Gravitational Wave Parameter Estimation · 40 Scientific American, October 2013 Photograph by Tktk...
Transcript of Gravitational Wave Parameter Estimation · 40 Scientific American, October 2013 Photograph by Tktk...
![Page 1: Gravitational Wave Parameter Estimation · 40 Scientific American, October 2013 Photograph by Tktk Tktk sad1013Ande3p.indd 40 8/21/13 1:12 PM Gravitational Wave Parameter Estimation](https://reader036.fdocuments.us/reader036/viewer/2022071011/5fc9f2b20e8032100673c5e9/html5/thumbnails/1.jpg)
40 Scientific American, October 2013 Photograph by Tktk Tktk
sad1013Ande3p.indd 40 8/21/13 1:12 PM
Gravitational Wave Parameter Estimation
Neil J. Cornish
![Page 2: Gravitational Wave Parameter Estimation · 40 Scientific American, October 2013 Photograph by Tktk Tktk sad1013Ande3p.indd 40 8/21/13 1:12 PM Gravitational Wave Parameter Estimation](https://reader036.fdocuments.us/reader036/viewer/2022071011/5fc9f2b20e8032100673c5e9/html5/thumbnails/2.jpg)
Outline• Signal analysis 101
• Bayesian Inference
• Bayesian Hierachical Modeling
• Noise modeling
• Tools of the trade
• Examples from LIGO/Virgo
![Page 3: Gravitational Wave Parameter Estimation · 40 Scientific American, October 2013 Photograph by Tktk Tktk sad1013Ande3p.indd 40 8/21/13 1:12 PM Gravitational Wave Parameter Estimation](https://reader036.fdocuments.us/reader036/viewer/2022071011/5fc9f2b20e8032100673c5e9/html5/thumbnails/3.jpg)
Signals in additive noise
data = signal + noise
d = h+ n
(randomly selected aLIGO data from 14 September 2015)
![Page 4: Gravitational Wave Parameter Estimation · 40 Scientific American, October 2013 Photograph by Tktk Tktk sad1013Ande3p.indd 40 8/21/13 1:12 PM Gravitational Wave Parameter Estimation](https://reader036.fdocuments.us/reader036/viewer/2022071011/5fc9f2b20e8032100673c5e9/html5/thumbnails/4.jpg)
data - signal = noise
d� h = n
p(d|h) = p(d� h) = p(n)
Its all about the residuals
![Page 5: Gravitational Wave Parameter Estimation · 40 Scientific American, October 2013 Photograph by Tktk Tktk sad1013Ande3p.indd 40 8/21/13 1:12 PM Gravitational Wave Parameter Estimation](https://reader036.fdocuments.us/reader036/viewer/2022071011/5fc9f2b20e8032100673c5e9/html5/thumbnails/5.jpg)
Noise models matter
p(d|h) = p(d� h) = p(n)
The likelihood is our statistical model for the noise
An incorrect noise model will bias the analysis. e.g. assuming that noise is stationary and Gaussian when it is not.
