Measuring X-ray variability in faint/sparsely-sampled AGN
We study the statistical properties of the Normalized Excess Variance of variability process characterized by a red-noise power spectral density (PSD), as the case of Active Galactic Nuclei (AGN). We perform Monte Carlo simulations of lightcurves, assuming both a continuous and a sparse sampling pattern and various signal-to-noise (S/N) ratios. We show that the normalized excess variance is a biased estimate of the variance even in the case of continuously sampled lightcurves. The bias depends on the PSD slope and on the sampling pattern, but not on the S/N ratio. We provide a simple formula to account for the bias, which yields unbiased estimates with an accuracy better than 15%. We show that the normalized excess variance estimates based on single lightcurves (especially for sparse sampling and S/N less than 3) are highly uncertain (even if corrected for bias) and we propose instead the use of an ensemble estimate, based on multiple lightcurves of the same object, or on the use of lightcurves of many objects. These estimates have symmetric distributions, known errors, and can also be corrected for biases.We use our results to estimate the ability to measure the intrinsic source variability in current data, and show that they could also be useful in the planning of the observing strategy of future surveys such as those provided by X-ray missions studying distant and/or faint AGN populations and, more in general, in the estimation of the variability amplitude of sources that will result from future surveys such as Pan-STARRS, and LSST.
💡 Research Summary
Active Galactic Nuclei (AGN) exhibit strong X‑ray variability that is often quantified using the Normalized Excess Variance (NXS). While NXS is widely employed because it is simple to compute, its statistical reliability—especially for faint sources or sparsely sampled light curves—has not been rigorously tested. In this paper the authors address this gap by performing extensive Monte‑Carlo simulations of AGN light curves whose underlying variability follows a red‑noise power‑spectral density (PSD) of the form P(f) ∝ f^–β, with β ranging from 1 to 2, a regime typical for AGN.
Two sampling schemes are explored. The first is a continuous, evenly spaced observation with no gaps, representing an idealised data set. The second mimics real X‑ray monitoring campaigns: observations are irregularly spaced, with long gaps and a limited total exposure, i.e., a sparse sampling pattern. For each combination of PSD slope, sampling pattern, and signal‑to‑noise ratio (S/N) – varied from 1 to 10 – thousands of synthetic light curves are generated. The NXS is then measured on each simulated curve, allowing the authors to map out its bias, variance, and distribution under realistic conditions.
The simulations reveal three robust findings. First, even in the continuous case NXS is a biased estimator of the true variance. The bias grows with the steepness of the PSD: for β ≈ 2 the measured NXS underestimates the intrinsic variance by up to ~30 %. Second, sparse sampling exacerbates the bias because the window function preferentially filters out low‑frequency power, further reducing the measured variance. Third, the bias is essentially independent of S/N; low S/N merely inflates the scatter of the NXS estimates without changing the systematic offset.
To correct for the systematic offset the authors derive an empirical correction formula that depends only on the PSD slope β and on a binary flag indicating whether the sampling is continuous or sparse. Applying this correction to the simulated data reduces the average bias to less than 15 % across the explored parameter space, and the corrected NXS values recover the true variance with an accuracy that meets the requirements of most AGN variability studies.
However, the authors caution that a single light curve—particularly when S/N < 3 and the sampling is sparse—produces an NXS distribution that is highly asymmetric and has large uncertainties, even after bias correction. To overcome this limitation they propose an “ensemble” approach: either combine multiple observations of the same AGN or aggregate light curves from many AGN with similar properties. In the ensemble case the distribution of NXS becomes nearly Gaussian, the errors are well‑behaved, and the same bias‑correction formula can be applied, yielding unbiased variance estimates with substantially reduced statistical noise.
The practical implications are illustrated by applying the methodology to existing X‑ray archives (XMM‑Newton, Chandra, Swift). For a typical AGN with β ≈ 1.8, a continuous 20 ks exposure achieving S/N ≈ 4 yields an NXS measurement accurate to ≈20 %. If the same total exposure is broken into three 5 ks segments separated by 15 ks gaps (a sparse pattern), the required S/N rises to ≈6 to retain comparable accuracy. These thresholds guide observers in planning exposure times and cadence for faint or high‑redshift AGN.
Finally, the authors discuss how their results can inform the design of upcoming surveys such as eROSITA, Athena, LSST, and Pan‑STARRS. By quantifying the trade‑offs between cadence, total exposure, and S/N, the correction formula and ensemble strategy enable survey planners to predict the achievable variability sensitivity for large AGN samples, ensuring that variability‑based selection or classification schemes are statistically robust.
In summary, the paper demonstrates that NXS is intrinsically biased, quantifies the dependence of this bias on PSD slope and sampling pattern, provides a simple correction that restores unbiased variance estimates, and highlights the superiority of ensemble measurements for low‑S/N, sparsely sampled data. These insights are directly applicable to current X‑ray datasets and will be essential for extracting reliable variability information from the next generation of time‑domain astronomical surveys.