Nicolas Yunes Montana State University
Testing Gravity 2015, January 15th, 2015 Yunes & Siemens, Living Reviews in Relativity 2014,
http://arxiv.org/abs/1304.3473
Gravitational Waves as Probes of Extreme Gravity
GW Probes of Extreme Gravity Yunes
An incomplete summary of what GWs will tell us about gravity
Clifford Will, Jim Gates, Stephon Alexander, Abhay Ashtekar, Sam Finn, Ben Owen, Pablo Laguna, Emanuele Berti, Uli Sperhake, Dimitrios Psaltis, Avi Loeb, Vitor Cardoso, Leonardo Gualtieri, Daniel Grumiller, David Spergel,
Frans Pretorius, Neil Cornish, Scott Hughes, Carlos Sopuerta, Takahiro Tanaka, Jon Gair, Paolo Pani, Antoine Klein, Kent Yagi, Laura Sampson, Luis
Lehner, Masaru Shibata, Curt Cutler, Haris Apostolatos,
2
Leo Stein, Sarah Vigeland, Katerina Chatziioannou, Philippe Jetzer, Leor Barack, Kostas Glampedakis, Stanislav Babak, Ilya Mandel, Chao Li, Eliu
Huerta, Chris Berry, Alberto Sesana, Carl Rodriguez, Georgios Lukes-Gerakopoulos, George Contopoulus, Chris van den Broeck, Walter del Pozzo,
Jon Veitch, Nathan Collins, Deirdre Shoemaker, Sathyaprakash, Devin Hansen, Enrico Barausse, Carlos Palenzuela, Marcelo Ponce, etc.
Standing on the Shoulders of...
YunesGW Probes of Extreme Gravity
Roadmap
3
What will we learn from GW tests of GR?
How do we use GWs to test GR?
How do GW tests differ from other
tests?
YunesGW Probes of Extreme Gravity 4
10-12 10-11 10-10 10-9 10-8 10-7 10-6 10-5 10-4 10-3 10-2 10-1 100
ε=M/r10-1310-1210-1110-1010-910-810-710-610-510-410-310-210-1
ξ1/2 =(
M/r3 )1/
2 [km
-1]
Double Binary Pulsar
Lunar Laser Ranging
LIGO BH-BH Merger
Sun's SurfaceEarth's Surface
LISA IMBH-IMBH Merger
Perihelion Precession of Mercury
LIGO NS-NS Merger
IMRIs IMBH-SCO
LAGEOSLISA SMBH-SMBH Merger
EMRIs SMBH-SCO
Pulsar Timing Arrays
1. Extreme Gravity:
[Baker, et al, Psaltis LRR]
How do GW tests Differ from Other Tests?
Sources: Compact Object Coalescence Supernova, deformed NSs, etc. (excluding pulsar timing in this talk)
Phases: Late Inspiral, Merger, Ringdown.
2. Clean:
Processes: Generation and Propagation of metric perturbation.
Absorption is negligible, lensing unimportant at low z, accretion disk and magnetic fields unimportant during inspiral.
YunesGW Probes of Extreme Gravity 5
4. Constraint Maps:
How do GW tests Differ from Other Tests?
If large # of sources detected. eg. preferred position tests.
5. Very Local Universe: z < 0.07 or D < 300 Mpc for NS/NS inspiral.
3. Localized: Distinct point sources in spacetime (not a background)
GW Probes of Extreme Gravity Yunes 6
[C. Hanna, PSU]
signal-to-noise ratio
(SNR)
detector noise (spectral noise
density)
data
template (projection of GW metric perturbation)
template param that characterize system
Matched Filtering:
⇢2 ⇠Z
s(f)h(f,�µ)
Sn(f)df
How do use GWs to test GR? Matched Filtering
• Create template “filters”
• Cross-correlate filters & data
• Find filter that maximizes the cross-correlation.
GW Probes of Extreme Gravity Yunes 7
80 100 120 140 160 180 200 220 240 260 280 300t/M
-2
-1
0
1
2
h+
InspiralMerger Ring down
Post-Newtonian
Num. Rel.BH Pert. Theory
gravitational wave
symmetric mass ratio distance to
the sourceinclination
angletotal mass
orbital freq.
orbital phase
h⇥(t) ⇠⌘M
DLcos ◆ (M!)2/3 cos 2�+ . . .
How do use GWs to test GR? Source Modeling
Inspiral: thousands of cycles, most SNR at low masses.