![Page 6: Gravitational Wave Parameter Estimation · 40 Scientific American, October 2013 Photograph by Tktk Tktk sad1013Ande3p.indd 40 8/21/13 1:12 PM Gravitational Wave Parameter Estimation](https://reader036.fdocuments.us/reader036/viewer/2022071011/5fc9f2b20e8032100673c5e9/html5/thumbnails/6.jpg)
Noise Models
p(d|h) = p(d� h) = p(n)
Example: Stationary, colored, Gaussian noise
p(n, Sn(f)) =Y
f
1
2⇡Sn(f)e�nf n⇤
f
Sn(f)
Note: This is a parametrized model, depends on the unknown spectrum Sn(f)
![Page 7: Gravitational Wave Parameter Estimation · 40 Scientific American, October 2013 Photograph by Tktk Tktk sad1013Ande3p.indd 40 8/21/13 1:12 PM Gravitational Wave Parameter Estimation](https://reader036.fdocuments.us/reader036/viewer/2022071011/5fc9f2b20e8032100673c5e9/html5/thumbnails/7.jpg)
Bayesian Inference
Noise model Signal model
Normalization - model evidence
Posterior probability forwaveform h
p(h|d) = p(d|h)p(h)p(d)
![Page 8: Gravitational Wave Parameter Estimation · 40 Scientific American, October 2013 Photograph by Tktk Tktk sad1013Ande3p.indd 40 8/21/13 1:12 PM Gravitational Wave Parameter Estimation](https://reader036.fdocuments.us/reader036/viewer/2022071011/5fc9f2b20e8032100673c5e9/html5/thumbnails/8.jpg)
Reconstructing GW150914 with wavelets
![Page 9: Gravitational Wave Parameter Estimation · 40 Scientific American, October 2013 Photograph by Tktk Tktk sad1013Ande3p.indd 40 8/21/13 1:12 PM Gravitational Wave Parameter Estimation](https://reader036.fdocuments.us/reader036/viewer/2022071011/5fc9f2b20e8032100673c5e9/html5/thumbnails/9.jpg)
Gravitational wave signal typesWell modeled - e.g. binary inspiral and merger
Poorly modeled - e.g. core collapse supernovae Stochastic- e.g. phase transition in early universe
p(h)
![Page 10: Gravitational Wave Parameter Estimation · 40 Scientific American, October 2013 Photograph by Tktk Tktk sad1013Ande3p.indd 40 8/21/13 1:12 PM Gravitational Wave Parameter Estimation](https://reader036.fdocuments.us/reader036/viewer/2022071011/5fc9f2b20e8032100673c5e9/html5/thumbnails/10.jpg)
Gravitational wave signal models
Template based
Burst signals
Stochastic signals
p(h) = �(h� h(~�)), p(~�)
p(h) = �(h�X
), p( )
p(h) =1p
det(2⇡Sh)e�
12 (h
†S�1h h), p(Sh)
![Page 11: Gravitational Wave Parameter Estimation · 40 Scientific American, October 2013 Photograph by Tktk Tktk sad1013Ande3p.indd 40 8/21/13 1:12 PM Gravitational Wave Parameter Estimation](https://reader036.fdocuments.us/reader036/viewer/2022071011/5fc9f2b20e8032100673c5e9/html5/thumbnails/11.jpg)
Bayesian Hierachical Modeling
p(h|d,Hi,✓) =p(d|h,Hi,✓)p(h|Hi,✓)
p(d|Hi,✓)Level 1 Inference
We have some model for the signal and the noise described by the parameters Hi ✓
The result of the analysis is a posterior distribution for the waveform, in addition to posterior distributions for the sky location, polarization and the noise properties
BayesWave performs Level 1 Inference
![Page 12: Gravitational Wave Parameter Estimation · 40 Scientific American, October 2013 Photograph by Tktk Tktk sad1013Ande3p.indd 40 8/21/13 1:12 PM Gravitational Wave Parameter Estimation](https://reader036.fdocuments.us/reader036/viewer/2022071011/5fc9f2b20e8032100673c5e9/html5/thumbnails/12.jpg)
Bayesian Hierachical Modeling
Level 2 Inference
But what if we are not really interested in the waveform samples ?h
p(d|Hi,✓) =
Zp(d|h,Hi,✓)p(h|Hi,✓)dhMarginalize over h
The marginal likelihood from Level 1 becomes the likelihood for Level 2
p(✓|d,Hi) =p(d|Hi,✓)p(✓|Hi)
p(d|Hi)
LALinference performs Level 2 Inference (Can recover Level 1 in post-processing)
![Page 13: Gravitational Wave Parameter Estimation · 40 Scientific American, October 2013 Photograph by Tktk Tktk sad1013Ande3p.indd 40 8/21/13 1:12 PM Gravitational Wave Parameter Estimation](https://reader036.fdocuments.us/reader036/viewer/2022071011/5fc9f2b20e8032100673c5e9/html5/thumbnails/13.jpg)
Bayesian Hierachical Modeling
Level 3 Inference
But what if we are not all that interested in the model parameters ?
