I Introduction
The detection of gravitational waves (GWs) from binary black hole (BBH) mergers and the first binary neutron star (BNS) inspiral Abbott et al. (2016a, b, 2017a, 2017b); The LIGO Scientific Collaboration et al. (2017a); Abbott et al. (2017c), by the advanced Laser Interferometer Gravitationalwave Observatory (aLIGO) Abbott et al. (2016c); The LIGO Scientific Collaboration et al. (2015) and the European advanced Virgo (aVirgo) detector Acernese et al. (2015), has ushered in a revolution in astrophysics.
These groundbreaking discoveries have provided conclusive evidence that stellar mass BBHs form and coalesce within the age of the Universe Abbott et al. (2016d), and that their astrophysical properties Abbott et al. (2016e, f) are consistent with Einstein’s theory of general relativity Einstein (1915). Furthermore, the detection of two colliding NSs with GWs and broadband electromagnetic observations has confirmed that BNS mergers are the central engines of short gamma ray bursts (sGRBs) Abbott et al. (2017c, d), and the cosmic factories where about half of all elements heavier than iron are produced The LIGO Scientific Collaboration et al. (2017b).
Ongoing improvements in the sensitivity of the aLIGO and aVirgo detectors at lower frequencies will enable detailed studies on the astrophysical content of GW signals. Since eccentricity modifies the amplitude and frequency evolution of GWs at lower frequencies, before it is radiated away due to GW emission (Peters, 1964; Peters and Mathews, 1963; Huerta et al., 2017a; Hinder et al., 2017), GW observations within the next few years with the aLIGO and aVirgo detectors will provide unique opportunities to search for and detect eccentric binary mergers.
Eccentricity is one of the cleanest signatures for the existence of compact binary systems formed in dense stellar environments (Clausen et al., 2013; Samsing et al., 2014; Samsing and RamirezRuiz, 2017; Samsing, 2017; Rodriguez et al., 2017). Therefore, identifying and carefully measuring the imprints of eccentricity in GW signals will enable new and detailed studies of astrophysical processes taking place in corecollapsed globular clusters and galactic nuclei, which would otherwise remain inaccessible (Samsing et al., 2014; Gondán et al., 2017; Hoang et al., 2017a; VanLandingham et al., 2016; Wen, 2003; Osburn et al., 2016; Maccarone et al., 2007; Strader et al., 2012; Chomiuk et al., 2013; Antonini et al., 2016; Rodriguez et al., 2015, 2016).
Current flagship matchedfiltering GW searches are highly optimized for the detection of quasicircular, spinaligned compact binary sources Usman et al. (2016), and burstlike GW signals Abbott et al. (2016g). GW sources that do not fall into these categories may be missed by these algorithms, as shown in Huerta et al. (2017a); Tiwari et al. (2016); Huerta et al. (2014a); Huerta and Brown (2013). Given the proven detection capabilities of the aLIGO and aVirgo detectors, it is timely and relevant to develop tools to confirm or rule out the existence of these GW sources Tiwari et al. (2016). This article focusses on the development of an inspiralmergerringdown (IMR) waveform model that is adequate for the detection and characterization of compact binary populations that form in dense stellar environments, and which are expected to enter the aLIGO frequency band with moderate eccentricities Antonini et al. (2016); Samsing et al. (2014); Antonini et al. (2014).
The model we introduce in this article, the ENIGMA (Eccentric, Nonspinning, InspiralGaussianprocess Merger Approximant) waveform model, builds on recent work to accurately describe quasicircular and moderately eccentric IMR BBH mergers Huerta et al. (2017a). ENIGMA does not require the calibration of free parameters for its construction; instead, it combines in a novel way analytical and numerical relativity (NR) using machine learning algorithms Moore and Gair (2014); Moore et al. (2016), which are a special class of algorithms that can learn from examples to solve new problems without being explicitly reprogrammed. In different words, the same algorithm can be used across science domains by just changing the dataset used to train it Goodfellow et al. (2016). Specifically, we use Gaussian Process Regression (GPR) Mackay (2003)
to interpolate numerical relativity simulations of the merger and ringdown across the waveform parameter space. GPR is well suited to this due to its flexibility, the fact that it makes minimal assumptions about the underlying data, and the ease with which it can be extended to higher dimensional interpolation problems. This last point will be particularly important when we come to extend the model to include BH spins in future work.
The main motivation to develop ENIGMA is to systematically quantify the importance of orbital eccentricity in the detection of GW sources with aLIGO. As we show in this article, ENIGMA can accurately reproduce the dynamics of quasicircular binaries and the true features of eccentric NR simulations. These features are of paramount importance to clearly associate deviations from quasicircularity to the physics of eccentric compact binary coalescence, and not to intrinsic waveform model inaccuracies. We use ENIGMA
to estimate the minimum value of eccentricity which may be discernable with aLIGO observations of eccentric BBH mergers.
In this article we use units . The binary components are labelled as and , where . We use the following combinations of : total mass , reduced mass , massratio , and symmetric massratio . This article is organized as follows: Section II contextualizes this work in light of recent efforts to model eccentric compact binary systems. Section III provides a brief description of our waveform model. In Sections IV, V we validate ENIGMA with a stateoftheart, IMR quasicircular waveform model, and a set of eccentric NR simulations, respectively. We discuss the detectability of moderately eccentric BBH mergers with aLIGO in Section VI. In Section VII we discuss the importance of including higherorder waveform multipoles for the detection of eccentric BBH mergers. We summarize our findings and outline future directions of work in Section VIII.
Ii Previous work
In this section we briefly summarize recent developments in the literature in connection to IMR waveform models that describe moderately eccentric compact binary mergers.
An IMR model describing highly eccentric compact binary mergers was introduced in East et al. (2013). This model used a geodesic based description for the inspiral dynamics, and a quasicircular merger waveform using the phenomenological approach described in Kelly et al. (2011). The model was used to study the detectability of burstlike signals in LIGO data.
In Ref. Cao and Han (2017), the authors introduce an effectiveonebody (EOB) model that combines the quasicircular dynamics of SEOBNRv1 Taracchini et al. (2012), and eccentric postNewtonian (PN) corrections up to 2PN order. They compare the model to three numerical relativity simulations: two of massratio and eccentricities at orbital frequencies , respectively; and one with and at an orbital frequency .
Ref. Hinderer and Babak (2017) presents a formalism to consistently incorporate eccentricity corrections to the EOB formalism. The radiative dynamics is restricted to 1.5PN order. No comparison to eccentric NR simulations is done.
In Ref. Hinder et al. (2017), the authors present an eccentric model that includes PN corrections up to 2PN in the fluxes of energy and angular momentum. They show that their model can reproduce eccentric numerical relativity simulations with massratios , and eccentricities , measured seven cycles before merger.
In this work, we show that ENIGMA reproduces the most recent version of EOB models, SEOBNRv4 Bohé et al. (2017), in the quasicircular limit. Our model is not based on phenomenological approximations for the description of the inspiral and merger evolution. Rather, we develop a hybrid inspiral model that encodes eccentric PN corrections up to 3PN order that includes tail, tailsoftails, and corrections due to nonlinear memory that enter at 2.5PN and 3PN order, as described in Huerta et al. (2017a). Furthermore, we improve the inspiral evolution by including up to 6PN quasicircular corrections using selfforce and black hole perturbation theory (BHPT) results. We model the merger phase using a machine learning algorithm that is trained with a dataset of NR simulations. This approach ensures that our merger model has the same fidelity of NR simulations. We validate our model using a set of 12 Einstein Toolkit Nakamura et al. (1987); Shibata and Nakamura (1995); Baumgarte and Shapiro (1998); Baker et al. (2006); Campanelli et al. (2006); Pollney et al. (2011); Wardell et al. (2016); Löffler et al. (2012); Ansorg et al. (2004); Diener et al. (2007); Schnetter et al. (2004); Thornburg (2004) eccentric NR simulations, with massratios up to and eccentricities ten orbits before merger.
Inspiral only models have steadily increased their accuracy Levin et al. (2011); Yunes et al. (2009); Huerta et al. (2014a); Huerta and Brown (2013); Hinder et al. (2010); Loutrel and Yunes (2017a); Loutrel et al. (2014); Osburn et al. (2016); Loutrel and Yunes (2017b). However, as extensively discussed in Huerta et al. (2014a, 2017a) and in this article, PNbased models, even including up to 3PN eccentric corrections, are not accurate enough to describe the dynamical evolution of eccentric compact binaries through merger, since the PN prescription breaks down at that point.
