New veto for continuous gravitational wave searches

We present a new veto procedure to distinguish between continuous gravitational wave (CW) signals and the detector artifacts that can mimic their behavior. The veto procedure exploits the fact that a long-lasting coherent disturbance is less likely than a real signal to exhibit a Doppler modulation of astrophysical origin. Therefore, in the presence of an outlier from a search, we perform a multistep search around the frequency of the outlier with the Doppler modulation turned off (DM-off), and compare these results with the results from the original (DM-on) search. If the results from the DM-off search are more significant than those from the DM-on search, the outlier is most likely due to an artifact rather than a signal. We tune the veto procedure so that it has a very low false dismissal rate. With this veto, we are able to identify as coherent disturbances $>99.9\%$ of the 6349 candidates from the recent all-sky low-frequency Einstein@Home search on the data from the Advanced LIGO O1 observing run [LIGO and Virgo Collaborations, Phys. Rev. D 96, 122004 (2017)]. We present the details of each identified disturbance in the Appendix.

Figure 1

The mismatch gives a measure of the dissimilarity between the DM-off waveforms and an astrophysical signal. The left panel here shows the mismatch as a function of the latitude of the source, where zero is at the equator and $\pi /2$ is at the poles. The right panel has the histograms for those mismatches. There is a dependence on the latitude and on frequency of the signal, as one would expect; sources at lower ecliptic latitudes and with higher frequencies have larger mismatches.

Figure 2

Comparison between the original search results (orange) and the DM-off search results (blue) for candidates 1 and 6 of [6]. Candidate 1 (top) did not pass the detection threshold set at $2\overline{\mathcal{F}}=15$ in [6]; candidate 6 (bottom) did, but was associated with a fake continuous wave signal in the data. After a DM-off search, the line that caused candidate 1 becomes more prominent while the signal from candidate 6 disappears into the noise. We use these two candidates to illustrate that the DM-off veto concept works.

Figure 3

Data containing fake noise plus 1500 simulated signals are searched with both the DM-off and the DM-on searches. The highest detection statistics are recorded as $2{\mathcal{F}}_{\mathrm{DM}-\mathrm{off}}$ and $2{\mathcal{F}}_{\mathrm{DM}-\mathrm{on}}$, respectively. For signals with $2{\mathcal{F}}_{\mathrm{DM}-\mathrm{on}}\lesssim 400$, the $2{\mathcal{F}}_{\mathrm{DM}-\mathrm{off}}$ is not distinguishable from the noise. The loudest value of $2{\mathcal{F}}_{\mathrm{DM}-\mathrm{off}}$ found for this group is 61.25. For louder $2{\mathcal{F}}_{\mathrm{DM}-\mathrm{on}}$, the $2{\mathcal{F}}_{\mathrm{DM}-\mathrm{off}}$ values are significantly above the noise and increase with $2{\mathcal{F}}_{\mathrm{DM}-\mathrm{on}}$. The veto threshold is indicated by the red dashed line, and was chosen so that none of the fake signals would have been rejected hence yielding a false dismissal rate of astrophysical signals smaller than one in 1500. Candidates that fall in the region above the red dashed line do not behave like signals, and therefore can be rejected.

Figure 4

An example of a candidate that does not survive Step 1. For this candidate, $2{\mathcal{F}}_{\mathrm{DM}-\mathrm{off}}$ in H1 exceeds the threshold (the horizontal red dashed line) and so this candidate is discarded. This particular candidate’s elevated $2{\mathcal{F}}_{\mathrm{DM}-\mathrm{on}}$ value is caused by a stationary line in H1.

Figure 5

An example of a candidate that does not survive Step 2. This candidate comes from a wandering line in H1.

Figure 6

$2{\mathcal{F}}_{\mathrm{DM}-\mathrm{off}}$ vs $2{\mathcal{F}}_{\mathrm{DM}-\mathrm{on}}$ for the 6349 candidates of the example that we illustrate in Sec. 3. The dashed line shows the veto threshold of Eq. (2), and the pink and blue points are described in Fig. 3. The region above the dashed line is the candidate-rejection region. Four signal candidates (blue circles) survive the DM-off veto, as well as fifth candidate that is due to a fake signal. The remaining 6345 candidates (gray dots) are rejected as being caused by detector artifacts.