Marginalize over
The marginal likelihood from Level 2 is the model evidence. It can be used for model selection
✓
✓ p(d|Hi) =
Zp(d|Hi,✓)p(✓|Hi)d✓
Oij =p(Hi|d)p(Hj |d) =
p(Hi)
p(Hj)
p(d|Hi)
p(d|Hj)
Prior Odds Bayes Factor
![Page 14: Gravitational Wave Parameter Estimation · 40 Scientific American, October 2013 Photograph by Tktk Tktk sad1013Ande3p.indd 40 8/21/13 1:12 PM Gravitational Wave Parameter Estimation](https://reader036.fdocuments.us/reader036/viewer/2022071011/5fc9f2b20e8032100673c5e9/html5/thumbnails/14.jpg)
Level 2 Inference: Template based analysis
These have the strongest priors and hence yield the most sensitive searches
Techniques such as MCMC and Nested Sampling can be used to map out the full posterior distribution, allowing us to compute mean, median and mode and credible intervals.
Integrating out the individual signal samples yields the marginal likelihood in terms of the model parameters
Likelihood
Evidence
Prior
p(h) = �(h� h(~�)), p(~�)
The marginal likelihood and the (hyper) prior on the model parameters then defines the posterior on the model parameters
p(d|~�) =Z
p(d|h)�(h� h(~�)) dh
p(~�|d) = p(d|~�)p(~�)p(d)
![Page 15: Gravitational Wave Parameter Estimation · 40 Scientific American, October 2013 Photograph by Tktk Tktk sad1013Ande3p.indd 40 8/21/13 1:12 PM Gravitational Wave Parameter Estimation](https://reader036.fdocuments.us/reader036/viewer/2022071011/5fc9f2b20e8032100673c5e9/html5/thumbnails/15.jpg)
Template based analysis - Parameter Posteriors
p(~�|d)
![Page 16: Gravitational Wave Parameter Estimation · 40 Scientific American, October 2013 Photograph by Tktk Tktk sad1013Ande3p.indd 40 8/21/13 1:12 PM Gravitational Wave Parameter Estimation](https://reader036.fdocuments.us/reader036/viewer/2022071011/5fc9f2b20e8032100673c5e9/html5/thumbnails/16.jpg)
Level 2 Inference: Stochastic Signals
Gaussian random waveform
Gaussian likelihood. The noise correlation matrix block diagonal between detectors
p(h) =1p
det(2⇡Sh)e�
12 (h
†S�1h h), p(Sh)
p(d|h) = 1pdet(2⇡C)
e�12 (d�h)†C�1(d�h)
Marginal likelihood
C(Ik)(Jl) = �IJ(Sn)kl
G(Ik)(Jl) = �IJ(Sn)kl + (Ik)(Jl)(Sh)
p(d|Sh) =
Zp(d|h)p(h) dh =
1pdet(2⇡G)
e�12 (d
†G�1d)
Overlap reduction function, Hellings-Downs curve[Cornish & Romano, PRD 87 122003 (2013)]
![Page 17: Gravitational Wave Parameter Estimation · 40 Scientific American, October 2013 Photograph by Tktk Tktk sad1013Ande3p.indd 40 8/21/13 1:12 PM Gravitational Wave Parameter Estimation](https://reader036.fdocuments.us/reader036/viewer/2022071011/5fc9f2b20e8032100673c5e9/html5/thumbnails/17.jpg)
Level 3 Inference: Rate of Black Hole Mergers
Likelihood for duty cycle
[Smith & Thrane, PRX kk (2018); Cornish, Physics (2018)]
Marginal Likelihood (Evidence) for Signal Model in data chunk j
Marginal Likelihood (Evidence) for Noise Model in data chunk j
Number of binary mergers
p(d|⇠) =MY
j=1
[⇠ p(dj |Signal) + (1� ⇠) p(dj |Noise)]
N = M⇠⇠Fraction of chunks with mergers
p(d|R) = MX
N
Zd⇠p (d|⇠)p(N)
e�RRN
N !Likelihood for merger rate