A key result in this article is the validation of ENIGMA with eccentric NR simulations. Recent NR studies on the physics of eccentric compact binary mergers include Hinder et al. (2008, 2010); East et al. (2012); East and Pretorius (2012); Gold et al. (2012); Gold and Brügmann (2013); Paschalidis et al. (2015); East et al. (2016); Radice et al. (2016); The LIGO Scientific Collaboration et al. (2016); Lewis et al. (2017); Huerta et al. (2017b). In this article we show that ENIGMA reproduces the true features of eccentric NR simulations throughout late inspiral, merger and ringdown, without requiring the use of eccentric NR simulations to calibrate it. In different words, a consistent combination of analytical relativity formalisms, boosted with a machinelearning based merger waveform provides a powerful framework to describe both quasicircular and eccentric BBH mergers.
Iii Waveform model construction
Our eccentric waveform model has two main components. The first component is an inspiral evolution scheme that combines results from PN theory Arun et al. (2009), the selfforce formalism Chua et al. (2017); Barausse et al. (2012); Bini and Damour (2014); Huerta and Gair (2011a, b); Huerta et al. (2014b); Huerta and Gair (2011c); Heffernan et al. (2017) and BHPT Fujita (2012); Huerta et al. (2012); Huerta and Gair (2010); Chua and Gair (2015); Huerta and Gair (2009). The second component is a merger waveform, which is constructed by interpolating a set of NRbased surrogate waveforms Blackman et al. (2015) using GPR Moore and Gair (2014); Moore et al. (2016); Mackay (2003); Doctor et al. (2017). The training dataset of NRbased surrogate waveforms describe BBHs with massratios .
In the following sections we present a succinct description of the improved inspiral evolution scheme, and a detailed description of the standalone GPRbased waveform model. Thereafter, we describe how to combine the inspiral and merger models to render a unified description of the dynamical evolution of moderately eccentric compact binary systems.
ENIGMA is tailored to carry out searches of compact binary systems that enter the aLIGO frequency band with moderate values of eccentricity. As discussed in Huerta et al. (2017a), this approximation covers an astrophysically motivated population of compact binary sources that are expected to enter the aLIGO frequency band with eccentricities at 10Hz, and circularize just prior to merger Antonini et al. (2016); Samsing et al. (2014); Antonini et al. (2014). Furthermore, using a set of eccentric NR simulations, in Section V we show that ENIGMA accurately describes BBH mergers that retain significant residual eccentricity prior to merger. This feature may prove useful anticipating astrophysically unconstrained formation mechanisms for eccentric BBH mergers.
iii.1 Inspiral evolution
We model the inspiral evolution within the adiabatic approximation, i.e., we assume that the radiation time scale is much longer than the orbital time scale, and therefore we use an averaged description of the radiation reaction over an orbital period Blanchet (2006). Furthermore, in order to combine a variety of recent results from analytical relativity in a consistent way, we express the equations of motion in a gaugeinvariant manner using as an expansion parameter the gaugeinvariant quantity , where represents the mean orbital frequency.
A first attempt to accurately describe the inspiral evolution of eccentric compact binaries with massratios was introduced in Ref. Huerta et al. (2017a). Below we succinctly describe the main ingredients of that model. Thereafter, we describe the new physics that we incorporate into ENIGMA.
The description of compact binary dynamics involves the relative orbital separation of the binary , which is related to the eccentric anomaly by
(1) 
On the other hand, the mean anomaly , which is related to the mean motion through the relation , is customarily described in terms of the eccentric anomaly as follows Blanchet (2006); Arun et al. (2009):
(2) 
The conservative dynamics, which are obtained from a 3PN order Hamiltonian for eccentric compact binary systems Blanchet (2006); Arun et al. (2009), determines the time evolution of the instantaneous angular velocity and the mean anomaly :
(3)  
(4) 
The instantaneous angular velocity is related to the mean orbital frequency through , where indicates average over an orbital period. represent the periastron precession and the relativistic precession, respectively, and are related via . The radiative evolution of the binary, which is driven by the energy and angular momentum carried out by GWs from the binary system, can be specified through the time evolution of eccentricity and the gaugeinvariant expansion parameter
(5)  
(6) 
The hereditary terms (HT) in Eqs. (5)(6) include tail, tailsoftails and nonlinear memory corrections at 1.5PN, 2.5PN and 3PN order. The 2.5PN and 3PN pieces of Eq. (5) were derived in Ref. Huerta et al. (2017a). As discussed above, we used the adiabatic approximation to model eccentric BBH mergers, therefore the functions only depend on , and Eqs. (5)(6) describe a closed system that can be solved given initial conditions for and . Once this is done, are used to numerically integrate Eq. (4). Thereafter, are substituted in Eq. (2) to determine by rootfinding. Finally, all these pieces are used to determine Eqs. (1) and (3).
Having this workflow in mind, we realize that to improve phase accuracy of any PNbased model, it is necessary to include higherorder PN corrections to Eqs. (5) and (6). This becomes apparent when we write the above expressions in the quasicircular limit. In the expressions below, we have augmented the conservative and radiative dynamics with selfforce and BHPT results, i.e.,
(7)  
(8)  
where is Euler’s constant, and the coefficients are presented in Appendix C of Ref. Huerta et al. (2017a). As is demonstrated in Section III.3, when this hybrid inspiral formalism is combined with the GPR mergerringdown waveform presented in Section III.2, the GW emission from quasicircular binaries with massratios in the range is described with excellent accuracy. As discussed in contemporary literature, this is a basic requirement for any waveform model that aims to establish a clear cut connection between eccentricity and deviations from quasicircular motion Huerta et al. (2017a, 2014a); Huerta and Brown (2013). In other words, accurate eccentric BBH modeling can only be done with models that simultaneously incorporate an accurate description of quasicircular motion.
In order to obtain GW strain from inspiral trajectory evolution, we use standard leadingorder, eccentric PN strain expressions for both polarizations as a baseline, —where represent the Inspiral waveform— and then include quasicircular PN corrections up to 3PN order , which are given by Eqs. (320)(323g) in Ref. Blanchet (2006):
(9) 
with
(10)  
(11)  
where , and represent the polar angles of the observer, and is the distance to the binary. This completes the description of the inspiral portion of our ENIGMA model. As mentioned above, ENIGMA has been developed to target eccentric compact binary systems that circularize prior to merger. This is a reasonable approach, since recent studies suggest that the eccentricity distribution of aLIGO sources may be bimodal, and the population with moderate values of eccentricity may enter the aLIGO frequency band with eccentricities at 10Hz. We show that our model is equipped to target this population, as well as compact binaries that may enter the aLIGO frequency band with eccentricities up to at 10Hz. Under this assumption, the following section describes the construction of a quasicircular merger waveform using GPR. The main motivation for the use of GPR is that they may be easily extended to higher dimensional interpolation problems which will be faced in future work simulating eccentric mergers of spinning black holes.
iii.2 Merger and Ringdown
The main limitation of the predecessor to the ENIGMA model Huerta et al. (2017a) was the treatment of the mergerringdown signal. In this section, a new machine learning based approach to modeling the merger–ringdown signal, using NR simulations is described. A desirable feature of this approach is the ability to identify regions of parameter space where the model performs poorly, and rapidly incorporate additional NR simulations into an updated model. This iterative procedure of improving the model is a feature of this approach that will be illustrated throughout this section.
In this paper only nonspinning BBHs are considered (although the approach is designed to be flexible enough to allow for the extension to spinning and precessing BBH systems in future); therefore the intrinsic parameter space of the IMR model is 3–dimensional (mass ratio, , eccentricity, , and mean anomaly, ; the total mass, , simply sets the overall dimensional scale of the problem). It is known that the mean anomaly affects the amplitude and phase of the waveform strain Brown and Zimmerman (2010); Huerta and Brown (2013). For a given value of anomaly, these effects are handled by the method we use to smoothly connect the inspiral and merger waveforms, as described in the following section.
As is well known (see, e.g. Peters (1964); Peters and Mathews (1963); Huerta et al. (2017a); Clausen et al. (2013); Hinder et al. (2017)) BBHs with moderate eccentricities early in the inspiral are efficiently circularized by GW emission and are almost circular at merger. This reduces the effective dimensionality of the merger–ringdown model to just 1 dimension (mass ratio).