Approximations: PN + PM
Accuracy: 3.5 PN (“3 loop” order)
Template:
YunesGW Probes of Extreme Gravity 8
How do use GWs to test GR?
Top-Down (test specific theory) vs. Bottom-Up (search for deviations).
hBD(f ;~�GR,�BD)
”◆0”
hD>4(f ;~�GR,�D>4)
hLV (f ;~�GR,�LV )
hppE(f ;~�GR,~�ppE)
1PN
2PN
3PN
4PN
-1PN
-2PN
-3PN
-4PN
0.5PN
1.5PN
2.5PN
3.5PN
-0.5PN
-1.5PN
-2.5PN
-3.5PN
0PNCurrent
ConstraintsGW
Constraints
GRBD MG
EDGB
CS
Gdot
LV
[Yunes & Pretorius, PRD 2009, Mirshekari, Yunes & Will, PRD 2012, Chatziioannou, Yunes & Cornish, PRD 2012]
YunesGW Probes of Extreme Gravity 10
Templates/Theories GR ppE
GR Business as usualQuantify the statistical significance that
the detected event is within GR. Anomalies?
Not GRQuantify fundamental bias
introduced by filtering non-GR events with GR templates
Can we measure deviations from GR characterized by non-GR signals?
Model Evidence.
[Yunes & Pretorius, PRD 2009, Chatziioannou, Yunes & Cornish, PRD 2012]
What will we learn from GW tests of GR?
Search for Generic Deviations: Parameterize post-Einsteinian (ppE)
YunesGW Probes of Extreme Gravity 11
What will we learn from GW tests of GR?
1. Gravitational Lorentz Violation: Primarily from propagation speed w/coincident EM
2. Graviton Mass: Primarily from modification of dispersion relation.
3. Dipolar Emission: From activation of scalar or vectorial modes.
[Nishizawa & Nakamura, 2014, Jacobson, 2004, Yagi, Blas, Barausse, Yunes, PRD 2013, Hansen, Yunes, Yagi, 2014]
[Will, PRD 1994, Yagi, et al PRL 2013, Hansen, Yunes, Yagi 2014]
[Finn & Sutton PRD 2002, Baskaran, et al, PRD 2008, Will, PRD 1998, Will & Yunes, CQG 2004, Berti, Buonanno & Will, CQG 2005]
YunesGW Probes of Extreme Gravity 12
What will we learn from GW tests of GR?
4. Higher Curvature Action: Effective theories (EDGB, CS)
5. Screening Strong-Field Mechanism: Scalarization
6. No-Hair Theorem: From binary black hole ringdown. (more difficult, requires high SNR)
[Damour & Esposito-Farese, CQG, 1992, Freire et al, MNRAS 2012, Sampson et al, PRD 2014]
[Alexander & Yunes, Phys. Rept, 2009 Yagi, PRD 2012, Yagi, et al, PRD 2012]
[Dreyer, et al, CQG, 2004, Berti et al, PRD 2006, Gossan et al, PRD 2012]
YunesGW Probes of Extreme Gravity 13
GW tests will constrain a variety of phenomena: Lorentz violation, graviton mass, dipole emission, higher curvature
action, screening mechanisms, no-hair theorem.
GW tests of GR differ from other tests in a variety of ways: probe extreme gravity, clean, localized, constraint maps, present day.
Doveryai, no proveryai
What does it all mean?
GW Probes of Extreme Gravity Yunes
What will we learn from GW tests of GR?
14
Nico’s (GW-Biased) GW Modified Theory Classification:
Nico’s (GW-Biased) Cosmological Modified Theory Classification:
Screened Unscreened
Late-time expansion, DEEg, chameleon, Vainshtein, etc.
Early-time cosmology, inflationEg, Chern-Simons, Gauss-Bonnet, etc.
“Weak Field” Strong-Field
Well-constrained by binary pulsars, so need screeningEg, Scalar-Tensor theories
Constrainable with GW observations, natural suppression without screening Eg, Chern-Simons, Gauss-Bonnet, etc.