![Page 18: Gravitational Wave Parameter Estimation · 40 Scientific American, October 2013 Photograph by Tktk Tktk sad1013Ande3p.indd 40 8/21/13 1:12 PM Gravitational Wave Parameter Estimation](https://reader036.fdocuments.us/reader036/viewer/2022071011/5fc9f2b20e8032100673c5e9/html5/thumbnails/18.jpg)
Noise Models
Spectra computed using 8 second segments, spaced 2 seconds apart, covering 256 surrounding the BNS merger (de-glitched data)
GW170817: Observation of Gravitational Waves from a Binary Neutron Star Inspiral, PRL 119 161101 (2017)
![Page 19: Gravitational Wave Parameter Estimation · 40 Scientific American, October 2013 Photograph by Tktk Tktk sad1013Ande3p.indd 40 8/21/13 1:12 PM Gravitational Wave Parameter Estimation](https://reader036.fdocuments.us/reader036/viewer/2022071011/5fc9f2b20e8032100673c5e9/html5/thumbnails/19.jpg)
Contending with non-stationary, non-Gaussian noise: Traditional approach
Non-stationary:
Adiabatic drifts in the PSD - work with short data segments
Non-Gaussian:
Glitches - vetoes and time-slides
![Page 20: Gravitational Wave Parameter Estimation · 40 Scientific American, October 2013 Photograph by Tktk Tktk sad1013Ande3p.indd 40 8/21/13 1:12 PM Gravitational Wave Parameter Estimation](https://reader036.fdocuments.us/reader036/viewer/2022071011/5fc9f2b20e8032100673c5e9/html5/thumbnails/20.jpg)
• Bayesian model selection • Three part model (signal, glitches, Gaussian noise) • Trans-dimensional Markov Chain Monte Carlo
• Wavelet decomposition • Glitches modeled by wavelets • Number, amplitude, duration and location of wavelets varies
Contending with non-stationary, non-Gaussian noise: Bayesian approach
(Currently the model does not account for adiabatic drifts in the PSD) [ Cornish & Littenberg, Class. Quant. Grav. 32 135012 (2015)]
BayesWave
![Page 21: Gravitational Wave Parameter Estimation · 40 Scientific American, October 2013 Photograph by Tktk Tktk sad1013Ande3p.indd 40 8/21/13 1:12 PM Gravitational Wave Parameter Estimation](https://reader036.fdocuments.us/reader036/viewer/2022071011/5fc9f2b20e8032100673c5e9/html5/thumbnails/21.jpg)
Spectral Modeling
10-48
10-46
10-44
10-42
10-40
10-38
10-36
50 100 150 200 250 300 350 400 450
Pow
er s
pect
ral d
ensi
ty (H
z-1)
frequency (Hz)
|h(f)|2
Cubic spline fit
Lorentzian fit
Data
Cubic spline fit
Lorentzian fit
Model it
![Page 22: Gravitational Wave Parameter Estimation · 40 Scientific American, October 2013 Photograph by Tktk Tktk sad1013Ande3p.indd 40 8/21/13 1:12 PM Gravitational Wave Parameter Estimation](https://reader036.fdocuments.us/reader036/viewer/2022071011/5fc9f2b20e8032100673c5e9/html5/thumbnails/22.jpg)
Time-Frequency Scalograms of LIGO data
Glitches
Model these too
X
![Page 23: Gravitational Wave Parameter Estimation · 40 Scientific American, October 2013 Photograph by Tktk Tktk sad1013Ande3p.indd 40 8/21/13 1:12 PM Gravitational Wave Parameter Estimation](https://reader036.fdocuments.us/reader036/viewer/2022071011/5fc9f2b20e8032100673c5e9/html5/thumbnails/23.jpg)
