Source-linked AI summary
On the artificial nature of aperiodic variability in XMM-Newton observations of M31 X-ray sources and the ultraluminous X-ray source NGC 4559 ULX-7
R. Barnard, S. Trudolyubov, U. C. Kolb, C. A. Haswell, J. P. Osborne, W. C. Priedhorsky
TL;DR
The paper investigates false LMXB-like power density spectra caused by combining non-synchronised XMM-Newton lightcurves. By re-analysing observations and simulations, it finds the artefact depends mainly on timing offsets, invalidates reported variability in M31 and NGC 4559 ULX-7, and confirms genuine variability remains detectable in NGC 5408 ULX1.
Problem
Previously reported Type A power density spectra in M31 X-ray sources may be artefacts of improperly adding or subtracting non-simultaneous lightcurves.
Method
The study re-analyses M31 observations with simultaneous and non-simultaneous lightcurves and combines simulated lightcurves across different intensities and timing offsets.
Results
The artefact depends more on offset than intensity; Method II produces Poisson-consistent power in a tested M31 source, while NGC 4559 ULX-7 shows no significant variability after correct combination.
Takeaways & Limitations
XMM-Newton timing analysis remains viable, but SAS-generated lightcurves must be synchronised before detector combination or background subtraction.
Abstract
from arXiv · showhide
Context: Power density spectra (PDS) that are characteristic of low mass X-ray binaries (LMXBs) have been previously reported for M31 X-ray sources, observed by XMM-Newton. However, we have recently discovered that these PDS result from the improper addition/subtraction of non-simultaneous lightcurves. Aims: To understand the properties and origins of the artefact. Methods: We re-analysed our XMM-Newton observations of M31 with non-simultaneous and simultaneous lightcurves, then combined simulated lightcurves at various intensities with various offsets and found that the artefact is more dependent on the offset than the intensity. Results: The lightcurves produced by the XMM-Newton Science Analysis Software (SAS) are non-synchronised by default. This affects not only the combination of lightcurves from the three EPIC detectors (MOS1, MOS2 and pn), but also background subtraction in the same CCD. It is therefore imperative that all SAS-generated lightcurves are synchronised by time filtering, even if the whole observation is to be used. We also find that the reported timing behaviour for NGC 4559 ULX-7 was also contaminated by the artefact; there is no significant variability in the correctly-combined lightcurves of NGC 4559 ULX-7. Hence, the classification of this source as an intermediate-mass black hole is no longer justified. Conclusions: While previous timing results from M31 have been proven wrong, and also the broken power law PDS in NGC 4559 ULX-7, XMM-Newton was able to detect aperiodic variability in just 3 ks of observations of NGC 5408 ULX1. Hence XMM-Newton remains a viable tool for analysing variability in extra-galactic X-ray sources.
1. Introduction
Earlier XMM-Newton studies reported Type A-like PDS in M31 and ULXs, but the paper identifies false variability caused by combining non-synchronised lightcurves. Correctly processed data distinguish artificial signals from genuine extragalactic variability.
- Type A PDS are broken power laws associated with disc-accreting X-ray binaries and can identify an X-ray source as an X-ray binary.Their break typically lies at 0.01–1 Hz, with approximately 10–40% rms power.
- Type A PDS had previously been reported in XMM-Newton observations of M31 sources and ULXs in NGC 4559 and NGC 5408.
- Improperly adding non-synchronised MOS1, MOS2, and pn lightcurves produces false Type A PDS.The paper investigates how the artefact arose and reanalyses affected observations.
- Method I produces an apparently broken PDS for r3-60, whereas Method II produces a flat PDS, demonstrating that the reported variability is artificial.
2. Manipulating lightcurves generated by SAS
The analysis found that SAS-generated lightcurves can differ according to how time selection is specified. Methods II and III synchronise lightcurves, unlike the original Method I procedure.
- M31 observations supplied 0.3–10 keV source and background lightcurves from MOS1, MOS2, and pn images with 2.6-second bins.
- The original Method I filtered event files using energy and time expressions, with additional instrument-dependent filtering.
- Method I is the paper’s name for the original event-filtering procedure.
- Methods II and III are equivalent time-filtering approaches but are not equivalent to Method I, prompting comparative analysis.Method II assigns observation bounds to event-list keywords, while Method III filters lightcurves with timemin and timemax.
3. Re-analysing the data
Reanalysis shows that non-synchronised lightcurves distort both detector combination and background subtraction. Synchronised Method II processing removes the apparent Type A variability in the tested M31 source.
- 3.1. Combined EPIC lightcurves: Method I uncertainties for combined EPIC lightcurves were underestimated by approximately 10–40%.The ratio varied within each lightcurve and its standard deviation decreased with increasing luminosity.
- 3.1. Combined EPIC lightcurves: Method II lightcurves are synchronised, whereas Method I lightcurves are not; synchronisation requires a common tstart in all detector event lists.
- 3.1. Combined EPIC lightcurves: A broken PDS fit for r3-60 had a break at 29±4 mHz, but the identical Method II PDS was flat at Leahy power 2 and consistent with Poisson noise.
