal AGN variability is exceedingly rare, with at most a few in $10^6$ AGN exhibiting it on 100 to 1000 day timescales. Models predict that the Gaia AGN light curve sample should contain dozens of true SMBH binaries with periods within the observational baseline, so the lack of strictly periodic light curves in the sample suggests that most short-period binary AGN do not have light curves dominated by simple sinusoidal periodicity.

💡 Research Summary

This paper critically evaluates the claim that 181 active galactic nuclei (AGN) identified in Gaia Data Release 3 (DR3) exhibit stable sinusoidal brightness variations on timescales of roughly 1000 days, a signature that could indicate close supermassive black‑hole (SMBH) binaries. The authors augment the Gaia G‑band light curves with much longer‑baseline photometry from the Zwicky Transient Facility (ZTF) and, where available, the Catalina Real‑Time Transient Survey (CRTS). Of the 181 candidates, 116 have sufficient ZTF coverage (median 453 r‑band and 316 g‑band points spanning ~2350 days, or ~3650 days when combined with Gaia).

First, the Gaia light curves are refitted with a simple sinusoidal model using MCMC, reproducing the periods reported by Huijse et al. (2025). The best‑fit sinusoid is then extrapolated forward to the ZTF and CRTS epochs. In virtually every case the observed ZTF (and CRTS, when available) fluxes deviate strongly from the sinusoidal prediction, displaying stochastic, red‑noise‑like variability rather than coherent periodicity. A detailed example (Gaia DR3 3870752104863883264) shows a good sinusoidal match within the Gaia window but a complete breakdown of the model in the subsequent ZTF data.

To quantify any lingering periodic signal, the authors compute Lomb–Scargle periodograms for the unbinned ZTF light curves, identify the peak power frequency, and refit a sinusoid (including an extra scatter term). Although >90 % of the ZTF periodograms yield formal false‑alarm probabilities (FAP) < 10⁻³ under a white‑noise assumption, the peak periods are generally inconsistent with the Gaia periods. Only three objects have Gaia and ZTF periods consistent within 1σ, and another three within 2σ; random shuffling of periods predicts a comparable number of coincidences, indicating that the apparent agreement is statistical noise. Moreover, the power at the Gaia‑derived period in the ZTF periodograms is indistinguishable from that obtained when periods are randomly reassigned, reinforcing the conclusion that the Gaia periods do not survive the longer baseline.

These findings imply that the Bayesian model comparison employed by Huijse et al.—which contrasted a rigid sinusoidal model against a damped random walk (DRW) stochastic model—substantially underestimates the false‑positive rate when only a few cycles are sampled. The DRW model, while widely used to describe AGN variability, appears insufficiently flexible to capture the full red‑noise behavior that can masquerade as periodicity on short baselines. Consequently, the purity of the Gaia‑selected sample is low despite the conservative Bayes‑factor threshold (log BPR > 3.0423).

The broader astrophysical implication is that truly sinusoidal, strictly periodic AGN variability is exceedingly rare: at most a few per million AGN on 100–1000 day timescales. This scarcity challenges optimistic predictions that the Gaia AGN sample should contain dozens of genuine SMBH binaries with observable periods. It also suggests that short‑period binary AGN, if they exist, likely produce more complex, non‑sinusoidal light‑curve signatures (e.g., modulated amplitudes, phase drifts, or additional stochastic components).

In summary, the paper demonstrates the necessity of multi‑survey, long‑baseline validation for periodicity claims in AGN, highlights the limitations of simple DRW stochastic models in estimating false‑positive rates, and calls for more sophisticated statistical frameworks that incorporate realistic red‑noise processes when searching for SMBH binary signatures in time‑domain data.