An initial training set, , of 19 quasi–circular NR simulations from the public SXS catalog Mroué et al. (2013) at mass ratios was used to build a mergerringdown model. Only the modes of the merger waveforms were used. The effect of neglecting higherorder waveform modes is quantified in Section VII. The time series were decomposed into the amplitude, , and (unwrapped) phase, . The waveforms were rescaled (in time) such that the total mass was unity, the peak luminosity occurs at , the amplitude was rescaled to satisfy and the phase was shifted to satisfy . The time series were sampled at points in the interval (with a higher sampling rate around merger). The training sets for amplitude and phase are
(12) 
GPR is an interpolation (or extrapolation) technique which makes minimal assumptions about the underlying function. It is used here to interpolate the data in Eq. 12 to obtain the waveform at any mass ratio, . A Gaussian process (GP) of a single variable is completely described by a covariance function, (and a mean function which is here assumed to be zero for simplicity). For the first training set, , the parameterization was used. The covariance function, and any free parameters therein, are free to be specified; however, they can also be learnt
from the training set by maximizing the probability of the training set being realized by the GP (maximizing the
GP evidence). This learning process can be computationally expensive, especially for large training sets or when comparing covariance functions with many free parameters; the techniques described in Moore et al. (2016) were used to accelerate this learning phase. The covariance functions considered here were the squaredexponential and Wendland polynomial functions used previously for waveform modeling in Moore et al. (2016); these covariance functions are all stationary, i.e. . For a discussion of GPR see Rasmussen and Williams (2005), or Moore and Gair (2014); Moore et al. (2016) in the context of GW signal modeling.As zero–mean GPs were used for the interpolation, the phase interpolation can be improved if a reference phase function is first subtracted from the training set phases to bring the values closer to zero; . After interpolation the phase can be recovered straightforwardly by adding to the result. The reference phase function was chosen to be the phase of the equal mass NR simulation; .
It was found that the GPR interpolation using the 19 waveforms in the set was not sufficiently accurate; typical errors on the interpolant were on the phase and on the amplitude (see Fig. 1, with ). A second training set, was created by adding more waveforms approximately uniformly in mass ratio; the new set, , consisted of simulations at mass ratios (a total of 55 simulations). As sufficiently accurate, and long duration NR simulations at these mass ratios were not readily available, NR surrogate waveforms Blackman et al. (2015) were used instead. Work is ongoing to replace the surrogate waveforms used in this study with full NR simulations. On this training set GPR was tested using different parameterizations^{1}^{1}1The parameterizations tried included the mass ratio, , a simple compactification on to the interval , , the symmetric mass ratio , which compactifies on to the interval ), and . of the training set, ; it was empirically found that yielded the largest GP evidence value and the best interpolant. From Fig. 1 it can be seen that indeed outperforms across most of the parameter space.
Finally, a third training set was constructed using 70 waveforms placed uniformly in the logarithm of the mass ratio; . The use of a stationary covariance function (with ) leads to large errors near the training set boundary (or ). Of course, this boundary is artificial because the parameter space admits the following identifications (or ). It was found that using the following nonstationary covariance function incorporating this symmetry into the GP leads to improved performance for nearly equal mass systems;
(13) 
The errors on the GPR interpolants for the amplitude and phase for the three training sets are shown in Fig. 1. As expected, the interpolant is generally the most accurate, especially for nearly equal mass systems (the interpolant is used for the remainder of the paper). A thorough validation of the combined inspiral–merger–ringdown model is presented in Sec. III.3. However, the finalized GPR interpolant for the mergerringdown performs extremely well; it reproduces the NR surrogate waveforms in the mass ratio range with overlaps computed over the entire duration .
iii.3 Complete waveform model
In this Section we describe the method followed to smoothly attach the hybrid inspiral model of Section III.1 with the GPRbased merger waveform of Section III.2.
The studies we have carried out to do this work indicate the regime of validity of our hybrid inspiral scheme, and furnish strong evidence that analytical and numerical relativity can be blended together to create a model that can accurately reproduce the true features of both quasicircular and eccentric compact binaries. We now describe the construction of a map that determines the optimal frequency at which the inspiral evolution can be blended with the merger evolution for a given combination of masses . In order to do this, a number of signal processing tools are needed which will now be described.
Given two signals and , and defining as aLIGO’s design power spectral density (PSD) Shoemaker (2015), and
as the Fourier transform of
, the noiseweighted inner product between and is given by(14) 
with and . The waveforms used in this study are generated with a sample rate of 8192Hz. Additionally, the normalized overlap is defined as
(15)  
(16) 
where indicate that the normalized waveform has been time and phaseshifted. Using these definitions, we blend our inspiral and merger models as follows:

Combine the inspiral and merger codes into a single library that generates an inspiral waveform and smoothly attaches a GPR merger waveform on the fly

At the point of attachment, , the inspiral, , and merger, , waveforms satisfy continuity and differentiability
(17) (18) 
The mass parameter space we consider to construct this map is , in steps of in the dimensions.

For each of the points of the aforementioned parameter space, we generated quasicircular ENIGMA waveforms considering a wide range of frequencies to connect the inspiral and merger waveforms, namely: in steps of

Thereafter, we computed overlaps between the set of waveforms described in the previous item and their SEOBNRv4 counterparts.

Finally, we picked the attachment frequency value that maximized the overlap for each point of the BBH parameter space under consideration.

Our ENIGMA code automatically translates into an optimal time of attachment , i.e., it determines the time in the GPR NRbased merger waveform that corresponds to
Finally, the complete IMR ENIGMA waveform can be written as
(19) 
where is the Heaviside step function, and represent the Inspiral and Merger waveforms. are time and phase shifts that need to be incorporated in the GPR merger waveforms to enforce continuity and differentiability. In the next section we explore the accuracy and robustness of this scheme in the quasicircular limit.
Iv Validation of Enigma in the quasicircular limit
The algorithm we described in the previous section to connect the inspiral and merger evolution utilizes a discrete grid of 1100 points, which uniformly covers the BBH parameter space . To ensure that this method is robust, we now compute the overlap between our ENIGMA model in the quasicircular limit, and SEOBNRv4 waveforms using a grid that covers the same region of parameter space, but now using 2500 points. For each combination, we test possible values of attachment frequency.
To determine at which frequency we connect the inspiral and merger waveforms, we consider the following: if we were to blend in the inspiral and merger waveforms using the lowest frequency of attachment in our dataset, BBH systems with component masses similar to the first and third GW transients detected by aLIGO would be described entirely by our quasicircular GPR merger waveforms. In that case, since we would only be using a quasicircular waveform, we would not be able to make any meaningful statements regarding the effects of eccentricity in the detection of BBH mergers with aLIGO. However, since we want to study eccentric binary mergers, we require a model that can tolerate small eccentricity values very late in the inspiral evolution. We can only do so if we attach the merger waveform as late as possible, i.e., using the largest possible frequency of attachment.
Based on these considerations, we have constructed a map that smoothly connects the inspiral and merger evolution, using the largest frequency of attachment, and which guarantees that the overlap between quasicircular ENIGMA waveforms and SEOBNRv4 waveforms is . Figure 2 presents these results. This is the level to which contemporary quasicircular models agree in many regions of BBH parameter space Bohé et al. (2017); Kumar et al. (2016). In the following section, we will show that this map works very well when we consider binary systems with nonnegligible eccentricity.
V Validation of Enigma with eccentric numerical relativity simulations
To show that ENIGMA reproduces the dynamics of eccentric BBHs throughout late inspiral, merger and ringdown, we use a catalog of eccentric NR simulations, generated with the open source, community software the Einstein Toolkit, and postprocessed with the open source software POWER Johnson et al. (2018). The parameters that describe these BBH NR simulations are provided in Table 1. To determine these parameters, we select the parameter combination that maximizes the overlap between ENIGMA and NR waveforms, without using any information based on the trajectory of the BHs. Appendix A provides a brief summary of the convergence and phase error of these NR simulations, which were generated with three different grid resolutions to assess their convergence. A detailed description of this NR catalog is provided in an accompanying paper Huerta et al. (2017b). To validate ENIGMA, we use the highest resolution run of each dataset.