- 3.2. Background Subtraction: Background-subtracted r3-125 lightcurves showed Type A PDS even though the source and background component PDS were flat.The result indicates that non-synchronised source and background lightcurves create subtraction artefacts.
4. Investigating the artificial variability with simulated lightcurves
Simulations show that artificial PDS variability is governed more by timing offsets than intensities and becomes stronger when more non-synchronous lightcurves are combined. Type A classification is most common at intermediate offsets.
- 4. Investigating the artificial variability with simulated lightcurves: Simulated source and background lightcurves covered intensities from 0.01 to 1.00 count s−1, lasted 60 ks, and used 2.6-second bins.
- 4.1. Adding two lightcurves with different start times: Offsets of 0.25–0.80 produced significant excess variability in all simulated combinations, while offsets of 0.05 or 0.95 did so in 12%.
- 4.1. Adding two lightcurves with different start times: Less than 10% of lightcurves at offsets 0.05–0.25 and 0.75–0.95 were Type A, compared with 30% at f = 0.50.Type A required rejection of a power-law fit and an acceptable broken-power-law fit.
- 4.2. Adding three lightcurves with different start times: Adding three non-synchronous lightcurves produced more than twice the false Type A variability of combining two, strengthening the artefact through detector combination or background subtraction.
5. The causes of the artefact
The artefact arises from non-synchronised lightcurve combination, which underestimates uncertainties, inflates PDS power, and suppresses high-frequency variability.
- Overview: Non-synchronised lightcurves cause three effects: underestimated uncertainties, excess PDS power, and suppressed high-frequency variability.These mechanisms jointly produce artificial Type A variability.
- Offset lightcurves: S(n) combines the source lightcurve A(n) with offset background bins B(n−1) and B(n), weighted by fractional offset f.The equation models addition or subtraction when lightcurves are not aligned.
- Offset lightcurves: For synchronised lightcurves, f = 0, so lcmath correctly combines corresponding source and background intervals.Synchronisation sets both start times to the same user-defined value.
- Uncertainty underestimation: When multiple background bins overlap one source interval, lcmath uses incorrect overlap-weighted uncertainties, causing them to be underestimated.The error affects both cases where uncertainties are absent and cases where uncertainties are supplied.
- PDS power: Underestimating Method I uncertainties by 1.1–1.4 overestimates PDS power by approximately 1.2–2.Leahy normalisation assumes Gaussian errors and scales inversely with the mean squared error.
- High-frequency suppression: After correcting the power excess, the Method I PDS falls below the Poisson level at high frequencies, suppressing variability and producing a break.For identified M31 sources, νb ranges from 20±4 to 57.8±0.5 mHz, independent of luminosity.
6. Ultra-luminous X-ray sources in NGC 4559 and NGC 5408
The authors reanalysed published ULX timing results using non-synchronised and synchronised lightcurves. The NGC 4559 variability was artificial, whereas NGC 5408 showed real broken-power-law variability.
- Reanalysis: The authors compared published-analysis methods using non-synchronised lightcurves with correctly synchronised lightcurves for ULXs in NGC 4559 and NGC 5408.The reanalysis tested whether reported broken power-law PDS persisted after synchronisation.
- NGC 4559 ULX-7: NGC 4559 ULX-7’s correctly analysed PDS is consistent with zero power, with a good-fit probability of 0.22.The expected noise power was 4.3, and the PDS used fractional-r.m.s.^2 normalisation.
- NGC 4559 ULX-7: The reported variability of NGC 4559 ULX-7 was therefore artificial rather than intrinsic timing behaviour.A separate reanalysis reproduced it only under an incorrect attribution to background flares.
- NGC 5408 ULX1: The NGC 5408 result indicates that XMM-Newton can detect genuine aperiodic variability despite the artefact affecting other analyses.The supplied passage directly establishes real variability through the poor zero-power fit and preferred broken-power-law fit.
7. Conclusions
The reported Type A variability in M31 and the broken-power-law PDS of NGC 4559 ULX-7 are artefacts of non-synchronised lightcurve handling, but XMM-Newton can still detect genuine extragalactic variability.
- Type A variability reported for M31 sources is attributable to errors when combining non-synchronised lightcurves.The artefact arises from improper addition or subtraction of lightcurves from XMM-Newton EPIC instruments.
- The apparent luminosity dependence of M31 variability likely reflects smaller offsets in brighter sources, because lightcurve start times follow the first event arrival.Brighter sources naturally produce smaller offsets between CCD lightcurves.
- The NGC 4559 ULX-7 PDS is consistent with zero power, providing no evidence for the variability previously reported there.The Method II PDS has χ2/dof = 20/16 and a null hypothesis probability of 0.22.
- The published broken-power-law PDS of NGC 4559 ULX-7 is artificial, whereas the NGC 5408 ULX shows genuine intrinsic variability.These findings support continued use of XMM-Newton for timing analysis of extragalactic X-ray sources.