Accuracy of numerical relativity waveforms with respect to space-based gravitational

wave detectors
Zun Wang,1, 2 Junjie Zhao,1, 2 and Zhoujian Cao ∗1, 2, 3, †
1
Institute for Frontiers in Astronomy and Astrophysics,
Beijing Normal University, Beijing 102206, China
2
Department of Astronomy, Beijing Normal University, Beijing 100875, China
3
School of Fundamental Physics and Mathematical Sciences,
Hangzhou Institute for Advanced Study, UCAS, Hangzhou 310024, China
The same to laser interferometer gravitational-wave observatory (LIGO), matched filtering tech-
nique will be critical to data analysis of gravitational wave detection by space-based detectors
including LISA, Taiji and Tianqin. Waveform templates are basis for such matched filtering tech-
nique. In order to construct ready-to-use waveform templates, numerical relativity waveforms are
start point. So the accuracy issue of numerical relativity waveforms is critically important. There
are many investigations about this issue with respect to LIGO. But unfortunately there are few
results on this issue with respect to space-based detectors. The current paper investigates this
problem. Our results indicate that the existing numerical relativity waveforms are as accurate as
99% with respect to space-based detectors including LISA, Taiji and Tianqin. Such accuracy level
arXiv:2401.15331v1 [gr-qc] 27 Jan 2024

is comparable to the one with respect to LIGO.

I. INTRODUCTION binary systems is available currently. This fact also ex-

plains to some extent why only events of coalescence of
Since the first successful detection of gravitational binary objects are observed till now. In contrast, cur-
waves in 2015, about 100 gravitational wave events have rent detection ability for supernovae gravitational waves
been reported. All these events are found by matched is quite weak [17]. Roughly the detection horizon is just
filtering technique. Till now matched filtering technique 1kpc (Fig. 5(a) of [17]). The partial reason for such fact
is still the standard data analysis trick for gravitational is the lack of accurate waveform template for supernovae
waves detection [1, 2]. As expected, the situation will gravitational waves.
be also true for space-based detectors in the near future Before the breakthrough of numerical relativity [18–
[3, 4]. 22], the waveform template problem is treated mainly
In order to let the matched filtering technique work, through post-Newtonian approximation [23]. As an en-
accurate and complete waveform templates are needed hanced post-Newtonian approximation method, effective
[5–8]. Besides the above mentioned matched filtering one body theory shows better convergent behavior [24].
method, there is time-frequency excess power identifica- After the success of numerical relativity simulation of
tion method [9]. But if accurate waveform is available binary black hole merger, the complete inspiral-merger-
and aid the analysis, the method will become more effi- ringdown behavior is revealed. The power of effective
cient [10]. In addition, machine learning method is a new one body theory to describe the waveform of coalescence
trick to treat gravitational wave data [11–16]. A data set of binary object is verified [25]. After that the numeri-
consisting a large amount of gravitational wave samples cal relativity waveforms are extensively used to construct
is critical important to let the machine learning method waveform templates for coalescence of binary objects.
work well. Since the real gravitational wave events are Till now, there are a bundle of waveform templates for
too few to play the role of such a data set, accurate wave- coalescence of binary objects available in the LIGO data
form templates are needed. In a short summary, no mat- analysis software. Among kinds of waveform template
ter what kind of data analysis means are taken, waveform models including EOBNR series [26–31], IMRphenom se-
templates are very important to gravitational wave data ries [32, 33], numerical relativity surrogate models [34–36]
analysis. and others, the numerical relativity waveforms are bases
A waveform template means the accurate waveform for the waveform templates construction. When people
with respect to time or frequency when a set of source talk about the accuracy of a waveform model, the nu-
parameters are specified. That is to say a waveform tem- merical relativity waveforms are treated as the standard
plate is valid only for a class of source falling in a specified answer. So the accuracy of numerical relativity wave-
parameters range. Till now gravitational wave astron- forms themselves are critically important to waveform
omy community only understand binary compact object template construction.
systems well. Consequently only waveform template of Both ground-based and space-based gravitational wave
detectors utilize matched filtering techniques for detect-
ing gravitational waves. Therefore, when calculating the
accuracy of the templates, the formulas used are very
∗ corresponding author similar to those of the matched filtering technique. Con-
† Zhoujian Cao: [email protected] sidering the detector noise is crucial when using the
 2