Simulation  

E0001  1.0  0.060  3.50  0.077  0.998 
J0005  1.5  0.067  3.30  0.078  0.997 
J0045  2.0  0.078  3.35  0.079  0.994 
E0013  2.5  0.070  3.00  0.081  0.997 
E0017  3.0  0.068  2.60  0.083  0.989 
K0001  3.5  0.060  3.20  0.081  0.991 
J0061  4.0  0.065  2.90  0.086  0.992 
J0065  4.5  0.080  3.10  0.088  0.981 
M0004  1.0  0.190  3.20  0.071  0.993 
J0047  2.0  0.120  2.70  0.078  0.956 
K0024  4.0  0.200  2.90  0.084  0.971 
L0020  5.5  0.210  3.10  0.087  0.951 
Figures 3 and 4 indicate that ENIGMA reproduces with excellent accuracy the latetime radiative evolution of NR simulations that describe BBH mergers with massratios and eccentricities ten orbits before merger. The last column of Table 1 indicates that the overlap between ENIGMA waveforms and their NR counterparts is .
We have also explored the performance of our new model to describe the evolution of BBH systems that retain significant eccentricity right before merger. Figure 5 and the last column of Table 1 show that ENIGMA can reproduce fairly well the dynamics of BBH systems with massratios and eccentricities ten orbits before merger; with overlaps . It is noteworthy that, even though ENIGMA was constructed to faithfully reproduce the dynamics of moderately eccentric BBH mergers, it can also describe BBH mergers that circularize right before merger. It is worth emphasizing that our ENIGMA model can do this because our inspiral scheme can accurately reproduce the true dynamics of both quasicircular and eccentric binaries very late in the inspiral evolution, which is clearly indicated in Figures 3, 4 and 5. These results imply that ENIGMA can be used both for matchedfiltering, burst and Deep Filtering searches Tiwari et al. (2016); George et al. (2017a); Klimenko et al. (2008); George and Huerta (2017a); George et al. (2017b); Shen et al. (2017); George and Huerta (2017b, c).
These results establish the accuracy of ENIGMA to describe eccentric BBH mergers, and indicate that the map we have developed to smoothly connect our hybrid inspiral scheme to a quasicircular, NRbased merger waveform is robust to describe moderately eccentric BBH mergers.
In terms of the astrophysically motivated systems that can be described with ENIGMA, Figure 6 presents two scenarios. Each panel includes an eccentric ENIGMA waveform and its quasicircular counterpart, generated with the surrogate waveform family Blackman et al. (2015). The insets show the properties of the BBH systems and the overlap between the two waveforms. The left panel presents a moderately eccentric BBH with component masses and at . The right panel depicts an eccentric BBH with component masses and at . We have chosen these examples to represent a variety of astrophysically motivates scenarios, i.e., BBHs that enter the aLIGO frequency band on nearly quasicircular orbits, and moderately eccentric BBH systems that may form in dense stellar environments through a variety of dynamical mechanisms Samsing et al. (2014); Samsing and RamirezRuiz (2017); Hoang et al. (2017b). Since the proposed formation scenarios for eccentric BBH mergers are astrophysically unconstrained, it is advantageous to use a waveform model that can cover as deep a parameter space as possible, and then let GW observations confirm or rule out potential formation channels. The analysis we carried out in this section indicates that ENIGMA can be used for these studies.
Figure 6 provides a clear visual description of the imprints of eccentricity. Compared to quasicircular signals, eccentricity reduces the timespan of GW signals, and induces significant modulations in frequency and amplitude at lower frequencies. The overlap values quoted in the inset have been computed using the design sensitivity of aLIGO from the quoted in the panels.
Having established the accuracy of our new waveform model, in the following section we use it to quantify the detectability of eccentric BBH mergers with GW observations.
Vi Detectability of eccentric binary black hole mergers
Having established the accuracy of ENIGMA in the quasicircular limit, and its ability to describe the dynamics of eccentric systems throughout merger with a set of eccentric NR simulations, we now want to quantify the minimum value of eccentricity at which a circular search is no longer effectual.
One can use a variety of criteria to accomplish this. In this study, we use fitting factor () calculations to establish a connection between eccentricity and deviations from quasicircularity. The is defined as Apostolatos (1995)
(20) 
The represents the overlap between a given GW signal , maximized continuously over chosen intrinsic and extrinsic parameters using templates of model . If the set contains all parameters that describe the template and model also models the signal , our continuous would be unity by construction. In practice, we perform this continuous maximization over
using particle swarm optimization
Kennedy and Eberhart (1995)^{2}^{2}2We use an opensource Python implementation of the particle swarm optimization algorithm: pyswarm Lee .. We find that using a swarm size of the converges to one part in .Figure 7 presents a gallery of calculations for a variety of eccentric BBH populations. In these panels we present BBH populations with component masses . The eccentricity range we explore is at .
A word of caution is in order to interpret these results. Using a set of eccentric NR simulations, we have established that ENIGMA can accurately describe eccentric BBH systems throughout late inspiral, merger and ringdown. We have also shown that, in the quasicircular limit, our model reproduces with excellent accuracy SEOBNRv4. Just as EOB models have been calibrated with quasicircular NR simulations for the late inspiral and merger evolution, and then applied to explore the dynamics of BBH systems at lower frequencies, we expect that our model will provide an adequate description of moderately eccentric BBH systems during the early inspiral dynamics. To accomplish this, we have introduced a hybrid inspiral scheme that combines recent developments in the modeling of eccentric and quasicircular BBHs, and which we have shown to perform very well to capture the dynamics of both eccentric and quasicircular BBH systems very late in the inspiral evolution.
0.010  0.994  0.999 
0.025  0.992  0.999 
0.050  0.989  0.998 
0.075  0.980  0.995 
0.100  0.963  0.991 
0.125  0.941  0.986 
0.150  0.912  0.980 
0.200  0.873  0.962 
0.250  0.829  0.937 
0.275  0.794  0.923 
0.325  0.736  0.886 
Table 2 presents key results extracted from the calculations presented in Figure 7. Just as we found in Huerta and Brown (2013)—left panel of Figure 2 therein—the anomaly does not have a significant impact in the results we present in this section.
If we assume that eccentricity corrections become significant when , then we can highlight the following results for the BBH parameter space :
A circular search will be effectual for eccentric BBH that satisfy at . We notice that eccentricity corrections become significant for , since at this particular threshold throughout the entire BBH parameter space. This eccentricity threshold is highlighted in Table 2. Putting this information in context with Figure 7, we notice that for this eccentricity value, some regions of the BBH parameter space have . These regions correspond to low mass and highly asymmetric massratio BBH systems. These results are consistent with previous results reported in the literature Huerta et al. (2017a), i.e., compact sources that generate longlived GW signals in the aLIGO frequency band will be the ones that present the most significant imprints of eccentricity. Shorter signals, represented by eccentric, massive BBH systems, will be better recovered by quasicircular templates. This is because eccentricity corrections have less time to accumulate, and thus have a lesser impact in the waveform phase. This pattern is clearly shown in the panels of Figure 7.
Table 2 indicates that for . As before, whereas massive BBH systems can be recovered with , low mass and asymmetric massratio BBH systems have . Finally, we notice that recovering a BBH population with would require the use of searches that specifically target eccentric binaries.
Figure 7 presents the five GW transients, consistent with BBH mergers, currently reported by aLIGO: GW150914, GW151226, GW170104, GW170814 and GW170608. These are indicated by a star, a diamond, a circle, a hexagon and a triangle, respectively. Table 3 indicates that these events can be recovered using spinning, quasicircular SEOBNRv4 templates with if at , respectively.
recovery with spinning SEOBNRv4 waveforms  
0.010  0.996  0.994  0.998  0.997  0.995 
0.025  0.996  0.993  0.997  0.996  0.994 
0.050  0.994  0.992  0.996  0.995  0.992 
0.075  0.992  0.984  0.993  0.991  0.986 
0.100  0.986  0.976  0.989  0.987  0.973 
0.125  0.982  0.984  0.982  
0.150  0.977  0.943  0.978  0.974  0.934 
0.175  0.928  0.920  
0.200  0.960  0.913  0.960  0.960  0.906 
To clearly emphasize the content of these results, in Figure 8 we present eccentric waveform signals that have with the quasicircular waveforms which the GW transients were assumed to be. The eccentric signals have at . The quasicircular SEOBNRv4 signals are generated from . The overlap values quoted in each panel are computed using the design sensitivity of aLIGO from . These results clearly indicate that the GWs we have detected, and assumed to be quasicircular, may have eccentricity content at lower frequencies that is not easily discerned by the time these signals become detectable by aLIGO and aVirgo. Indeed, all these signals can be recovered with .
