CBC R&D presentation

Basic description of what we have done

Method: we compare the cWB point estimate waveform for a given GW/LVT event with a distribution of its PE posterior samples injected on O2 data and reconstructed by cWB.

Confidence belts: Departure of the cWB point estimate (red curve) outside of the corresponding confidence belts
gives us a conservative measure of the statistical agreement between the burst waveform and the PE samples

Issues

the first event for which our confidence in detection is based almost entirely on cWB

we should make as strong a case as we can in terms of testing consistency with a real signal

Why a burst pipeline is finding a BBH with higher significance than MF pipelines?

Interesting explanations: deviations from GR, features not included in current template banks, such as HM, precession or eccentricity, etc.

"Boring" explanation: currently, cWB is more optimized than MF pipelines wrt heavier BBH systems

LVT170729: plot

Minor inconsistency in the merger

Extra pixel

~30% of LVT170729 LALInference PE samples when injected on chunk 19 are found by cWB with higher significance than the original event (i.e. IFAR=56 yrs)

SImilar results for PE samples from the IMRPhenomHM run

What about PyCBC and GstLAL? If a significant fraction of the PE samples are consistent with their significance estimate (IFAR~1 yr) then we will know that for events like LVT170729 the discrepancy in significance that we observe is to be expected.

PE runs

Code implementations

code versions