Trans-dimensional Markov Chain Monte Carlo
![Page 24: Gravitational Wave Parameter Estimation · 40 Scientific American, October 2013 Photograph by Tktk Tktk sad1013Ande3p.indd 40 8/21/13 1:12 PM Gravitational Wave Parameter Estimation](https://reader036.fdocuments.us/reader036/viewer/2022071011/5fc9f2b20e8032100673c5e9/html5/thumbnails/24.jpg)
Example from LIGO S5 science run
![Page 25: Gravitational Wave Parameter Estimation · 40 Scientific American, October 2013 Photograph by Tktk Tktk sad1013Ande3p.indd 40 8/21/13 1:12 PM Gravitational Wave Parameter Estimation](https://reader036.fdocuments.us/reader036/viewer/2022071011/5fc9f2b20e8032100673c5e9/html5/thumbnails/25.jpg)
Removing a glitch was preferable to vetoing GW170817
![Page 26: Gravitational Wave Parameter Estimation · 40 Scientific American, October 2013 Photograph by Tktk Tktk sad1013Ande3p.indd 40 8/21/13 1:12 PM Gravitational Wave Parameter Estimation](https://reader036.fdocuments.us/reader036/viewer/2022071011/5fc9f2b20e8032100673c5e9/html5/thumbnails/26.jpg)
Likelihood for non-stationary Gaussian Noise
p(d|h) = 1pdet(2⇡C)
e�12 (d�h)†C�1(d�h)
Cost of computing the likelihood is far less in a representation where the noise correlation matrix C is diagonal
e.g. Colored stationary noise has a diagonal noise correlation matrix in the Fourier domain
Pulsar Timing has to deal with colored, non-stationary data and un-even sampling - analysis performed directly in the time domain. Clever tricks have been developed to speed up the costly matrix inversions and sums
Adiabatic drifts in the PSD - Locally Stationary Noise
![Page 27: Gravitational Wave Parameter Estimation · 40 Scientific American, October 2013 Photograph by Tktk Tktk sad1013Ande3p.indd 40 8/21/13 1:12 PM Gravitational Wave Parameter Estimation](https://reader036.fdocuments.us/reader036/viewer/2022071011/5fc9f2b20e8032100673c5e9/html5/thumbnails/27.jpg)
Likelihood for Non-stationary Gaussian Noise
[2. “Wavelet processes and adaptive estimation of the evolutionary wavelet spectrum”, Nason, von Sachs, & Kroisandt, J. R. Statist. Soc. Series B62, 271 (2000)]
[1. “Fitting time series models to nonstationary processes”. Dahlhaus, Ann. Statist., 25, 1 (1997)]
p(d|h) = 1pdet(2⇡C)
e�12 (d�h)†C�1(d�h)
Cost of computing the likelihood is far less in a representation where the noise correlation matrix C is diagonal
For a large class of discrete wavelet transformations and locally stationary noise
C(i,j)(k,l) = �ij�kl Cik
[1]
Time
[2]
Frequency
This is the likelihood used by the LIGO coherent WaveBurst algorithm
![Page 28: Gravitational Wave Parameter Estimation · 40 Scientific American, October 2013 Photograph by Tktk Tktk sad1013Ande3p.indd 40 8/21/13 1:12 PM Gravitational Wave Parameter Estimation](https://reader036.fdocuments.us/reader036/viewer/2022071011/5fc9f2b20e8032100673c5e9/html5/thumbnails/28.jpg)
Likelihood for Non-stationary Gaussian Noise
p(d|h) = 1pdet(2⇡C)
e�12 (d�h)†C�1(d�h)
C(i,j)(k,l) = �ij�kl CikCan we use a discrete wavelet based likelihood?