matched filtering technique to search for signals. Hence, Then similar matching factor to (1) can be defined to
when calculating the accuracy of the waveform tem- quantify the closeness between two complex waveforms.
plates, it is necessary to take into account the sensi- The only difference to (1) is the maximum should be
tivity of different detectors. The noise characteristics taken with respect to the initial phase besides the shifted
of space-based gravitational wave detectors differ signifi- time. The initial phase describes the phase difference
cantly from those of ground-based detectors. Therefore, between the two polarization modes h+ (t) and h× (t) at
the purpose of this article is to investigate whether the the initial time.
numerical relativity waveform’s accuracy can meet the re- Assume we have two complex waveforms h1,2 = h1,2+ −
quirement of space-based detectors. When ones discuss ih1,2× , the linearity of inner product (2) results in
the accuracy of a waveform model, a specific detector
should be referred. The accuracy issue of numerical rela- hh1 |h2 i = hh1+ |h2+ i + hh1× |h2× i
tivity waveforms has been extensively studied against ad- − ihh1+ |h2× i − ihh1× |h2+ i. (5)
vanced LIGO detectors [37]. But this issue has not been
investigated against space-based detectors. The current In the mean time Eq. (2) can be equivalently expressed
paper aims to do such an investigation and lays down a as
foundation of waveform template construction for space-
h̃1 h̃∗2
Z
based detectors.
hh1 |h2 i = 4ℜ df, (6)
In the next section we introduce the waveform accu- S(f )
racy estimation method. After that we apply the method
in Sec. III to calculate waveform accuracy of the wave- where ℜ means taking the real part. So we have
forms of SXS numerical relativity catalog. LISA, Taiji
and Tianqin detectors are all considered. Finally the hh1 |h2 i = hh1+ |h2+ i + hh1× |h2× i, (7)
summary and conclusion are given in the last section.
which corresponds to

II. MATCHING FACTOR AND ACCURACY

maxhh1 |h2 i = maxhh1+ |h2+ i + maxhh1× |h2× i. (8)
t,φ t+ t×
INDICATOR
Here t means the time shift for the complex waveform, φ
Following the idea of matched filtering data analy- means the initial phase difference of the two polarization
sis trick, matching factor has been extensively used to modes, and t+,× are the time shifts for the two polariza-
quantify how close between two given waveforms. With tion waveforms. Due to the above relations, we have
respect to a detector sensitivity S(f ) which describes
the one sided power spectrum of the detector noise, the FF+ kh1+ k · kh2+ k + FF× kh1× k · kh2× k
FF = , (9)
matching factor of two real waveforms h1 (t) and h2 (t) kh1 k · kh2 k
can be expressed as hh1+ |h2+ i
FF+ ≡ max , (10)
t kh1+ k · kh2+ k
hh1 |h2 i
FF ≡ max , (1) hh1× |h2× i
t kh1 k · kh2 k FF× ≡ max . (11)
fup
t kh1× k · kh2× k
h̃1 h̃∗2 + h̃∗1 h̃2
Z
hh1 |h2 i = 2 df, (2)
flow S(f ) That is to say we can calculate the matching factors FF+
p and FF× for two polarization modes individually, then
khk ≡ hh|hi, (3) use the above equation to combine the final matching
factor we wanted. In the current work we follow this way
˜ means the Fourier transformation, the
where the “(·)” and use PyCBC tool to calculate the matching factor.
∗
“ ” means taking the complex conjugate, and the max- Similar to any other computing science topics, the only
imum is taken with respect to the time shift to align errors involved in numerical relativity include truncation
the two waveforms. (flow , fup ) corresponds to the fre- errors and round off errors. Truncation errors are due
quency band where the two waveforms should be com- to the numerical approximation of derivatives. Round
pared. Within PyCBC software [38], the command line off errors are due to the memory limit of computers. In
‘pycbc.filter.matchedfilter.match’ can be used to do the practice, ones need to make sure the real calculation dom-
above calculation of the matching factor FF. Together inated by truncation errors. Consequently the final error
with the value of the matching factor, the time shift is related to the numerical solution is proportional to some
also returned by the command line. power of the resolution used in the numerical calculation.
For a theoretical waveform template, two polarization The power index is nothing but the convergence order of
waveforms will be given h+ (t) and h× (t). Usually people the involved numerical algorithm. So we can use the dif-
are used to the complex waveform defined as ference between the results of two different resolutions to
quantitatively estimate the error of the numerical solu-
h ≡ h+ − ih× . (4) tion.
 3

