Searching for unresolved massive black hole pairs through AGN photometric variability

Abstract

Since their discovery, AGN light curves are known to be intrinsically variable. In the optical/UV band, this variability is consistent with correlated or red noise and is particularly well described by the damped random walk (DRW) model. In this work, we evaluate the feasibility of a new method for identifying spatially unresolved couples of AGN through a fully Bayesian time-domain analysis of the observed light curves (LCs). More specifically, we check whether observed LCs are better described by a single DRW, which we interpret as emitted by a single massive black hole (MBH), or a pair of independent DRWs, generated by a pair of MBHs. We test the method on mock LCs associated with a single MBH and pairs generated with different cadences and lengths of observational campaigns. We constrained the occurrence of false positives, that is, the percentage of single MBH LCs that show substantial evidence in favour of the unresolved MBH pair scenario, finding a fraction of 0.2% and 0.59% in the even and uneven sampling scenarios. We discuss how well the method recovers the model parameters, showing that about 51% and 7% of the simulated LCs have all the recovered parameters within 20% of their true values in our best scenario of evenly sampled LCs for the single MBH and MBH pair scenarios, respectively. We finally study the region of the parameter space in which the detection of an MBH pair is possible, finding that such objects can be correctly identified if the timescales of the process describing the noise are very different, with a ratio smaller than ~0.2, and the variability amplitudes are similar, with their ratio bigger than ~0.2. When limiting to such a region of the parameter space, the fraction of pairs with all the recovered parameters within 20% of the injected values increases up to about 14% and 8% for evenly and unevenly sampled LCs, respectively.

Figures (9)

• Figure 1: Examples of light curves generated using celerite varying the process parameters: the damping timescale $\tau$, panel (a), and the variability amplitude $\hat{\sigma}$, panel (c). In the right panels, the retrieved periodograms (computed as discussed in Appendix \ref{['app:periodogram_computation']}) as well as the theoretical expected PSD (dashed lines) are shown. The time series ($\hat{F}$) are centred at zero and have arbitrary units, the parameter $\sigma$ and power ($P$) should be considered in terms of the same dimensional scale as $\hat{F}$, with $P$ having units of the square of the units of $\hat{F}$.
• Figure 2: Distribution of relative errors for the retrieved parameters in the single MBH evenly sampled
• Figure 3: Distribution of relative errors for the retrieved parameters in the UBHD evenly sampled case. In green, the distribution for the cases in which the UBHD model is favoured.
• Figure 4: Map of Bayes factors for 1500 unevenly sampled realisations of UBHD light curves with a 10-year baseline. The colour scale represents the base-10 logarithm of the Bayes factor, while the axes show the ratio of the model parameters between the two light curves. The red square, diamond, triangle and circle correspond to the light curves used to compute the periodograms shown in panels a,b,c, and d of Figures \ref{['fig:double_kernel_PSD']} and \ref{['fig:PSD_four_panel']}, respectively.
• Figure 5: Same kind of map as reported in Figure \ref{['fig:even_3000']} for 580 light curves in the region of the parameter space where the UBHDs can be correctly identified.