Model the wavelet spectrum as a smooth function in frequency and time. E.g. Trans-dimensional Bicubic spline
Cik
![Page 29: Gravitational Wave Parameter Estimation · 40 Scientific American, October 2013 Photograph by Tktk Tktk sad1013Ande3p.indd 40 8/21/13 1:12 PM Gravitational Wave Parameter Estimation](https://reader036.fdocuments.us/reader036/viewer/2022071011/5fc9f2b20e8032100673c5e9/html5/thumbnails/29.jpg)
Likelihood for Non-stationary Gaussian Noise
p(d|h) = 1pdet(2⇡C)
e�12 (d�h)†C�1(d�h)
Need fast wavelet transforms of the signals for computational efficiency
Use SPA to derive analytic wavelet domain waveforms?
Only compute wavelets along predicted t-f track?
Reduce order modeling?
C(i,j)(k,l) = �ij�kl CikCan we use a discrete wavelet based likelihood?
![Page 30: Gravitational Wave Parameter Estimation · 40 Scientific American, October 2013 Photograph by Tktk Tktk sad1013Ande3p.indd 40 8/21/13 1:12 PM Gravitational Wave Parameter Estimation](https://reader036.fdocuments.us/reader036/viewer/2022071011/5fc9f2b20e8032100673c5e9/html5/thumbnails/30.jpg)
Bayesian Inference: Tools of the Trade
Prior Likelihood
Posterior Evidence
MCMC
p(h|M)
p(h|d,M) p(d|M)
p(d|h)
![Page 31: Gravitational Wave Parameter Estimation · 40 Scientific American, October 2013 Photograph by Tktk Tktk sad1013Ande3p.indd 40 8/21/13 1:12 PM Gravitational Wave Parameter Estimation](https://reader036.fdocuments.us/reader036/viewer/2022071011/5fc9f2b20e8032100673c5e9/html5/thumbnails/31.jpg)
Markov Chain Monte Carlo
⇥x
⇥y
H
Transition Probability(Metropolis-Hastings)
Prior Proposal
Likelihood
Always go up,Sometime come down
H = min�
1,p(�y)p(d|�y)q(�x|�y)p(�x)p(d|�x)q(�y|�x)
�
Yields PDF p(�x|d) for parameters�x given data d
![Page 32: Gravitational Wave Parameter Estimation · 40 Scientific American, October 2013 Photograph by Tktk Tktk sad1013Ande3p.indd 40 8/21/13 1:12 PM Gravitational Wave Parameter Estimation](https://reader036.fdocuments.us/reader036/viewer/2022071011/5fc9f2b20e8032100673c5e9/html5/thumbnails/32.jpg)
MCMC Recipe
Ingredients:
Local posterior approximation
Differential evolution proposals
Parallel tempering
Global likelihood maps
Directions:Mix all the proposals together. Check consistency by recovering the prior and producing diagonal PP plots. Results are ready when distributions are stationary.
![Page 33: Gravitational Wave Parameter Estimation · 40 Scientific American, October 2013 Photograph by Tktk Tktk sad1013Ande3p.indd 40 8/21/13 1:12 PM Gravitational Wave Parameter Estimation](https://reader036.fdocuments.us/reader036/viewer/2022071011/5fc9f2b20e8032100673c5e9/html5/thumbnails/33.jpg)
Proposal Distributions
Propose jumps along eigendirections of K, scaled by eigenvalues
Quadratic approximation to the posterior using the augmented Fisher Information Matrix
Use a Non-Markovian Pilot search (hill climbers, simulated annealing, genetic algorithms etc) to crudely map the posterior/likelihood and use this as a proposal distribution for a Markovian follow-up [Littenberg & Cornish, PRD 80, 063007, (2009)]
Local posterior approximation
Global likelihood maps
q(~y|~x) = 1pdet(2⇡K�1)
e
� 12Kij(x
i�y
i)(xj�y
j)
Time-frequency maps, Maximized likelihood maps
![Page 34: Gravitational Wave Parameter Estimation · 40 Scientific American, October 2013 Photograph by Tktk Tktk sad1013Ande3p.indd 40 8/21/13 1:12 PM Gravitational Wave Parameter Estimation](https://reader036.fdocuments.us/reader036/viewer/2022071011/5fc9f2b20e8032100673c5e9/html5/thumbnails/34.jpg)
BayesWave Global Map Proposal
![Page 35: Gravitational Wave Parameter Estimation · 40 Scientific American, October 2013 Photograph by Tktk Tktk sad1013Ande3p.indd 40 8/21/13 1:12 PM Gravitational Wave Parameter Estimation](https://reader036.fdocuments.us/reader036/viewer/2022071011/5fc9f2b20e8032100673c5e9/html5/thumbnails/35.jpg)
Proposal Distributions
[Braak (2005)]Differential evolution
![Page 36: Gravitational Wave Parameter Estimation · 40 Scientific American, October 2013 Photograph by Tktk Tktk sad1013Ande3p.indd 40 8/21/13 1:12 PM Gravitational Wave Parameter Estimation](https://reader036.fdocuments.us/reader036/viewer/2022071011/5fc9f2b20e8032100673c5e9/html5/thumbnails/36.jpg)
Parallel Tempering
Primary Mode
Secondary Mode
[Swendsen & Wang, 1986]
Ordinary MCMC techniques side-step the need to compute the evidence. PT uses multiple, coupled chains to improve mixing, and also allows the evidence to be computed.