GW Probes of Extreme Gravity Yunes
Screening in Cosmology ≉ Screening in GWs
15
Weak Field, Low Energy
In Cosmology
Strong Field, High Energy
Solar System
Binary Black Hole
Mergers
Galactic Dynamics
GRNot GR
In Gravitational Wave Physics
Solar System
Binary Black Hole
MergersNot GRGR
YunesGW Probes of Extreme Gravity 16
Weak Field Theories
GW Probes of Extreme Gravity Yunes
Example: Scalar Tensor Theories
17
Definition:
Main Effect:
Dominant Observables:
Spontaneous Scalarization
Stars acquire scalar charge +
Grav. and Inertial center of mass do not coincide
Faster Orbital Decay
Damour+Esposito-Farese, PRD 54 (’96) Palenzuela, et al, PRD 97 (’13), 89 (’14).
Screened Dipole Gravitational Wave
Emission
Effective Coupling to Matter:
GW Probes of Extreme Gravity Yunes
Constraints on Weak Field Theories
18
Scalarizable Scalar-Tensor:
(similar constraints for TeVeS and for massive Brans-Dicke) Freire, et al, MRAS 18 (’12).
Alsing, et al, PRD 85, (’12).
YunesGW Probes of Extreme Gravity 19
Strong Field Theories
GW Probes of Extreme Gravity Yunes
Example: Quadratic Gravity
20
Definition:
Main Effects:
Dominant Observables:
Chirping of Gravitational Wave Phase
Alexander & Yunes, Phys. Rept 480 (’09) Yunes & Stein, PRD 83 (’11)
certain choices of couplings lead to Einstein-Dilaton-Gauss-Bonnet theory or dynamical Chern-Simons gravity.
dCS. Gravitational Parity Violation, inverse no-hair theorem.
Requires observation of late inspiral & merger
GW Probes of Extreme Gravity Yunes
Constraints on Strong Field Theories
21
Extremely weak from Solar System (GPB)
dCS
Projected GW constraints
Yagi, Yunes & Tanaka, PRL 109 (’12)
Current constraints
Constraint Contours on
in km.
YunesGW Probes of Extreme Gravity 22
Parametrized Post-Einsteinian
YunesGW Probes of Extreme Gravity 23
Strong FieldWeak Field
GR Signal/ppE Templates, 3-sigma constraints, SNR = 20
Yunes & Hughes, 2010, Cornish, Sampson, Yunes & Pretorius, 2011 Sampson, Cornish, Yunes 2013.
Newt 1PN 1.5 2 2.5 3 3.5
aLIGO projected bounds
Double Binary Pulsar bounds
Projected Gravitational Wave Constraints
h(f) = hGR(f) (1 + ↵fa) ei�fb
GW Probes of Extreme Gravity Yunes
At our doorstep...
• is the Pontryagin density.
• is either a dynamical field that evolves.
24
Firs
t Loc
k
LIG
O In
dia?
aLIG
Oac
cept
ance
aLIG
Oin
stal
latio
nst
art
Stra
in N
oise
[1/s
qrt(
Hz)
]
S5 S6
YearN
ow
Earl
y ru
ns:
LIG
O O
nly?
Full
Lock
Adv
Kag
ra
Virg
o
YunesGW Probes of Extreme Gravity 25
Non GR injection, extracted with GR templates (blue) and ppE templates (red). GR template extraction is “wrong” by much more than the systematic
(statistical) error. “Fundamental Bias”
Non-GR Signal/GR Templates, SNR = 20
12
0.01
0.1
10 20 30 40 50
beta
unc
erta
inty
SNR
actual values1/SNR
FIG. 14: The scaling of the parameter estimation error inthe ppE parameter β for an aLIGO simulation with ppE pa-rameters (a,α, b,β) = (0, 0,−1.25, 0.1). The parameter errorsfollow the usual 1/SNR scaling.
0
0.01
0.02
0.03
0.04
0.05
0.06
0.005 0.01 0.015 0.02 0.025 0.03
1
100
10000
1e+06
1.0
- FF
Baye
s Fa
ctor
β
Bayes FactorFitting Factor
FIG. 15: The log Bayes factors and (1 − FF) plotted as afunction of β for a ppE injection with parameters (a,α, b,β) =(0, 0,−1.25,β). The predicted link between the fitting factorand Bayes factor is clearly apparent.
the log Bayes factor is equal to
logB = χ2min/2 + ∆ logO
= (1− FF2)SNR2
2+ ∆ logO . (25)
Thus, up to the difference in the log Occam factors,∆ logO, the log Bayes factor should scale as 1−FF whenFF ∼ 1. This link is confirmed in Figure 15.
E. Parameter Biases
If we assume that nature is described by GR, but intruth another theory is correct, this will result in therecovery of the wrong parameters for the systems we arestudying. For instance, when looking at a signal thathas non-zero ppE phase parameters, a search using GRtemplates will return the incorrect mass parameters, asillustrated below.
