Download presentation
Presentation is loading. Please wait.
Published byMaximilian Hunter Modified over 9 years ago
1
GWDAW91Thursday, December 16 First Comparison Between LIGO &Virgo Inspiral Search Pipelines F. Beauville on behalf of the LIGO-Virgo Joint Working Group
2
GWDAW92Thursday, December 16 LIGO – Virgo Mock Data Challenge First LIGO – Virgo mock data challenge: Inspiral Project o Exchange simulated data with injected events, exchange triggers o Quantitative comparison and cross-check between pipelines Beginning of September o Simulated LIGO data (S. Chatterji, S. Fairhurst) and Virgo data (A.Viceré) Mid-October: face to face meeting o Exchange first results & Define trigger productions for quantitative study Search for inspiral mass range 1 to 3 M , Minimal Match 95%, SNR threshold 6 Starting frequency 30 Hz for Virgo data, 40 Hz for LIGO data Trigger production for both data sets: o LIGO pipeline (S. Fairhurst, N. Christensen) o Virgo flat search (L. Bosi) o Virgo Multi-Band pipeline (F. B.) Comparison study: o Mostly LIGO pipeline vs. Virgo Multi-Band pipeline o Some LIGO pipeline vs. Virgo flat search pipeline
3
GWDAW93Thursday, December 16 Data Generation 3 hours of LIGO & Virgo noise o Design sensitivities with line features 16384 Hz & 20 kHz Inspiral events injection o 2nd Order Post-Newtonian o 26 in LIGO data (every ~400s) distances 20, 25, 30, 35 Mpc masses 1.4-1.4 and 1.0-2.0 M Starting frequency 40 Hz o 11 in Virgo data(every ~ 900 s) SNR = 10, distance=24.83 Mpc masses 1.4 – 1.4 M starting frequency 24 Hz
4
GWDAW94Thursday, December 16 The LIGO Pipeline Use the LIGO inspiral pipeline. – Split data into analysis chunks of 15 overlapping segments. – Segment length at least 4 times longest template duration (depends on f_low). – Create template bank for each analysis chunk. – Filter data with 2PN templates generated in frequency domain. – Cluster triggers within duration of template. – No clustering between templates in the bank. The LIGO pipeline for LIGO data ● LIGO – f low = 40 Hz – longest template = 45 seconds – segment length = 256 seconds ● Virgo – f low = 30 Hz – longest template = 96 seconds – segment length = 512 seconds
5
GWDAW95Thursday, December 16 The Virgo Multi-band Pipeline Initialization o Spectrum (on 1800 s of noise) o Grid of full frequency band (VIRTUAL) templates o Grids of (REAL) templates for each frequency band Processing o Run synchronously each grid of REAL templates on data Data chunk twice the longest REAL template o Check if any REAL template triggers o Recombine associated VIRTUAL templates Coherent sum of real templates outputs: Obtain VIRTUAL templates triggers o Cluster those triggers, in time and between templates REAL VIRTUAL
6
GWDAW96Thursday, December 16 Triggers Production (Overview) LIGO dataVirgo data LIGO pipeline ~3500 Templates UWM Cluster (1 GHz Pentium II) 6 jobs => 6 time slices total time ~ 46 h ~9500 Templates CalTech Cluster (Xeon 2.66 GHz) 3 jobs => 3 time slices total time~ 88 h Virgo MultiBand ~1950 (VIRTUAL)Templates Xeon 2 GHz 2 jobs => 2 mass space regions total time ~15h Memory 1.1 G ~6190 (VIRTUAL) Templates Xeon 2 GHz 7 jobs => 7 mass space regions total time 38 h Memory ~ 5 G Virgo flat search 2556 templates Cluster of 7 Xeon 1.7 GHz 8 jobs total time 16 h Memory 13 G 8103 templates Cluster of 7 Xeon 1.7 GHz 23 jobs total time 48h Memory 38 G
7
GWDAW97Thursday, December 16 Triggers List Confrontation In each list, triggers are tagged “true” if (ending) time is less than 20 ms from an injection Fake triggers True trigger SNR LIGO Pipeline lists: Many triggers / injection (No clustering between templates in the bank.) Virgo Multi-Band lists: not more than 1 / injection (Clustering also between templates in the bank.) For Each injection, Compare: otrigger from Virgo Multi-Band obest SNR trigger from LIGO Pipeline Fake triggers True trigger SNR
8
GWDAW98Thursday, December 16 Cross-Validation (SNR) Same events detected, Strong SNR correlation between pipelines oVirgo data: All events found by both pipeline oLIGO data: 1 event missed by both pipelines + 1 missed by Virgo MB (near threshold) Also for Virgo flat search Virgo data LIGO data
9
GWDAW99Thursday, December 16 Cross-Validation (distance) Strong Correlation Also Good normalization agreement between pipelines Virgo data LIGO data Also for Virgo flat search 1.11
10
GWDAW910Thursday, December 16 Compare Arrival Time Compare offsets with injection time No correlation, Multi-Band spread 6X larger than LIGO pipeline o Known (fixable) template-dependant error on time offset in Virgo Multi-Band Analysis implementation
11
GWDAW911Thursday, December 16 Compare Masses Virgo data LIGO data LIGO Data o large spread for Virgo Multi- Band (showing grid resolution) vs. reduced spread & error for LIGO pipeline Virgo Data o Similar (slightly correlated) spread for both pipelines. Brings evidence of template grids effect o LIGO pipeline has templates close to injected events in both set, not Virgo MB. Need for more elaborate post-processing o Less dependant on grid choice
12
GWDAW912Thursday, December 16 Post-processing Exists in LIGO pipeline for distance and chirpmass o heavy, not processed systematically for all true triggers In development in Virgo Multi-Band analysis for masses o based on SNR from different templates within a cluster Posterior PDF estimate for a Virgo injection distance=25Mpc, chirpmass=1.2188, m1=m2=1.4. For both Virgo and LIGO injections with these mass parameters the chirp mass estimate is very close, but consistently low.
13
GWDAW913Thursday, December 16 Achievements Learned to process each others’ data, exchange and compare triggers o Event format translator Pipelines cross-validation: o Same events found by different pipelines o Correlated triggers ( SNR, distance ) o Remaining differences ( 10 % in worst case) under discussion => Leads to better understanding of each other pipeline Put focus on specific needs o Fix arrival time estimation in multi-band analysis o Post-processing Got ready for next step: o Injections of astrophysical events (from given location in the sky)
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.