In the current work we use the matching factor between

the two numerical relativity waveforms of two different
resolutions to quantify the accuracy of the numerical rel-
ativity waveforms. Specifically to the SXS waveform cat-
alogs [37, 39], the finest and second finest resolutions are
used.

III. ACCURACY OF NUMERICAL

RELATIVITY WAVEFORMS

A. Fourier transforms of numerical relativity

waveforms
FIG. 1: Frequency waveform of SXS:BBH:2106. This wave-
form corresponds to a quasi-circular coalescing binary black
Numerical relativity (NR) waveforms are presented in
hole system with mass ratio 1, dimensionless spin χ ~1 =
time domain. In order to calculate the matching factor (0, 0, 0.8998) and χ
~ 2 = (0, 0, 0.5). In the plot, M means the
explained in the last section, we need transform these total mass of the binary. The horizontal axis has no special
waveforms to frequency domain. In practice, we use fast meaning. It just indicates different NR simulations.
Fourier transformation to get the waveforms in frequency
domain.
In order to reduce the Gibbs effect and spectral leakage means the total mass of the binary system. Different nu-
resulting from truncation in the time domain, we apply merical relativity waveforms begin at different frequency
the Plank window σT (t) to the time domain waveform corresponding to M fmin. M fmin ranges from about 0.002
before the Fourier transformation. The Plank window to 0.012. Most waveforms admit M fmin ≈ 0.006. Lower
σT (t) is set as [40, 41] M fmin means the corresponding binary system begins at
larger separation and the waveform is longer. Roughly
0, t < t1

 M fmax falls in the quasi-normal modes stage. The spe-
 σstart (t), t1 ≤ t < t2

cific value of M fmax depends on the specific numerical

σ(t) = 1, t2 ≤ t < t3 (12) simulation. In the viewpoint of the resolution require-



 σend (t), t 3 ≤ t < t4 ment of the binary system in question, if the numerical
0, t4 ≤ t resolution is higher the value of M fmax is larger. Rela-
tively the numerical setting is random, so the behavior
where σstart is the segment that smoothly increases from of M fmax shown in Fig. 2(b) is random.
0 to 1 between t1 and t2 , and σend is the segment that
smoothly decreases from 1 to 0 between t3 and t4 :
h   i−1 C. Accuracy of numerical relativity waveforms
σstart (t) = exp tt−t2 −t1
1
+ t2 −t1
t−t 2
+ 1 , with respect to LIGO
h   i−1 (13)
σend (t) = exp tt−t3 −t4
3
+ tt−t
3 −t4
4
+1 . For comparison convenience, we also investigate the
accuracy of numerical relativity waveforms with respect
Then, We further zero pad the waveform to the nearest to LIGO detectors. Specifically we use the designed sen-
power of 2. sitivity of advanced LIGO [42]. The frequency band of
LIGO is (10, 8192)Hz.
Note that only the numerical relativity waveform
B. Frequency range of numerical relativity falling in the range (M fmin , M fmax ) is trustable. Con-
waveforms sidering the source character for LIGO, we investigate
M ∈ (10, 200)M⊙. In Fig. 3 we show the trustable fre-
Typically we get waveforms in frequency domain like quency range for binary system with total mass M =
the one shown in Fig. 1. Apparently only the part be- 10M⊙. For other total mass systems we need only rescale
tween the two vertical dash lines is reliable. The left the vertical axis proportional to the inverse of the sys-
vertical line corresponds to the lowest frequency fmin of tem total mass 1/M . From Fig. 3(a) we can see clearly
the numerical relativity waveform which is determined by that the numerical relativity simulation can not cover
the length of the waveform. The right vertical line corre- the whole frequency range of LIGO detection. This is
sponds the highest frequency fmax where the numerical due to the well known expensive computational cost of
error begins to dominate. numerical relativity. Consequently numerical relativity
There are 1872 waveforms in the SXS catalog [39] who only starts near merger. For early inspiral part, people
have more than one resolution result. In Fig. 2(a) and (b) rely on post-Newtonian approximation to construct the
we plot M fmin and M fmax of these waveforms. Here M waveform template. In the current work, we just care
 4