2.8 2.82 2.84 2.86 2.88 2.9 2.92 2.94ln(M)
BF = 0.3β = 1
2.75 2.8 2.85 2.9 2.95ln(M)
BF = 5.6β = 5
2.65 2.7 2.75 2.8 2.85 2.9 2.95ln(M)
BF = 322β = 10
2.4 2.5 2.6 2.7 2.8 2.9ln(M)
BF = 3300β = 20
FIG. 16: Histograms showing the recovered log total massfor GR and ppE searches on ppE signals. As the source getsfurther from GR, the value for total mass recovered by theGR search moves away from the actual value.
Perhaps the most interesting point to be made withthis study is that the GR templates return values of thetotal mass that are completely outside the error rangeof the (correct) parameters returned by the ppE search,even before the signal is clearly discernible from GR. Werefer to this parameter biasing as ‘stealth bias’, as it isnot an effect that would be easy to detect, even if onewere looking for it.
This ‘stealth bias’ is also apparent when the ppE αparameter is non-zero. As one would expect, when a GRtemplate is used to search on a ppE signal that has non-zero amplitude corrections, the parameter that is mostaffected is the luminosity distance. We again see the biasof the recovered parameter becoming more apparent asthe signal differs more from GR. In this study, becausewe held the injected luminosity distance constant insteadof the injected SNR, the uncertainty in the recovered lu-minosity distance changes considerably between the dif-ferent systems. In both cases shown, however, the re-covered posterior distribution from the search using GRtemplates has zero weight at the correct value of lumi-nosity distance, even though the Bayes factor does notfavor the ppE model over GR.
V. CONCLUSION
The two main results of this study are that the ppEwaveforms can constrain higher order deviations from GR(terms involving higher powers of the orbital velocity)much more tightly than pulsar observations, and thatthe parameters recovered from using GR templates torecover the signals from an alternative theory of gravitycan be significantly biased, even in cases where it is notobvious that GR is not the correct theory of gravity. Wealso see that the detection efficiency of GR templates canbe seriously compromised if they are used to characterizedata that is not described by GR.
13
30 40 50 60 70 80 90 100 110DL (Gpc)
BF = 1 ppEGR
injected value
30 40 50 60 70 80 90 100 110DL(Gpc)
BF = 53 ppEGR
injected value
FIG. 17: Histograms showing the recovered values for lumi-nosity distance from GR and ppE searches on a LISA binaryat redshift z = 7. Both signals have a = 0.5, and were in-jected with a luminosity distance of 70.5 Gpc. The top plothas α = 3.0 and the bottom has α = 2.5. As the Bayes factorfavors the ppE model more strongly, the bias in the recov-ered luminosity distance from the GR search becomes morepronounced.
The current study makes several simplifying assump-tions about the waveforms: we consider only the inspi-ral stage for non-spinning black holes on circular orbits,and include just the leading order ppE corrections to thewaveforms. In future work we plan to include a marginal-ization over these higher order corrections. Including thismarginalization will be more realistic, as the ppE for-malism allows for many higher order corrections to thewaveform. Marginalizing over the higher order terms willweaken the bounds on the leading order ppE parame-ters, though probably not by that much since they aresub-dominant terms.
Another subject that we will examine in the futureis the affect on our analysis of multiple detections. Si-multaneously characterizing several systems with differ-ent mass ratios should allow us to constrain all six ppEparameters and not just the four we used in this study.Looking at several systems simultaneously should also al-low us to detect deviations from GR that are smaller than
those we could confidently infer with a single detection,as the evidence for the ppE hypothesis will accumulatewith the additional data.
We also plan to perform a study similar to that doneby Arun et al. [24–26], in which the exponents ai, bi arefixed at the values found in the PN expansion of GR, andcompare their Fisher matrix based bounds to those fromBayesian inference.
Finally, we will look at LISA observations of galacticwhite-dwarf binaries to see if the brighter systems, whichmay have SNRs in the hundreds, may allow us to beatthe pulsar bounds across the entire ppE parameter space.The brightest white-dwarf systems will have u ∼ 10−8 →10−7 (for comparison the ‘golden’ double pulsar system,PSR J0737-3039A has u = 3.94× 10−9), and these smallvalues for u make the ppE effects, which scale as ua andub, much larger than for black hole inspirals when a, b <0.