Recovery with spinantialigned SEOBNRv4 waveforms  

1  1.78  0.26  0.043  0.083 
2  3.07  3.67  0.449  0.738 
3  0.41  0.96  0.288  0.454 
4  2.06  1.51  0.093  0.145 
5  4.75  5.14  0.051  0.401 
Furthermore, as shown in Table 4, the astrophysical parameters of both waveform signals, eccentric and quasicircular, are very similar, with differences in the component masses of a few percent. The key difference is that our ENIGMA model currently describes nonspinning, eccentric binaries. In different words, GWs emitted by quasicircular, spinantialigned BBH systems can easily mimic the features of moderately eccentric, nonspinning BBH mergers. We have also computed results using a template bank of SEOBNRv4 waveforms with the spin of the binary components set to zero. Our results are consistent using both spinaligned and nonspinning template banks, with deviations level, in a similar spirit to the results reported in Huerta et al. (2017a).
In conclusion, even though recently detected GW transients can be effectively recovered with quasicircular, spinning SEOBNRv4 templates, it is plausible that these events may have significant residual eccentricity at , and circularize by the time these systems become detectable by aLIGO, i.e., at . We expect that when aLIGO Abbott et al. (2016c); The LIGO Scientific Collaboration et al. (2015) and aVirgo Acernese et al. (2015) reach design sensitivity within the next few years, and we can start observing the evolution of compact binary mergers from , we will be able to quantify eccentricity corrections, and clearly associate them to currently unconstrained astrophysical formation channels of compact binary populations in dense stellar environments.
Vii Importance of higherorder waveform multipoles for the detection of eccentric binary black hole mergers
There are several studies in the literature that have explored the importance of including higherorder waveform multipoles for the detection of quasicircular binaries Brown et al. (2013); Capano et al. (2013); London et al. (2017). There is, however, no study in the literature that has shed light on this important topic in the context of eccentric binary mergers. In this section, we provide a succinct introduction to this problem using our catalog of eccentric NR simulations.
The NR higherorder waveform multipoles we use in this section exhibit similar convergent behavior to the we used in Section V to validate the ENIGMA model. The waveform multipoles we use in this section are those that contribute more significantly to the waveform strain, namely and .
Figure 9 presents two cases. The top row presents the higherorder waveform multipoles of simulation M0004 in our catalog, as described in Table 1, i.e., and twelve orbits before merger. To clearly show the low impact of higherorder waveform multipoles for equal mass, eccentric BBH mergers, we use two different panels: the top left panel only shows the subdominant multipoles, whereas the right panel shows the amplitudes of all multipoles in comparison to the leading mode, clearly indicating that higherorder waveform multipoles contribute up to of the signal power, with the greatest contribution near merger. In different words, these modes do not significantly contribute to the detectability of equal mass, eccentric BBH signals.
The bottom panels of Figure 9 show the higherorder waveform multipoles of the L0020 simulation: and ten orbits before merger. These results show that the subdominant multipoles contribute up to of the total signal power in the vicinity of merger. Therefore, the inclusion of higherorder waveform multipoles will be necessary for searches of eccentric BBH mergers that have asymmetric massratios. This finding is consistent with results in the context of quasicircular BBH mergers Brown et al. (2013); Capano et al. (2013); London et al. (2017), i.e., higherorder waveform multipoles contribute significantly to the total waveform strain of asymmetric massratio systems, having an important impact on their SNR, and therefore on their detectability. A detailed study of this important source modeling topic is well underway, and will be presented in an accompanying paper Huerta et al. (2017c).
Viii Conclusions
We have developed ENIGMA, a complete waveform model to search for and characterize compact binary populations that form in dense stellar environments, and which are expected to enter the aLIGO frequency band with moderate values of eccentricity.
Our model is a combination of analytical and numerical relativity results. It is constructed under the assumption that moderately eccentric compact binary systems circularize prior to merger. In this context, we describe the inspiral evolution using a variety of recent results from PN theory, the selfforce and BHPT. On the other hand, to describe the quasicircular merger evolution, we use GPR to create a standalone, merger waveform that is trained with a dataset of quasicircular NR waveforms. We described a method to put together these two pieces so as to provide a complete description of moderately eccentric compact binary systems. Our results demonstrate that ENIGMA describes with excellent accuracy the dynamics of both quasicircular and moderately eccentric systems in a single, unified framework.
We have validated ENIGMA with a set of eccentric NR simulations that describe BBH mergers with massratios and eccentricities ten orbits before merger. To the best of our knowledge, this is the only model in the literature that can reproduce the dynamics of eccentric compact binary mergers for this combination of highly eccentric, and very asymmetric massratio systems. We have also validated ENIGMA in the quasicircular limit using SEOBNRv4 waveforms, and have shown that both waveform families have overlaps , assuming an initial filtering frequency .
Having validated ENIGMA both in the quasicircular limit, and with eccentric NR simulations, we used it to quantify the threshold at which the effect of eccentricity is negligible, and existing circular searches are effectual for moderately eccentric BBH mergers. Our studies show that BBH populations with at will be recovered with spinning, quasicircular SEOBNRv4 templates with . At this level, a circular search will be effectual. On the other hand, BBH populations with at will be recovered with a maximum . However, within this same population, low mass, and asymmetric massratio BBH populations are recovered with . BBH populations with eccentricities at , will be recovered with . Finally, BBH populations with will require the use of dedicated eccentric searches, since these events would be poorly recovered with .
Our calculations indicate that GW150914, GW151226, GW170104, GW170814 and GW170608 can be recovered with spinning, quasicircular SEOBNRv4 templates with if the eccentricity of these events satisfies at , respectively. We have also shown that the first five GW transients detected by aLIGO and aVirgo could be misclassified as quasicircular systems, since spin corrections can easily mimic the properties of moderately eccentric signals when they become detectable by groundbased GW detectors. We argue that future improvements to the sensitivity of aLIGO and aVirgo will be critical to identify and quantify eccentricity content in GW signals at lower frequencies.
We have carried out a preliminary analysis of the importance of including higherorder waveform multipoles, and have shown that these will be important for searches of eccentric BBH mergers whose components have asymmetric massratios. This result is consistent with similar analysis carried out in the context of quasicircular BBH mergers.
Having completed the description of nonspinning, eccentric BBH mergers, ongoing work is focused on the development of an extended version of ENIGMA that will enable the characterization of spinning, eccentric BBH systems. This new model will be useful to explore whether the degeneracy between eccentricity and spin corrections we have found in these studies can be resolved.
Ix Acknowledgements
This research is part of the Blue Waters sustainedpetascale computing project, which is supported by the National Science Foundation (awards OCI0725070 and ACI1238993) and the State of Illinois. Blue Waters is a joint effort of the University of Illinois at UrbanaChampaign and its National Center for Supercomputing Applications. The eccentric numerical relativity simulations used in this article were generated with the open source, community software, the Einstein Toolkit on the Blue Waters petascale supercomputer and XSEDE (TGPHY160053). We acknowledge support from the NCSA and the SPIN (Students Pushing Innovation) Program at NCSA. PK gratefully acknowledges support for this research at CITA from NSERC of Canada, the Ontario Early Researcher Awards Program, the Canada Research Chairs Program, and the Canadian Institute for Advanced Research. CJM has received funding from the European Union’s Horizon 2020 research and innovation programme under the Marie SkłodowskaCurie grant No 690904, and from STFC Consolidator Grant No. ST/L000636/1. RH is supported by NSF grant 1550514. We thank the NCSA Gravity Group for useful feedback and suggestions, and Ian Hinder for a painstaking review of this manuscript.
Appendix A Convergence of numerical relativity simulations
Using Richardson Extrapolation, Figure 10 presents the waveform phase error of a sample of eccentric NR simulations that we have generated to validate ENIGMA. We notice that the accumulated waveform phase error at merger, which is defined as the amplitude peak of the mode waveform, is . A detailed analysis of these simulations is presented in an accompanying paper Huerta et al. (2017b).
References
 Abbott et al. (2016a) B. P. Abbott, R. Abbott, T. D. Abbott, M. R. Abernathy, F. Acernese, K. Ackley, C. Adams, T. Adams, P. Addesso, R. X. Adhikari, and et al., Physical Review Letters 116, 061102 (2016a), arXiv:1602.03837 [grqc] .
 Abbott et al. (2016b) B. P. Abbott, R. Abbott, T. D. Abbott, M. R. Abernathy, F. Acernese, K. Ackley, C. Adams, T. Adams, P. Addesso, R. X. Adhikari, and et al., Physical Review Letters 116, 241103 (2016b), arXiv:1606.04855 [grqc] .
