enhancing the sensitivity of searches for gravitational ... · ott1 cwb cwb+bw 108 107 106 105 104...
TRANSCRIPT
Kiranjyot Gill
Enhancing the Sensitivity of Searches for Gravitational Waves from Core-Collapse
Supernovae with a Bayesian classification of candidate events
Kiranjyot Gill
03/17/2017March LVC 2017
For LIGO folk: S5 data used
Collaborators: Wenhui Wang, Oscar Valdez, Marek Szczepanczyk, Michele Zanolin & Soma Mukherjee
NAME
EVENT NAME
TALK TITLE DATENAME DATETALK TITLEKiranjyot Gill 03/17/2017March LVC 2017
The Need for BayesWaveGoal for LVC-SN Searches: reduction of the false alarm rate produced
by cWB in order to improve the ROC curve for GW detectionProcedure
1) cWB outputs a ‘ranking statistic’ that is used to separate the background noise from the injected triggers
2) All surviving triggers that are above the nominal value of the ranking statistic are then post-processed through BW
3) BW initially produces results using a scatterplot that differentiates between glitches, noise, and signals present in the data
4) This secondary classification is applied to the cWB ROC curve in hopes of improving the false alarm rate - and essentially the detection efficiency of each waveform family
NAME
EVENT NAME
TALK TITLE DATENAME DATETALK TITLEKiranjyot Gill 03/17/2017March LVC 2017
10�8 10�7 10�6 10�5 10�4 10�30.35
0.40
0.45
0.50
0.55
FAR
Effic
ienc
y
cWB ROC
If cWB+BW efficiency better than original cWB FAR = improvement!
old cWB FAR
new cWB+BW point with improved efficiency
The Need for BayesWave
NAME
EVENT NAME
TALK TITLE DATENAME DATETALK TITLEKiranjyot Gill 03/17/2017March LVC 2017
BayesWave Breakdown
SN source already known
LIGO-G1401157
catered to SN searches
NAME
EVENT NAME
TALK TITLE DATENAME DATETALK TITLEKiranjyot Gill 03/17/2017March LVC 2017
Seeing through the eyes of BW
NAME
EVENT NAME
TALK TITLE DATENAME DATETALK TITLEKiranjyot Gill 03/17/2017March LVC 2017
CCSNe “signals”
Successfully identified “background noise”
Seeing through the eyes of BW
NAME
EVENT NAME
TALK TITLE DATENAME DATETALK TITLEKiranjyot Gill 03/17/2017March LVC 2017
Seeing through the eyes of BW
NAME
EVENT NAME
TALK TITLE DATENAME DATETALK TITLEKiranjyot Gill 03/17/2017March LVC 2017
Seeing through the eyes of BW
NAME
EVENT NAME
TALK TITLE DATENAME DATETALK TITLEKiranjyot Gill 03/17/2017March LVC 2017
O1/O2 Search Pool of Waveforms
Rotating Core-Collapse
Scheidegger+10 sch1: R1E1CA_L_thetaX.XXX_phiX.XXX sch2: R3E1AC_L_thetaX.XXX_phiX.XXX sch3: R4E1FC_L_thetaX.XXX_phiX.XXX
Dimmelmeier+08 dim1: signal_s15a2o05_ls dim2: signal_s15a2o09_ls dim3: signal_s15a3o15_ls
NAME
EVENT NAME
TALK TITLE DATENAME DATETALK TITLEKiranjyot Gill 03/17/2017March LVC 2017
O1/O2 Search Pool of Waveforms
Neutrino-driven Explosion Mueller+12
mul1: L153_thetaX.XXX_phiX.XXX mul2: N202_thetaX.XXX_phiX.XXX mul3: W154_thetaX.XXX_phiX.XXX
Ott+13 ott1: s27fheat1p05_thetaX.XXX_phiX.XXX
