Density-clustering of continuous gravitational wave candidates from large surveys

Searches for continuous gravitational waves target nearly monochromaticgravitational wave emission from e.g. non-axysmmetric fast-spinning neutronstars. Broad surveys often require to explicitly search for a very large numberof different waveforms, easily exceeding $\sim10^{17}$ templates. In suchcases, for practical reasons, only the top, say $\sim10^{10}$, results aresaved and followed-up through a hierarchy of stages. Most of these candidatesare not completely independent of neighbouring ones, but arise due to somecommon cause: a fluctuation, a signal or a disturbance. By judiciouslyclustering together candidates stemming from the same root cause, thesubsequent follow-ups become more effective. A number of clustering algorithmshave been employed in past searches based on iteratively finding symmetric andcompact over-densities around candidates with high detection statistic values.The new clustering method presented in this paper is a significant improvementover previous methods: it is agnostic about the shape of the over-densities, isvery efficient and it is effective: at a very high detection efficiency, it hasa noise rejection of $99.99\%$ , is capable of clustering two orders ofmagnitude more candidates than attainable before and, at fixed sensitivity itenables more than a factor of 30 faster follow-ups. We also demonstrate how tooptimally choose the clustering parameters.