 Abbott et al. (2017a) B. P. Abbott, R. Abbott, T. D. Abbott, M. R. Abernathy, F. Acernese, K. Ackley, C. Adams, T. Adams, P. Addesso, R. X. Adhikari, et al., Physical Review Letters 118, 221101 (2017a).
 Abbott et al. (2017b) B. P. Abbott, R. Abbott, T. D. Abbott, F. Acernese, K. Ackley, C. Adams, T. Adams, P. Addesso, R. X. Adhikari, V. B. Adya, and et al., Physical Review Letters 119, 141101 (2017b), arXiv:1709.09660 [grqc] .
 The LIGO Scientific Collaboration et al. (2017a) The LIGO Scientific Collaboration, the Virgo Collaboration, B. P. Abbott, R. Abbott, T. D. Abbott, F. Acernese, K. Ackley, C. Adams, T. Adams, P. Addesso, and et al., ArXiv eprints (2017a), arXiv:1711.05578 [astroph.HE] .
 Abbott et al. (2017c) B. P. Abbott, R. Abbott, T. D. Abbott, F. Acernese, K. Ackley, C. Adams, T. Adams, P. Addesso, R. X. Adhikari, V. B. Adya, and et al., Physical Review Letters 119, 161101 (2017c), arXiv:1710.05832 [grqc] .
 Abbott et al. (2016c) B. P. Abbott, R. Abbott, T. D. Abbott, M. R. Abernathy, F. Acernese, K. Ackley, C. Adams, T. Adams, P. Addesso, R. X. Adhikari, and et al., Physical Review Letters 116, 131103 (2016c), arXiv:1602.03838 [grqc] .
 The LIGO Scientific Collaboration et al. (2015) The LIGO Scientific Collaboration, J. Aasi, et al., Classical and Quantum Gravity 32, 074001 (2015), arXiv:1411.4547 [grqc] .
 Acernese et al. (2015) F. Acernese et al., Classical and Quantum Gravity 32, 024001 (2015), arXiv:1408.3978 [grqc] .
 Abbott et al. (2016d) B. P. Abbott, R. Abbott, T. D. Abbott, M. R. Abernathy, F. Acernese, K. Ackley, C. Adams, T. Adams, P. Addesso, R. X. Adhikari, and et al., Astrophys. J. Lett 818, L22 (2016d), arXiv:1602.03846 [astroph.HE] .
 Abbott et al. (2016e) B. P. Abbott, R. Abbott, T. D. Abbott, M. R. Abernathy, F. Acernese, K. Ackley, C. Adams, T. Adams, P. Addesso, R. X. Adhikari, and et al., Physical Review X 6, 041015 (2016e), arXiv:1606.04856 [grqc] .
 Abbott et al. (2016f) B. P. Abbott, R. Abbott, T. D. Abbott, M. R. Abernathy, F. Acernese, K. Ackley, C. Adams, T. Adams, P. Addesso, R. X. Adhikari, and et al., Physical Review Letters 116, 241102 (2016f), arXiv:1602.03840 [grqc] .
 Einstein (1915) A. Einstein, Königlich Preussische Akademie der Wissenschaften Zu Berlin, Sitzungberichte 1915, 844 (1915).
 Abbott et al. (2017d) B. P. Abbott, R. Abbott, T. D. Abbott, F. Acernese, K. Ackley, C. Adams, T. Adams, P. Addesso, R. X. Adhikari, V. B. Adya, and et al., Astrophys. J. Lett 848, L12 (2017d), arXiv:1710.05833 [astroph.HE] .
 The LIGO Scientific Collaboration et al. (2017b) The LIGO Scientific Collaboration, the Virgo Collaboration, B. P. Abbott, R. Abbott, T. D. Abbott, F. Acernese, K. Ackley, C. Adams, T. Adams, P. Addesso, and et al., ArXiv eprints (2017b), arXiv:1710.05836 [astroph.HE], arXiv:1710.05836 [astroph.HE] .
 Peters (1964) P. C. Peters, Phys. Rev. 136, B1224 (1964).
 Peters and Mathews (1963) P. C. Peters and J. Mathews, Phys. Rev. D 131, 435 (1963).
 Huerta et al. (2017a) E. A. Huerta, P. Kumar, B. Agarwal, D. George, H.Y. Schive, H. P. Pfeiffer, R. Haas, W. Ren, T. Chu, M. Boyle, D. A. Hemberger, L. E. Kidder, M. A. Scheel, and B. Szilagyi, Phys. Rev. D 95, 024038 (2017a).
 Hinder et al. (2017) I. Hinder, L. E. Kidder, and H. P. Pfeiffer, ArXiv eprints (2017), arXiv:1709.02007 [grqc] .
 Clausen et al. (2013) D. Clausen, S. Sigurdsson, and D. F. Chernoff, MNRAS 428, 3618 (2013), arXiv:1210.8153 [astroph.HE] .
 Samsing et al. (2014) J. Samsing, M. MacLeod, and E. RamirezRuiz, Astrophys. J. 784, 71 (2014), arXiv:1308.2964 [astroph.HE] .
 Samsing and RamirezRuiz (2017) J. Samsing and E. RamirezRuiz, Astrophys. J. Lett 840, L14 (2017), arXiv:1703.09703 [astroph.HE] .
 Samsing (2017) J. Samsing, ArXiv eprints (2017), arXiv:1711.07452 [astroph.HE] .
 Rodriguez et al. (2017) C. L. Rodriguez, P. AmaroSeoane, S. Chatterjee, and F. A. Rasio, ArXiv eprints (2017), arXiv:1712.04937 [astroph.HE] .
 Gondán et al. (2017) L. Gondán, B. Kocsis, P. Raffai, and Z. Frei, ArXiv eprints (2017), arXiv:1705.10781 [astroph.HE] .
 Hoang et al. (2017a) B.M. Hoang, S. Naoz, B. Kocsis, F. A. Rasio, and F. Dosopoulou, ArXiv eprints (2017a), arXiv:1706.09896 [astroph.HE] .
 VanLandingham et al. (2016) J. H. VanLandingham, M. C. Miller, D. P. Hamilton, and D. C. Richardson, ArXiv eprints (2016), arXiv:1604.04948 [astroph.HE] .
 Wen (2003) L. Wen, Astrophys. J. 598, 419 (2003), arXiv:astroph/0211492 .
 Osburn et al. (2016) T. Osburn, N. Warburton, and C. R. Evans, Phys. Rev. D 93, 064024 (2016), arXiv:1511.01498 [grqc] .
 Maccarone et al. (2007) T. J. Maccarone, A. Kundu, S. E. Zepf, and K. L. Rhode, Nature (London) 445, 183 (2007), astroph/0701310 .
 Strader et al. (2012) J. Strader, L. Chomiuk, T. J. Maccarone, J. C. A. MillerJones, and A. C. Seth, Nature (London) 490, 71 (2012), arXiv:1210.0901 [astroph.HE] .
 Chomiuk et al. (2013) L. Chomiuk, J. Strader, T. J. Maccarone, J. C. A. MillerJones, C. Heinke, E. Noyola, A. C. Seth, and S. Ransom, Astrophys. J. 777, 69 (2013), arXiv:1306.6624 [astroph.HE] .
 Antonini et al. (2016) F. Antonini, S. Chatterjee, C. L. Rodriguez, M. Morscher, B. Pattabiraman, V. Kalogera, and F. A. Rasio, Astrophys. J. 816, 65 (2016), arXiv:1509.05080 .
 Rodriguez et al. (2015) C. L. Rodriguez, M. Morscher, B. Pattabiraman, S. Chatterjee, C.J. Haster, and F. A. Rasio, Physical Review Letters 115, 051101 (2015), arXiv:1505.00792 [astroph.HE] .
 Rodriguez et al. (2016) C. L. Rodriguez, S. Chatterjee, and F. A. Rasio, Phys. Rev. D 93, 084029 (2016), arXiv:1602.02444 [astroph.HE] .
 Usman et al. (2016) S. A. Usman, A. H. Nitz, I. W. Harry, C. M. Biwer, D. A. Brown, M. Cabero, C. D. Capano, T. Dal Canton, T. Dent, S. Fairhurst, M. S. Kehl, D. Keppel, B. Krishnan, A. Lenon, A. Lundgren, A. B. Nielsen, L. P. Pekowsky, H. P. Pfeiffer, P. R. Saulson, M. West, and J. L. Willis, Classical and Quantum Gravity 33, 215004 (2016), arXiv:1508.02357 [grqc] .