The chance to test the validity of Einstein’s theoryof gravity is one of the most exciting opportunities thatgravitational wave astronomy will afford to the scientificcommunity. Without the appropriate tools, however, ourability to perform these tests is sharply curtailed. Thisanalysis has shown that the ppE template family couldbe an effective means of detecting and characterizing de-viations from GR, and also that assuming that our GRwaveforms are correct could lead to lessened detectionefficiency and biased parameter estimates if gravity isdescribed by and alternative theory. We have identifiedseveral areas of future investigation, and will continue tostudy this area in depth.
Acknowledgments
We thank Patrick Brady, Curt Cutler, Ben Owen,David Spergel, Xavier Siemens, Paul Steinhardt andMichelle Vallisneri for detailed comments and sugges-tions. We are very grateful to Martin Weinberg and WillFarr for making their direct evidence integration codesavailable to us, and for helping us to understand the re-sults. N. J. and L. S. acknowledge support from theNSF Award 0855407 and NASA grant NNX10AH15G.N. Y. and F. P. acknowledge support from the NSF grantPHY-0745779, and FP acknowledges the support of theAlfred P. Sloan Foundation.
[1] C. M. Will, Living Reviews in Relativity 9 (2006), URLhttp://www.livingreviews.org/lrr-2006-3.
[2] N. Yunes and F. Pretorius, Phys. Rev. D80, 122003(2009), 0909.3328.
[3] B. F. Schutz, J. Centrella, C. Cutler, and S. A. Hughes(2009), 0903.0100.
[4] C. M. Will, Phys. Rev. D57, 2061 (1998), gr-qc/9709011.[5] C. M. Will and N. Yunes, Class. Quant. Grav. 21, 4367
(2004), gr-qc/0403100.[6] E. Berti, A. Buonanno, and C. M. Will, Class. Quant.
Grav. 22, S943 (2005), gr-qc/0504017.[7] A. Stavridis and C. M. Will, Phys. Rev. D80, 044002
(2009), 0906.3602.[8] K. G. Arun and C. M. Will, Class. Quant. Grav. 26,
155002 (2009), 0904.1190.[9] D. Keppel and P. Ajith, Phys. Rev. D82, 122001 (2010),
Cornish, Sampson, Yunes & Pretorius, 2011
Fundamental Bias
YunesGW Probes of Extreme Gravity 26
Sampson, 2013
Ignoring Fundamental Bias...injection=(not-ruled out) ppE template=GR
Fitting Factor Deteriorates
Physical Parameters Completely Biased
YunesGW Probes of Extreme Gravity 27Vallisneri & Yunes, 2013
Stealth Bias
Fundamental Bias that we can’t detect!SNR needed to detect a
GR deviationSNR needed for fundamental bias error
to be larger than systematic error
Overt BiasNegligible Bias
Stealth Bias
YunesGW Probes of Extreme Gravity 28
Sampson, Cornish & Yunes, 2013
Bayes Factor between a 1-parameter ppE template and a GR template (red) and between a 2-parameter ppE template and a GR template (blue), given a non-GR
injection with 3 phase deformations, as a function of the magnitude of the leading-order phase deformation.
Simple ppE Performance
GW Probes of Extreme Gravity Yunes
The Need for Accuracy
• “C-tensor”:
29Quantum Noise (Amelino-Camelia)
GW Probes of Extreme Gravity Yunes
Gravitational Wave Detectors
• is the Pontryagin density.
• is either a dynamical field that evolves.
30
Bounce light off mirrors and look for interference pattern when the light
recombines.
LHO
LLO Virgo/AdV
GEO
KAGRA
Ligo-India
Detectors
YunesGW Probes of Extreme Gravity 31
Yunes & Pretorius, PRD 2009 Mirshekari, Yunes & Will, PRD 2012 Chatziioannou, Yunes & Cornish, PRD 2012
I. Parametrically deform the Hamiltonian.
II. Parametrically deform the RR force.
III. Deform waveform generation.
IV. Parametrically deform g propagation.
h = hGR (1 + �fa) ei�fb
Result: To leading PN order and leading GR deformation
Parameterized post-Einsteinian Framework
A = AGR + �A�AH,RR = ↵H,RRv
aH,RR
h = F+h+ + F⇥h⇥ + Fshs + . . .
E2g = p2gc
4 + ↵p↵g
h(f) = hGR(f) (1 + ↵fa) ei�fb