2
inferred eccentricities cannot be directly compared
with each other. A detailed comparison between
the two waveform models is given in Knee et al.
[68].
•Parameter space degeneracy We quantified the
impact of including higher order modes in terms
of fitting factor calculations that aim to pinpoint
an optimal quasicircular NRHybSur3dq8 waveform
signal [69] whose astrophysical parameters best re-
produce the complex morphology of moderately or
highly eccentric numerical relativity waveforms.
These three complementary studies underscore the im-
portance of improving our understanding of compact bi-
nary mergers in dense stellar environments. It is not
enough to hope for the best and expect that burst or ma-
chine learning searches identify complex signals in gravi-
tational wave data [46, 48]. It is also necessary to develop
a comprehensive toolkit that encompasses numerical rel-
ativity waveforms, semi-analytical or machine learning
based models, and signal processing tools to detect and
then infer the astrophysical properties of eccentric com-
pact binary mergers. Not doing so would be a disservice
to the proven detection capabilities of advanced gravita-
tional wave detectors, and would limit the science reach
of gravitational wave astrophysics. To contribute to this
important endeavor, we release our catalog of numerical
relativity waveforms along with this article.
This article is organized as follows. We describe our
approach to create a catalog of eccentric numerical rel-
ativity waveforms in Sec. II. Sec. IV presents our wave-
form catalog, and a systematic study on the importance
of including higher order wave modes in terms of SNR
calculations. In Sec. V we study whether surrogate mod-
els based on quasicircular, spinning, nonprecessing binary
black hole numerical relativity waveforms can capture the
physics of spinning, nonprecessing eccentric mergers. We
summarize our findings and future directions of work in
Sec. VI.
II. NUMERICAL SETUP AND SIMULATION
DETAILS
We used the Einstein Toolkit to generate a catalog
of numerical relativity waveforms. Initial data for the bi-
naries was computed using the TwoPunctures code. The
evolution was done with the CTGamma code implement-
ing the 3+1 BSSN formulation. The outer boundary of
the simulation domain was placed far enough (2500M) to
avoid any contamination of the signal until 200M after
the merger. Each simulation was run at three resolutions
to check for convergence (see appendix A): N= 36,40,44
where Nis the resolution across the finest grid radius.
The highest resolution simulations were used for all anal-
yses. Further details of the simulation setup are given in
[3]. Waveforms extracted at future null infinity were com-
puted for 1 < l ≤4 and 1 ≤ |m| ≤ lmodes using the
POWER code [70] by extrapolating the observed signals
from 7 detectors located 100–700M. m= 0 modes were
not used since these modes (so-called memory modes)
are many orders of magnitude smaller than the dominant
modes of the waveform making a reliable estimation diffi-
cult due to numerical resolution (for more details see Sec.
6.2 in [71]). A plot of all the h+simulation waveforms
is shown in Fig. 1. Note that the simulations are also
dimensionalized in units of M.
Table I describes the properties of our waveform cata-
log, including the mass-ratio, individual spins and orbital
eccentricity of each binary (measured from both wave-
form templates). The library consists of 27 simulations
across 3 mass ratios, q={2,4,6}, and a combination of
nonprecessing individual spins, namely ±0.6 and ±0.3,
for the primary (heavier) and secondary (lighter) binary
components, respectively.
III. ECCENTRICITY MEASUREMENTS
Orbital eccentricity in a Keplerian interpretation can
only be defined for a BBH system during the early in-
spiral, where the orbits of the binaries are nearly closed
(the adiabatic approximation). This definition breaks
down close to the merger, which is when our simulations
begin. Thus, the definitions of eccentricity used to gen-
erate the initial conditions are ill-defined, even though
they produce eccentric simulations.
Using evolution information of the binary, such as the
separation between the components, throughout the sim-
ulation to obtain a measure of orbital eccentricity is not
useful, as such a concept is gauge-dependent by assum-
ing a coordinate system. To obtain a useful measure of
eccentricity, we calibrate our numerical simulations to
the spin-aligned eccentric EOB models TEOBResumS and
SEOBNRE. For both of these models, a reference eccentric-
ity e0and reference GW frequency fref are used as in-
puts to generate adiabatic initial conditions of the binary
from which the waveform is computed. As investigated
in Knee et al. [68], each waveform model’s definition of e0
may vary, due to different conventions of fref and initial
condition constructions.
The method is similar to that used in [18, 66]. The key
idea consists of using `=|m|= 2 waveforms to compute
the fitting factor between a given numerical relativity
waveform, and an array of templates. In this work, we
have assigned fref = 10Hz, which is at the lower end of
the detectability range for LIGO. To estimate the eccen-
tricity of our numerical relativity waveforms, we need to
compute a few objects. The first of them is the inner
product between one of our numerical relativity wave-
forms, hNR
22 , and a waveform template, htemplate
22 , given
by:
hhNR
22 |htemplate
22 i=RZt2
t1
hNR
22 h* template
22 .(1)
Where Rrepresents the real component. Note that the