 Abbott et al. (2016g) B. P. Abbott, R. Abbott, T. D. Abbott, M. R. Abernathy, F. Acernese, K. Ackley, C. Adams, T. Adams, P. Addesso, R. X. Adhikari, and et al., Phys. Rev. D 93, 122004 (2016g), arXiv:1602.03843 [grqc] .
 Tiwari et al. (2016) V. Tiwari, S. Klimenko, N. Christensen, E. A. Huerta, S. R. P. Mohapatra, A. Gopakumar, M. Haney, P. Ajith, S. T. McWilliams, G. Vedovato, M. Drago, F. Salemi, G. A. Prodi, C. Lazzaro, S. Tiwari, G. Mitselmakher, and F. Da Silva, Phys. Rev. D 93, 043007 (2016), arXiv:1511.09240 [grqc] .
 Huerta et al. (2014a) E. A. Huerta, P. Kumar, S. T. McWilliams, R. O’Shaughnessy, and N. Yunes, Phys. Rev. D 90, 084016 (2014a), arXiv:1408.3406 [grqc] .
 Huerta and Brown (2013) E. A. Huerta and D. A. Brown, Phys. Rev. D 87, 127501 (2013), arXiv:1301.1895 [grqc] .
 Antonini et al. (2014) F. Antonini, N. Murray, and S. Mikkola, Astrophys. J. 781, 45 (2014), arXiv:1308.3674 [astroph.HE] .
 Moore and Gair (2014) C. J. Moore and J. R. Gair, Physical Review Letters 113, 251101 (2014), arXiv:1412.3657 [grqc] .
 Moore et al. (2016) C. J. Moore, C. P. L. Berry, A. J. K. Chua, and J. R. Gair, Phys. Rev. D 93, 064001 (2016), arXiv:1509.04066 [grqc] .
 Goodfellow et al. (2016) I. Goodfellow, Y. Bengio, and A. Courville, Deep Learning (MIT Press, 2016).
 Mackay (2003) D. J. C. Mackay, Information Theory, Inference and Learning Algorithms, by David J. C. MacKay, pp. 640. ISBN 0521642981. Cambridge, UK: Cambridge University Press, October 2003. (2003) p. 640.
 East et al. (2013) W. E. East, S. T. McWilliams, J. Levin, and F. Pretorius, Phys. Rev. D 87, 043004 (2013), arXiv:1212.0837 [grqc] .
 Kelly et al. (2011) B. J. Kelly, J. G. Baker, W. D. Boggs, S. T. McWilliams, and J. Centrella, Phys. Rev. D 84, 084009 (2011), arXiv:1107.1181 [grqc] .
 Cao and Han (2017) Z. Cao and W.B. Han, ArXiv eprints (2017), arXiv:1708.00166 [grqc] .
 Taracchini et al. (2012) A. Taracchini, Y. Pan, A. Buonanno, E. Barausse, M. Boyle, T. Chu, G. Lovelace, H. P. Pfeiffer, and M. A. Scheel, Phys. Rev. D 86, 024011 (2012), arXiv:1202.0790 [grqc] .
 Hinderer and Babak (2017) T. Hinderer and S. Babak, ArXiv eprints (2017), arXiv:1707.08426 [grqc] .
 Bohé et al. (2017) A. Bohé, L. Shao, A. Taracchini, A. Buonanno, S. Babak, I. W. Harry, I. Hinder, S. Ossokine, M. Pürrer, V. Raymond, T. Chu, H. Fong, P. Kumar, H. P. Pfeiffer, M. Boyle, D. A. Hemberger, L. E. Kidder, G. Lovelace, M. A. Scheel, and B. Szilágyi, Phys. Rev. D 95, 044028 (2017), arXiv:1611.03703 [grqc] .
 Nakamura et al. (1987) T. Nakamura, K. Oohara, and Y. Kojima, Progress of Theoretical Physics Supplement 90, 1 (1987).
 Shibata and Nakamura (1995) M. Shibata and T. Nakamura, Phys. Rev. D 52, 5428 (1995).
 Baumgarte and Shapiro (1998) T. W. Baumgarte and S. L. Shapiro, Phys. Rev. D 59, 024007 (1998), grqc/9810065 .
 Baker et al. (2006) J. G. Baker, J. Centrella, D.I. Choi, M. Koppitz, and J. van Meter, Physical Review Letters 96, 111102 (2006), grqc/0511103 .
 Campanelli et al. (2006) M. Campanelli, C. O. Lousto, P. Marronetti, and Y. Zlochower, Physical Review Letters 96, 111101 (2006), grqc/0511048 .
 Pollney et al. (2011) D. Pollney, C. Reisswig, E. Schnetter, N. Dorband, and P. Diener, Phys. Rev. D 83, 044045 (2011), arXiv:0910.3803 [grqc] .
 Wardell et al. (2016) B. Wardell, I. Hinder, and E. Bentivegna, “Simulation of GW150914 binary black hole merger using the Einstein Toolkit,” (2016).
 Löffler et al. (2012) F. Löffler, J. Faber, E. Bentivegna, T. Bode, P. Diener, R. Haas, I. Hinder, B. C. Mundim, C. D. Ott, E. Schnetter, G. Allen, M. Campanelli, and P. Laguna, Classical and Quantum Gravity 29, 115001 (2012), arXiv:1111.3344 [grqc] .
 Ansorg et al. (2004) M. Ansorg, B. Brügmann, and W. Tichy, Phys. Rev. D 70, 064011 (2004), arXiv:grqc/0404056 .
 Diener et al. (2007) P. Diener, E. N. Dorband, E. Schnetter, and M. Tiglio, J. Sci. Comput. 32, 109 (2007), arXiv:grqc/0512001 .
 Schnetter et al. (2004) E. Schnetter, S. H. Hawley, and I. Hawke, Class. Quantum Grav. 21, 1465 (2004), arXiv:grqc/0310042 .
 Thornburg (2004) J. Thornburg, Class. Quantum Grav. 21, 743 (2004), arXiv:grqc/0306056 .
 Levin et al. (2011) J. Levin, S. T. McWilliams, and H. Contreras, Classical and Quantum Gravity 28, 175001 (2011), arXiv:1009.2533 [grqc] .
 Yunes et al. (2009) N. Yunes, K. G. Arun, E. Berti, and C. M. Will, Phys. Rev. D 80, 084001 (2009), arXiv:0906.0313 [grqc] .
 Hinder et al. (2010) I. Hinder, F. Herrmann, P. Laguna, and D. Shoemaker, Phys. Rev. D 82, 024033 (2010).
 Loutrel and Yunes (2017a) N. Loutrel and N. Yunes, Classical and Quantum Gravity 34, 044003 (2017a), arXiv:1607.05409 [grqc] .
 Loutrel et al. (2014) N. Loutrel, N. Yunes, and F. Pretorius, Phys. Rev. D 90, 104010 (2014), arXiv:1404.0092 [grqc] .
 Loutrel and Yunes (2017b) N. Loutrel and N. Yunes, Classical and Quantum Gravity 34, 135011 (2017b), arXiv:1702.01818 [grqc] .
 Hinder et al. (2008) I. Hinder, B. Vaishnav, F. Herrmann, D. M. Shoemaker, and P. Laguna, Phys. Rev. D 77, 081502 (2008), arXiv:0710.5167 [grqc] .
 East et al. (2012) W. E. East, F. Pretorius, and B. C. Stephens, Phys. Rev. D 85, 124009 (2012), arXiv:1111.3055 [astroph.HE] .
 East and Pretorius (2012) W. E. East and F. Pretorius, Astrophys. J. Lett 760, L4 (2012), arXiv:1208.5279 [astroph.HE] .
 Gold et al. (2012) R. Gold, S. Bernuzzi, M. Thierfelder, B. Brügmann, and F. Pretorius, Phys. Rev. D 86, 121501 (2012), arXiv:1109.5128 [grqc] .
 Gold and Brügmann (2013) R. Gold and B. Brügmann, Phys. Rev. D 88, 064051 (2013), arXiv:1209.4085 [grqc] .
 Paschalidis et al. (2015) V. Paschalidis, W. E. East, F. Pretorius, and S. L. Shapiro, Phys. Rev. D 92, 121502 (2015), arXiv:1510.03432 [astroph.HE] .
