Variability in the supermassive black hole binary candidate SDSS J2320+0024: No evidence of periodic modulation

TL;DR

The paper tests the claim that SDSS J2320+0024 hosts a supermassive black hole binary by distinguishing genuine periodicity from red-noise variability in its light curve. It adopts a fully Bayesian Gaussian-process framework, comparing DRW, modified DRW, quasi-periodic, and purely periodic kernels, and uses nested sampling to compute Bayesian evidences. The results show no robust preference for a periodic signal over red-noise models; a short-timescale peak near ~290 days in the QPO model is broad and not decisively significant, and a strictly periodic model is not favored. The authors conclude that, while this weakens the binary interpretation based on photometric periodicity alone, it does not rule out SBHB scenarios and advocate joint photometric-spectroscopic analyses and further monitoring to clarify the nature of J2320+0024.

Abstract

Supermassive black hole binaries (SBHBs) are a natural outcome of galaxy mergers, and they are expected to be among the loudest gravitational-wave sources at low frequencies. The source SDSS J2320+0024 was recently proposed as a promising SBHB candidate due to a possible periodicity in its light curve and variability in the MgII emission line. In this work, we reanalysed the optical (g, r, and i bands) light curves of J2320+0024 within the framework of Bayesian model selection. When periodicity was searched for together with red noise, analysis of the g-band light curve reveals a peak in the posterior of the period at ~290 days. The posterior profile is too broad to yield a preference for periodic models over models that include only red noise. Furthermore, the same peak is not present in the analysis of the r-band and i-band light curve. A periodic model without red noise identified a different (~1100 days) periodicity, but this model is statistically significantly disfavoured relative to the other models tested. In summary, we find no significant evidence in favour of a true periodic signal over red-noise variability. Our analysis questions the robustness of the previously proposed periodicity and emphasises the importance of rigorous statistical treatment. While our findings challenge the binary interpretation for J2320+0024, they do not rule it out. A statistically robust joint analysis of the photometric light curves and evolving broad-line profiles would shed further light on the true nature of this object.

Figures (7)

• Figure 1: Best-fit to the g-band light curve of J2320+0024. Black points represent the data with their errors, the solid blue line and shaded area show the median, 16th, and 84th percentiles of the posterior distributions, and the red dashed line indicates the maximum-likelihood model. From top to bottom: DRW (a), modified DRW (b), QPO (c), and Periodic (d) . See Table \ref{['tab:parameters']} for the estimated best-fit parameters.
• Figure 2: Corner plot of the posteriors for the different models. The DRW model is shown in blue, modified DRW in red, QPO in green, and Periodic in orange. Contours are drawn to include 90% of the posterior probability. The dashed black lines indicate the period reported in Fatovic2025. For clarity, although the fit is performed on standardised data, the marginal posterior distribution of the period is shown in units of $\log_{10}$ days.
• Figure 3: Best-fit for the r-band light curve of J2320+0024. The black points represent the data with their errors, the blue solid line and the shaded area represent the median, 16th, and 84th percentiles from the posterior distributions, and the red dashed line refers to the maximum likelihood model. From top to bottom: DRW (a), modified DRW (b), QPO (c), and Periodic (d), we refer to Tab. \ref{['tab:parameters']} for the estimated best-fit parameters.
• Figure 4: Corner plot of the posteriors of the different models. DRW is blue, modified DRW is red, QPO is green, and Periodic is orange. Contours are drawn at a level to include 90% of the posterior probability, while the black dashed lines refer to the period reported in Fatovic2023. For reasons of clarity, although the fit is performed on standardised data, we report the marginal posterior distribution of the period in units of $\log_{10}$ Days. Please note the different range of $\log_{10}{T}$ in the corner plot. No short ($\simeq273$ days) oscillations are identified by the nested sampling algorithm.
• Figure 5: Best-fit for the i-band light curve of J2320+0024. The black points represent the data with their errors, the blue solid line and the shaded area represent the median, 16th, and 84th percentiles from the posterior distributions, and the red dashed line refers to the maximum likelihood model. From top to bottom: DRW (a), modified DRW (b), QPO (c), and Periodic (d), we refer to Tab. \ref{['tab:parameters']} for the estimated best-fit parameters.