Yakunin+15 yak1: B12WH07 yak2: B15WH07 yak3: B20WH07 yak4: B25WH07
NAME
EVENT NAME
TALK TITLE DATENAME DATETALK TITLEKiranjyot Gill 03/17/2017March LVC 2017
BW Post-Processing Results
0.380.400.420.440.460.480.500.520.540.560.58
Effic
ienc
y
Dim1
0.42
0.44
0.46
0.48
0.50
0.52
0.54
Effic
ienc
y
Dim2
10�8 10�7 10�6 10�5 10�4
FAR
0.28
0.32
0.36
0.40
0.44
0.48
0.52
Effic
ienc
y
Dim3
cWBcWB+BW
0.38
0.40
0.42
0.44
0.46
0.48
0.50
0.52
0.54
0.56
Effic
ienc
y
Sch1-p
0.1
0.2
0.3
0.4
0.5
0.6
Effic
ienc
y
Sch1-v
10�8 10�7 10�6 10�5 10�4
FAR
0.25
0.30
0.35
0.40
0.45
0.50
Effic
ienc
y
Sch1-c
cWBcWB+BW
NAME
EVENT NAME
TALK TITLE DATENAME DATETALK TITLEKiranjyot Gill 03/17/2017March LVC 2017
10�8 10�7 10�6 10�5 10�4
0.35
0.40
0.45
0.50
0.55
Effic
ienc
y
FAR
Ott1cWBcWB+BW
10�8 10�7 10�6 10�5 10�4
0.30
0.35
0.40
0.45
0.50
Effic
ienc
y
FAR
MurphycWBcWB+BW
0.35
0.40
0.45
0.50
0.55
Effic
ienc
y
Sch2
10�8 10�7 10�6 10�5 10�4
0.30
0.35
0.40
0.45
Effic
ienc
y
Sch3
cWBcWB+BW
BW Post-Processing Results
NAME
EVENT NAME
TALK TITLE DATENAME DATETALK TITLEKiranjyot Gill 03/17/2017March LVC 2017
Waveform FAR cWB+BW
sch1-wf12 10-6 13.184% increase
sch2 10-6 10.243% increase
sch3 10-6 1.1643% increase
dim1 10-6 4.522% increase
dim2 10-6 3.062% increase
dim3 10-6 10.434% increase
murphy 10-6 12.412% increase
ott1 10-6 1.193% increase
cWB+BW ROC Improvement
NAME
EVENT NAME
TALK TITLE DATENAME DATETALK TITLEKiranjyot Gill 03/17/2017March LVC 2017
First paper is on the way!
NAME
EVENT NAME
TALK TITLE DATENAME DATETALK TITLEKiranjyot Gill 03/17/2017March LVC 2017
Linear polarization v Non-linear polarization
O1/O2 Search Pool of Waveforms
NAME
EVENT NAME
TALK TITLE DATENAME DATETALK TITLEKiranjyot Gill 03/17/2017March LVC 2017
Example: Mul1 & Mul3 (linear polarized BW code)[W. Wang runs]
Mishandled CCSNe Waveforms
injections are identified as
glitches instead of signals
The waveforms are initially produced at 10 kpc
(rescaled distances with scale factors)
NAME
EVENT NAME
TALK TITLE DATENAME DATETALK TITLEKiranjyot Gill 03/17/2017March LVC 2017
0 10 20 30 40 50 60
10
20
30
40
50
60
70
80
90
100
L1 SNR
H1
SNR
Yak1Yak2Yak3
0 10 20 30 40 50 60
10
20
30
40
50
60
70
80
90
100
L1 SNR
H1
SNR
Sch1Sch2Sch3
0 5 10 15 20
5
10
15
20
25
30
35
L1 SNR
H1
SNR
Mul1Mul2Mul3
0 20 40 60 80 100 120 140
20406080
100120140160180200220240
L1 SNR
H1
SNR
Ott
Framing the Problem with an SNR Outlook
NAME
EVENT NAME
TALK TITLE DATENAME DATETALK TITLEKiranjyot Gill 03/17/2017March LVC 2017
0 10 20 30 40 50 60
10
20
30
40
50
60
70
80
90
100
L1 SNR
H1
SNR
Yak1Yak2Yak3
0 10 20 30 40 50 60
10
20
30
40
50
60
70
80
90
100
L1 SNR
H1
SNR
Sch1Sch2Sch3
0 5 10 15 20
5
10
15
20
25
30
35
L1 SNR
H1
SNR
Mul1Mul2Mul3
0 20 40 60 80 100 120 140
20406080
100120140160180200220240
L1 SNR
H1
SNR
Ott
Framing the Problem with an SNR Outlook
Range of SNRs tested = problem is universal regardless of distance!