 East et al. (2016) W. E. East, V. Paschalidis, F. Pretorius, and S. L. Shapiro, Phys. Rev. D 93, 024011 (2016), arXiv:1511.01093 [astroph.HE] .
 Radice et al. (2016) D. Radice, F. Galeazzi, J. Lippuner, L. F. Roberts, C. D. Ott, and L. Rezzolla, MNRAS 460, 3255 (2016), arXiv:1601.02426 [astroph.HE] .
 The LIGO Scientific Collaboration et al. (2016) The LIGO Scientific Collaboration, the Virgo Collaboration, B. P. Abbott, R. Abbott, T. D. Abbott, M. R. Abernathy, F. Acernese, K. Ackley, C. Adams, T. Adams, and et al., ArXiv eprints (2016), arXiv:1611.07531 [grqc] .
 Lewis et al. (2017) A. G. M. Lewis, A. Zimmerman, and H. P. Pfeiffer, Classical and Quantum Gravity 34, 124001 (2017), arXiv:1611.03418 [grqc] .
 Huerta et al. (2017b) E. A. Huerta et al., (2017b).
 Arun et al. (2009) K. G. Arun, L. Blanchet, B. R. Iyer, and S. Sinha, Phys. Rev. D 80, 124018 (2009), arXiv:0908.3854 [grqc] .
 Chua et al. (2017) A. J. K. Chua, C. J. Moore, and J. R. Gair, Phys. Rev. D 96, 044005 (2017), arXiv:1705.04259 [grqc] .
 Barausse et al. (2012) E. Barausse, A. Buonanno, and A. Le Tiec, Phys. Rev. D 85, 064010 (2012), arXiv:1111.5610 [grqc] .
 Bini and Damour (2014) D. Bini and T. Damour, Phys. Rev. D 89, 064063 (2014), arXiv:1312.2503 [grqc] .
 Huerta and Gair (2011a) E. A. Huerta and J. R. Gair, Phys. Rev. D 83, 044020 (2011a), arXiv:1009.1985 [grqc] .
 Huerta and Gair (2011b) E. A. Huerta and J. R. Gair, Phys. Rev. D 83, 044021 (2011b), arXiv:1011.0421 [grqc] .
 Huerta et al. (2014b) E. A. Huerta, P. Kumar, J. R. Gair, and S. T. McWilliams, Phys. Rev. D 90, 024024 (2014b), arXiv:1403.0561 [grqc] .
 Huerta and Gair (2011c) E. A. Huerta and J. R. Gair, Phys. Rev. D 84, 064023 (2011c), arXiv:1105.3567 [grqc] .
 Heffernan et al. (2017) A. Heffernan, A. C. Ottewill, N. Warburton, B. Wardell, and P. Diener, ArXiv eprints (2017), arXiv:1712.01098 [grqc] .
 Fujita (2012) R. Fujita, Progress of Theoretical Physics 128, 971 (2012), arXiv:1211.5535 [grqc] .
 Huerta et al. (2012) E. A. Huerta, P. Kumar, and D. A. Brown, Phys. Rev. D 86, 024024 (2012), arXiv:1205.5562 [grqc] .
 Huerta and Gair (2010) E. A. Huerta and J. R. Gair, in Laser interferometer space antenna. Proceedings, 8th International LISA Symposium, Stanford, USA, June 28July 2, 2010 (2010) arXiv:1009.5882 [grqc] .
 Chua and Gair (2015) A. J. K. Chua and J. R. Gair, Classical and Quantum Gravity 32, 232002 (2015), arXiv:1510.06245 [grqc] .
 Huerta and Gair (2009) E. A. Huerta and J. R. Gair, Phys. Rev. D 79, 084021 (2009), arXiv:0812.4208 .
 Blackman et al. (2015) J. Blackman, S. E. Field, C. R. Galley, B. Szilágyi, M. A. Scheel, M. Tiglio, and D. A. Hemberger, Physical Review Letters 115, 121102 (2015), arXiv:1502.07758 [grqc] .
 Doctor et al. (2017) Z. Doctor, B. Farr, D. E. Holz, and M. Pürrer, ArXiv eprints (2017), arXiv:1706.05408 [astroph.HE] .
 Blanchet (2006) L. Blanchet, Living Reviews in Relativity 9, 4 (2006).
 Brown and Zimmerman (2010) D. A. Brown and P. J. Zimmerman, Phys. Rev. D 81, 024007 (2010).
 Mroué et al. (2013) A. H. Mroué, M. A. Scheel, B. Szilágyi, H. P. Pfeiffer, M. Boyle, D. A. Hemberger, L. E. Kidder, G. Lovelace, S. Ossokine, N. W. Taylor, A. Zenginoğlu, L. T. Buchman, T. Chu, E. Foley, M. Giesler, R. Owen, and S. A. Teukolsky, Physical Review Letters 111, 241104 (2013), arXiv:1304.6077 [grqc] .
 Moore et al. (2016) C. J. Moore, A. J. K. Chua, C. P. L. Berry, and J. R. Gair, Royal Society Open Science 3 (2016), 10.1098/rsos.160125.
 Rasmussen and Williams (2005) C. E. Rasmussen and C. K. I. Williams, Gaussian Processes for Machine Learning (Adaptive Computation and Machine Learning) (The MIT Press, 2005).
 Shoemaker (2015) D. Shoemaker, “Advanced LIGO anticipated sensitivity curves,” (2015), https://dcc.ligo.org/LIGOT0900288v2/public.
 Kumar et al. (2016) P. Kumar, T. Chu, H. Fong, H. P. Pfeiffer, M. Boyle, D. A. Hemberger, L. E. Kidder, M. A. Scheel, and B. Szilagyi, Phys. Rev. D93, 104050 (2016), arXiv:1601.05396 [grqc] .
 Johnson et al. (2018) D. Johnson, E. A. Huerta, and R. Haas, Classical and Quantum Gravity 35, 027002 (2018), arXiv:1708.02941 [grqc] .
 George et al. (2017a) D. George, H. Shen, and E. A. Huerta, ArXiv eprints (2017a), arXiv:1711.07468 [astroph.IM] .
 Klimenko et al. (2008) S. Klimenko, I. Yakushin, A. Mercer, and G. Mitselmakher, Classical and Quantum Gravity 25, 114029 (2008), arXiv:0802.3232 [grqc] .
 George and Huerta (2017a) D. George and E. A. Huerta, ArXiv eprints (2017a), arXiv:1701.00008 [astroph.IM] .
 George et al. (2017b) D. George, H. Shen, and E. A. Huerta, ArXiv eprints (2017b), arXiv:1706.07446 [grqc] .
 Shen et al. (2017) H. Shen, D. George, E. A. Huerta, and Z. Zhao, ArXiv eprints (2017), arXiv:1711.09919 [grqc] .
 George and Huerta (2017b) D. George and E. A. Huerta, ArXiv eprints (2017b), arXiv:1711.03121 [grqc] .
 George and Huerta (2017c) D. George and E. A. Huerta, ArXiv eprints (2017c), arXiv:1711.07966 [grqc] .
 Hoang et al. (2017b) B.M. Hoang, S. Naoz, B. Kocsis, F. A. Rasio, and F. Dosopoulou, ArXiv eprints (2017b), arXiv:1706.09896 [astroph.HE] .
 Apostolatos (1995) T. A. Apostolatos, Phys. Rev. D 52, 605 (1995).
 Kennedy and Eberhart (1995) J. Kennedy and R. Eberhart, in Neural Networks, 1995. Proceedings., IEEE International Conference on, Vol. 4 (1995) pp. 1942–1948 vol.4.
 (115) A. Lee, “Pyswarm,” http://pythonhosted.org/pyswarm/.
 Brown et al. (2013) D. A. Brown, P. Kumar, and A. H. Nitz, Phys. Rev. D 87, 082004 (2013), arXiv:1211.6184 [grqc] .
 Capano et al. (2013) C. Capano, Y. Pan, and A. Buonanno, ArXiv eprints (2013), arXiv:1311.1286 [grqc] .
 London et al. (2017) L. London, S. Khan, E. FauchonJones, C. García, M. Hannam, S. Husa, X. Jiménez Forteza, C. Kalaghatgi, F. Ohme, and F. Pannarale, ArXiv eprints (2017), arXiv:1708.00404 [grqc] .
 Huerta et al. (2017c) E. A. Huerta et al., (2017c).
Comments
There are no comments yet.