⇡(~�|d)T = p(d|~�)1/T p(~�)
Explore tempered posterior
log p(d) =
Z 1
0E[log p(d|~�)]� d�
Compute model evidence
(Here )� = 1T
![Page 37: Gravitational Wave Parameter Estimation · 40 Scientific American, October 2013 Photograph by Tktk Tktk sad1013Ande3p.indd 40 8/21/13 1:12 PM Gravitational Wave Parameter Estimation](https://reader036.fdocuments.us/reader036/viewer/2022071011/5fc9f2b20e8032100673c5e9/html5/thumbnails/37.jpg)
MCMC
![Page 38: Gravitational Wave Parameter Estimation · 40 Scientific American, October 2013 Photograph by Tktk Tktk sad1013Ande3p.indd 40 8/21/13 1:12 PM Gravitational Wave Parameter Estimation](https://reader036.fdocuments.us/reader036/viewer/2022071011/5fc9f2b20e8032100673c5e9/html5/thumbnails/38.jpg)
Parallel Tempering
![Page 39: Gravitational Wave Parameter Estimation · 40 Scientific American, October 2013 Photograph by Tktk Tktk sad1013Ande3p.indd 40 8/21/13 1:12 PM Gravitational Wave Parameter Estimation](https://reader036.fdocuments.us/reader036/viewer/2022071011/5fc9f2b20e8032100673c5e9/html5/thumbnails/39.jpg)
Posterior Predictive Checks
P-P plots, BW Sky localization Gaussianity of BW residuals (noise model)
![Page 40: Gravitational Wave Parameter Estimation · 40 Scientific American, October 2013 Photograph by Tktk Tktk sad1013Ande3p.indd 40 8/21/13 1:12 PM Gravitational Wave Parameter Estimation](https://reader036.fdocuments.us/reader036/viewer/2022071011/5fc9f2b20e8032100673c5e9/html5/thumbnails/40.jpg)
GW170104
Parameter estimation: Examples from LIGO/Virgo
mostly measuring M
mostly measuring M
![Page 41: Gravitational Wave Parameter Estimation · 40 Scientific American, October 2013 Photograph by Tktk Tktk sad1013Ande3p.indd 40 8/21/13 1:12 PM Gravitational Wave Parameter Estimation](https://reader036.fdocuments.us/reader036/viewer/2022071011/5fc9f2b20e8032100673c5e9/html5/thumbnails/41.jpg)
follows line of constant M
BNS GW170817 - Parameter estimation
![Page 42: Gravitational Wave Parameter Estimation · 40 Scientific American, October 2013 Photograph by Tktk Tktk sad1013Ande3p.indd 40 8/21/13 1:12 PM Gravitational Wave Parameter Estimation](https://reader036.fdocuments.us/reader036/viewer/2022071011/5fc9f2b20e8032100673c5e9/html5/thumbnails/42.jpg)
h(f) = A�(f)ei��(f)
Dominant Harmonic
�2(f) = 2�ftc � �c ��
4+
3128 u5
�
k=0
(�k(��) + �lk(��) ln u)uk
A2(f) =M2
DL u7/2
�
k=0
(�k(��) + �lk(��) ln u)uk
u = (�Mf)1/3 � v
Why measuring BH spin is hard I: Information from the Phase