NAME
EVENT NAME
TALK TITLE DATENAME DATETALK TITLEKiranjyot Gill 03/17/2017March LVC 2017
Sch2 BW Waveform Reconstruction
NAME
EVENT NAME
TALK TITLE DATENAME DATETALK TITLEKiranjyot Gill 03/17/2017March LVC 2017
Let’s take a closer look…
NAME
EVENT NAME
TALK TITLE DATENAME DATETALK TITLEKiranjyot Gill 03/17/2017March LVC 2017
Let’s take a closer look…
BW does not reconstruct!
NAME
EVENT NAME
TALK TITLE DATENAME DATETALK TITLEKiranjyot Gill 03/17/2017March LVC 2017
Re-evaluate Waveform Reconstruction Efforts
Example studies:
Sch2
Yakunin+15
(ex: yak1)
BW does not reconstruct!
NAME
EVENT NAME
TALK TITLE DATENAME DATETALK TITLEKiranjyot Gill 03/17/2017March LVC 2017
Re-evaluate Waveform Reconstruction Efforts
(ex: yak2)
BW does not reconstruct!
A more worrisome reconstruction
example…
NAME
EVENT NAME
TALK TITLE DATENAME DATETALK TITLEKiranjyot Gill 03/17/2017March LVC 2017
Key Introduction to Existing BW Code
Linear polarization v Non-linear polarization
Original BW code was catered toward IMBH searches ✴ code assumed elliptical polarization
For the SN searches, we set ε = 0 for the linear polarized wf models (i.e., Dim)
✴ nice approximation for the initial stages of the rapidly-rotating (RR) wf models
✴ cannot make the same assumption for the later stages of the same RR models as we do not know the behavior of the waveform in its later stages (no simulation group has computed that far out yet that we know of) - via talks with Radice
NAME
EVENT NAME
TALK TITLE DATENAME DATETALK TITLEKiranjyot Gill 03/17/2017March LVC 2017
Key Introduction to Existing BW Code
Linear polarization v Non-linear polarization
BW assumes that all signals are elliptically polarized i.e. h× = εh+eiπ/2
✴ where ε ∈ [0, 1] is the ellipticity parameter ✴ 0 - linearly polarized signals ✴ 1 - circularly polarized signals
LIGO-P1600181
For linearly polarized waveforms, with either the + or × component, would be detectable within a LIGO-only network, and therefore made the elliptical
constraint a fair approximation.
NAME
EVENT NAME
TALK TITLE DATENAME DATETALK TITLEKiranjyot Gill 03/17/2017March LVC 2017
Key Introduction to Existing BW Code
Linear polarization v Non-linear polarization
LIGO-P1600181
it is not universally applicable in the case of SNe since the focus of our study is
more on realistic and phenomenological waveforms, which are not all linearly
polarized (such as Mueller 2012). Introduction of non-linear polarization capabilities
hardcoded into the BW pipeline
NAME
EVENT NAME
TALK TITLE DATENAME DATETALK TITLEKiranjyot Gill 03/17/2017March LVC 2017
Current CCSNe-focused BW Testing
List of Priors to be modified: ✴ Sky Location ✴ Glitch SNR ✴ Signal SNR ✴ Number of wavelets ✴ Waveform Type ✴ Clustering
(currently being tested with Tyson)
(currently being tested)
(currently being tested)(Done)
(Done)
(Done)
The quest to maximize the estimation of appropriate
parameters of the waveforms of
interest
NAME
EVENT NAME
TALK TITLE DATENAME DATETALK TITLEKiranjyot Gill 03/17/2017March LVC 2017
Extra Slides
NAME
EVENT NAME
TALK TITLE DATENAME DATETALK TITLEKiranjyot Gill 03/17/2017March LVC 2017
10�8 10�7 10�6 10�5 10�4 10�30.35
0.40
0.45
0.50
0.55
FAR
Effic
ienc
y
cWB ROC[1]
cWB+BW ROC Improvement
[2]
Estimate efficiency for cWB+BW ROC with thresholds set in cWB+BW fixed by [1] & [2]
NAME
EVENT NAME
TALK TITLE DATENAME DATETALK TITLEKiranjyot Gill 03/17/2017March LVC 2017