Gravitational Wave Transients state-of-the-arts: detection confidence and signal reconstruction G.A.Prodi, University of Trento and INFN, for the LIGO Scientific Collaboration and the Virgo Collaboration plenty of credits to many colleagues...
Outline 1. Basics of gravitational wave transient signal searches 2. Reconstruction of signal properties 3. Assessment of the confidence of events 4. Case study: end-to-end test by blind injection challenge 2
Noise transients The identification of gravitational wave transients is limited by noise transients at detectors of instrumental, environmental and unknown origin in accidental coincidence. Noise tails >> Gaussian No evidence for foreground non-accidental transients over 2 yrs observation time Strategies: clean by vetoing bad periods or noise transients clean by network data analysis clean by selection of target transient signals characteristics The search is still dominated d by non-gaussian tails 3
Assumptions on transient signals specific features of the target signal are useful to reduce the noise background contributed by each detector Search for any gw signal in band: minimal constraints (max. signal duration) constraining waveform targeted to modeled signals: e.g. template banks for NS, BH coalescence ring down of NS, BH co onstrain ing occurren nce Externally triggered searches: constraining time and direction e.g. GRBs, SGR, High Energy Neutrinos, Supernovae, Actual detection probability depends on source population p and horizon of the search 4
Directional sensitivity & sky coverage On the wavefront of a gravitational wave: two polarization components polarization states: linear, elliptical, unpolarized h + h x each interferometric detector senses sensitivity of detectors for a one combination h sample direction: det = F+ h+ + F h V1 sense different coefficients function polarization H1 of sky direction components L1 sensitivity to polarizations of H1 L1 V1 network: most favorable polarization least favorable / most favorable Phys. Rev. D 72, 122002 (2005) 5
Coherent Network data analysis Information from different detectors is useful both for noise rejection and for signal reconstruction Coincidence analysis: Coherent analyses: detectors analyzed detectors separately synthesized together careful not to miss signals lower computational load Likelihood maximization over the sky for a simulated event in H1 L1 V1 network fully exploit general network properties and GW properties separate the coherent response to a GW from the incoherent noises of detectors many implementations are possible (e.g. Likelihood Maximization, Bayesian evidence) simpler implementation for networks of many detectors: redundant information available more complex implementation for few detectors: lack of information on one polarization, ambiguities of solutions in the sky (current 2 3 sites network needs ad hoc constraints) t 6
Coherent Network data analysis Reconstruction of source direction for un-modeled transients (simulation) figures of merit for position uncertainty (simulated data&signals for SNRs [10-30] at 50% probability) Cumulative fraction of the sky with uncertainty area H1-L1-V1 [deg] J1-H1-L1-V1 area [deg 2 ] The extension to a 4 th detector site greatly improves the performance of source direction reconstruction (J1 is LCGT, see next talk) Klimenko et al., PRD 83, 102001 (2011) 7
Coherent Network data analysis Signal reconstruction by un-modeled transient search (simulation) reconstruction ti at one detector t (whitened strain at detector) software injection with: network SNR 30 BH-BH coalescence (11+11 M sun) ) waveform most SNR comes from the sweetspot of detectors ( 100-200 Hz) More on signal reconstruction: talk by Alan Weinstein tomorrow 8
Assessment of the Confidence Accidental background of transient searches standard time slides technique time shift data of detectors in the network repeat the analysis reference distribution for accidental events measure confidence of on-source events: False Alarm Rate follow-up on-source events with low FAR: excluding possible instrumental or environmental causes Critical issues: number of useful time slides is limited lower limit on FAR studies: minimum time shift step O(1s) due to detectors autocorrelation time maximum time shift O(1d) due to non stationary noise of detectors check for pollution by foreground or signal events in the network same network event may repeat itself with negligible differences in different time slides (multiple events) correlation among different time slides is possible even with independent detector noises 9
Case Study: blind injection LIGO-Virgo observations passed an end-to-end test: Blind injection challenge http://www.ligo.org/news/blind-injection.php injection.php Secret signal injections were performed in the detectors (0, 1, or a few signals) to mimic the effect of real gravitational waves impinging on the network. Complete off-line searches on subperiod of observation null results preliminary internal review GW Alert deep internal review draft paper on results final approval of paper for submission blind injection/s? not a blind injection else lessons learned etc. submit paper! nice try 10
Case Study: blind injection strong candidate event found on Sept 16, 2010, 06:42 UTC GW100916 Analysis completed, paper draft approved in 6 months of harder work...... but it was a blind injection at least we passed the test ready for the first transient detection 11
Case Study: blind injection strong candidate Sept 16, 2010, 06:42 UTC GW100916 detected and reconstructed with a latency of a few minutes by a coherent pipeline searching un-modeled bursts Low latency checking procedures confirmed the interest in the signal: chirping in frequency as expected from compact binary inspiral louder than most noise transient events (FAR 2/yr ) detectors were operating smoothly Spectrograms of whitened strain output: H1 SNR 14 L1 SNR 10 V1 SNR 4 Total SNR in the network 17 (both un-modeled and template searches) information was released to partner astronomers for follow-up observations within 45 minutes (talk by Marica Branchesi) 12
Case Study: blind injection GW100916 Signal-to-Noise Ratio in a detector is set by: spectral sensitivity directional sensitivity Spectral sensitivity of detectors Directional sensitivity of fdetectors t for the reconstructed position H1 Expected signal strength at SNR 15 in H1 SNR 10 in L1 L1 V1 13
Case Study: blind injection GW100916 readily available information Sky map of the reconstructed source location Wide spots distributed on the ring of 7ms time delay between H1 and L1 plotted using Aladin tool http://aladin.u-strasbg.fr Telescope shoots were requested on the main spot, to cover nearby galaxies 14
Case Study: blind injection GW100916 assessing the confidence Template search for coalescing compact bynaries shows very low False Alarm Rate: 1/7000 yrs over all trials performed FAR was considered low enough to report the observation of a Binary BH coalescence. FAR of the unmodeled burst search is not significant enough ( 1/yr), being open to a much larger class of waveforms coincident events in H1 L1 seen with chirp mass [3.48,7.4]M sun background background excluding data around GW100916 time (SNR&compliance to template) GW100916 FAR 1/40,000yrs in the H1-L1 coincidence 15
Case Study: blind injection GW100916 achievements proved with high confidence that the event can not be of instrumental or environmental origin developed tools to investigate extremely low False Alarm Rates <1/10,000 000 yrs faster progresses on signal parameter estimation, solving the inverse problem and e.g. disentangling ambiguities in the large parameter space of compact binary coalescences (plenary talk by Alan Weinstein) success of the low latency search and test for prompt Electromagnetic Follow-up of gw candidates The interferometric gravitational wave observatory is capable of confident detection of transient gravitational waves capturing astrophysics of their sources 16
Case Study: blind injection Take a look at our open release of data! http://www.ligo.org/science/gw100916/index.php 17