![Page 43: Gravitational Wave Parameter Estimation · 40 Scientific American, October 2013 Photograph by Tktk Tktk sad1013Ande3p.indd 40 8/21/13 1:12 PM Gravitational Wave Parameter Estimation](https://reader036.fdocuments.us/reader036/viewer/2022071011/5fc9f2b20e8032100673c5e9/html5/thumbnails/43.jpg)
0PN
1PN
1.5PN
2PN
�371532256
+ �55384
���2/5u�3
3128
u�5 Measure chirp mass
Measure individual masses
Measure spin combination
Measure individual spins
��
3�
8� 1
32
�113(1 ±
�1� 4�)� 76�
�L · ��1,2
���3/5 u�2
�1529336521676032
+2714521504
� +30853072
�2 + �(L · ��1,2, ��1 · ��2, �21,2)
���4/5 u�1
Post-Newtonian Expansion u = (�Mf)1/3 � v
![Page 44: Gravitational Wave Parameter Estimation · 40 Scientific American, October 2013 Photograph by Tktk Tktk sad1013Ande3p.indd 40 8/21/13 1:12 PM Gravitational Wave Parameter Estimation](https://reader036.fdocuments.us/reader036/viewer/2022071011/5fc9f2b20e8032100673c5e9/html5/thumbnails/44.jpg)
�e↵ = m1 �1 cos ✓LS1 +m2 �2 cos ✓LS2
Out-of-plane spin combination
�p =1
2(�2? + ↵�1? + |�2? � ↵�1?|)
↵ =
✓m1
m2
◆(4M �m2)
(4M �m1)
In-plane spin combination
[component (anti)aligned with angular momentum]
Why measuring BH spin is hard II: Information from precision
![Page 45: Gravitational Wave Parameter Estimation · 40 Scientific American, October 2013 Photograph by Tktk Tktk sad1013Ande3p.indd 40 8/21/13 1:12 PM Gravitational Wave Parameter Estimation](https://reader036.fdocuments.us/reader036/viewer/2022071011/5fc9f2b20e8032100673c5e9/html5/thumbnails/45.jpg)
�
���
���
���
���
���
���
���
� ��� � ��� � ��� �
��������������������
�����������
Face-on Face-offEdge-on
We are more likely to detect Face-on/off systems than Edge-on systems
We are also more likely to detect more massive systems 70 Mpc for 1.4+1.4 Msun 300 Mpc for 10+10 Msun 700 Mpc for 30+30 Msun
Selection Effects
More massive, less cycles in-band, Harder to measure precession
Harder to measure precession
![Page 46: Gravitational Wave Parameter Estimation · 40 Scientific American, October 2013 Photograph by Tktk Tktk sad1013Ande3p.indd 40 8/21/13 1:12 PM Gravitational Wave Parameter Estimation](https://reader036.fdocuments.us/reader036/viewer/2022071011/5fc9f2b20e8032100673c5e9/html5/thumbnails/46.jpg)
�e↵
�p
Spin posteriors for GW170104
![Page 47: Gravitational Wave Parameter Estimation · 40 Scientific American, October 2013 Photograph by Tktk Tktk sad1013Ande3p.indd 40 8/21/13 1:12 PM Gravitational Wave Parameter Estimation](https://reader036.fdocuments.us/reader036/viewer/2022071011/5fc9f2b20e8032100673c5e9/html5/thumbnails/47.jpg)
|�e↵ | < 0.35
All four events consistent with low spin, no precession
90% confidence, all 4
�p ⇠ 0
Evidence for spin precession remains elusive
�e↵