Transcript Gravitational Waves – a data analysis perspective

Data Characterization in
Gravitational Waves
Soma Mukherjee
Max Planck Institut fuer Gravitationsphysik
Golm, Germany.
Talk at
University of Texas, Brownsville.
March 26, 2003
Soma Mukherjee 26/3/03
What data do we have ?
Consists of information from the main
gravitational wave channel and ~1000
auxiliary channels.


Science run data (S1 and S2) from the
three LIGO and GEO interferometers.
Several (E1-E9) Engineering run data.
Soma Mukherjee 26/3/03
What does data analysis involve ?
Detector Characterization :


Looking at ALL channels all the time for detector diagnostic
Calibration
Data Characterization :


Checking the stability of the data
Data decomposition
Astrophysical Searches :


Algorithm development
Post search analysis
 Vetoes

Upper limits
Simulations
Soma Mukherjee 26/3/03
Computational Aspects
Very large data volume demands
Automation
Speed
Parallel processing
Efficient database
Systems available : LDAS, DMT, DCR, GODCS
Data Mining
Soma Mukherjee 26/3/03
Aspects that I work on
Data stability


Non-stationarity – detection and measure
Implications in Astrophysics
Burst Upper Limit*

Post detection analysis – Data Mining
 Exploratory
 Classification
 Coincidence
Externally Triggered Search

*
Association with Gamma Ray Bursts
http://www.aei.mpg.de/~soma/bursts.html
Robust Detection of Noise
Floor Drifts in Interferometric
Data
Soma Mukherjee, 26/3/03
Why :
Interferometric data has three components : Lines,
transients, noise floor.
Study of a change in any one of these without
elimination of the other two will cause interference.
Lines dominate.
Presence of transients change the central tendency.
“SLOW” nonstationarity of noise floor interesting in the
analysis of several astrophysical searches, e.g.
Externally triggered search.
To be able to simulate the non-stationarity to test the
efficiencies of various algorithms.
Soma Mukherjee,
26/3/03
Method :
MNFT :
1. Bandpass and resample given timeseries x(k).
2. Construct FIR filter than whitens the noise floor.
Resulting timeseries : w(k)
3. Remove lines using notch filter. Cleaned timeseries
: c(k)
4. Track variation in second moment of c(k) using
5.
Running Median*.
Obtain significance levels of the sampling
distribution via Monte Carlo simulations.
* Mohanty S.D., 2002, CQG
Soma Mukherjee 26/3/03
Sequence :
Low pass and
resample
Thresholds set by
Simulation.
Soma Mukherjee,
GWDAW7, Kyoto,
Japan, 19/12/02
Estimate spectral
noise floor using
Running Median
Design FIR
Whitening filter.
Whiten data.
Compute Running
Median of the
squared timeseries.
Clean lines.
Highpass.
9
Soma Mukherjee, 26/3/03
Data :
Locked segments from :
LIGO S1 : L1 and H2
LIGO S2 : L1 and H1
Soma Mukherjee, 26/3/03
Soma Mukherjee, 26/3/03
Soma Mukherjee, 26/3/03
Soma Mukherjee, 26/3/03
Soma Mukherjee 26/3/03
Soma Mukherjee 26/3/03
Soma Mukherjee 26/3/03
Soma Mukherjee 26/3/03
Soma Mukherjee 26/3/03
With transients added
Soma Mukherjee 26/3/03
Soma Mukherjee 26/3/03
Computation of :
G(m)=V(Z t+m – Z t)/V(Z t+1 – Z t)
Z t : t th sample of a timeseries.
m: Lag.
Soma Mukherjee 26/3/03
Comments :
Soma Mukherjee 26/3/03
Threshold setting by single simulation.
Discussions underway for incorporation in
the externally triggered burst search
analysis.
Automation.
Use MBLT for line removal.
C++ codes underway.
Incorporation in the DCR in near future.
Soma Mukherjee 26/3/03
Questions wrt Astrophysical
Search
Threshold and tolerance.
… being worked up on.
Work in the area of Burst
Upper Limits
Soma Mukherjee 26/3/03
Components of a Burst search pipeline
Conditioned Data
Coincidence
Search Filter
Production of list
of events
Event database
Generate Veto
Soma Mukherjee 26/3/03
Veto generation – Classification
Data from main (h(t))
channel
Data from Auxiliary
Channels
……….
Triggers
Trigger characterization (amplitude, frequency,
shape information, duration, time of arrival…)
Classification
Instrumental Source
Identification
Soma Mukherjee 26/3/03
Classification continued …
Future plans
Construction of a distance measure in
multi-parameter space.
Identification of non-redundant
parameters.
Discover statistically significant clusters.
Correlate bursts from different sources
that fall into the same cluster.
Soma Mukherjee 26/03/03
More future plans
Continue analysis of Science data.
More emphasis on injection and
simulation in the burst analysis.
Suitable modification to the existing
algorithms to accommodate nonstationarity.
Development of efficient post-detection
algorithms.