FIG. 2: Frequency lower and upper limit of the 1872 numeri-

cal relativity waveforms in SXS catalog. The top panel is the FIG. 3: Trustable frequency range of numerical relativity
lower limit M fmin . The bottom panel shows both the lower waveforms of the 1872 numerical relativity waveforms in SXS
limit (black dots) and the upper limit (blue dots). catalog for M = 10M⊙ binary system. The top plot is the
lower limit M fmin . The bottom plot shows both the lower
limit (black dots) and the upper limit (blue dots).
about the accuracy of numerical relativity, so we take
the integrand bound in (2) as  2 !
6 f
1+ , (17)
flow = max(10, fmin), (14) 10 f∗
fup = min(8192, fmax). (15) f∗ = c/(2πL). (18)
In the first panel of Fig. 4 we plot the mismatch factors For LISA [48] we have

M ≡ 1 − FF (16) POMS = (1.5 × 10−11 m)2 Hz−1 , (19)

2 !
4 × 10−4 Hz

with respect to LIGO between the highest resolution sim- Pacc = (3 × 10−15 ms−2 )2 1+ Hz−1 ,
ulation and the second highest resolution simulation. f
(20)
9
L = 2.5 × 10 m. (21)
D. Accuracy of numerical relativity waveforms
with respect to space-based detectors For Taiji [49] we have

Regarding space-based detectors, we consider LISA [4, POMS = (8 × 10−12 m)2 Hz−1 , (22)
43], Taiji [44] and Tianqin [45, 46] as examples. We do  −4
2 !
4 × 10 Hz
not involve realistic response functions as [47], instead we Pacc = (3 × 10−15 ms−2 )2 1+ Hz−1 ,
use sky averaged sensitivity [48] to do the estimation. f
Specifically we use the following approximated sen- (23)
sitivity for space based gravitational wave detectors 9
L = 3 × 10 m. (24)
(Eq. (13) of [48])
  For Tianqin we have [45]
10 2 Pacc
Sn (f ) = POMS + 2(1 + cos (f /f∗ )) ×
3L2 (2πf )4 POMS = (1 × 10−12 m)2 Hz−1 , (25)
 5

FIG. 4: The mismatch factors between the highest resolution simulation and the second highest resolution simulation. There
are in all 1872 SXS waveforms are investigated here. From top to bottom, from left to right the subfigures correspond to LIGO,
LISA, Taiji and Tianqin respectively. The blue lines in all of the subfigures correspond to SXS:BBH:1131.

2 !
1 × 10−4 Hz can be approximated as [50]

−2 2
Pacc = (1 × 10 −15
ms ) 1+ Hz−1 ,
f α
+βf sin(κf )
Sc (f ) =Af −7/3 e−f ×
(26)
√ [1 + tanh (γ (fk − f ))] Hz−1 (28)
L= 3 × 108 m. (27) −45
A = 9 × 10 , (29)
α = 0.133, (30)
β = 243, (31)
κ = 482, (32)
γ = 917, (33)
fk = 0.00258. (34)
Note that parameters α, β, κ, γ depend on observation
time. The values listed above correspond to observation
time half year. The overall noise sensitivity of space-
based detectors can be estimated as
S = Sn + Sc . (35)
Due to the similar reason for LIGO, we take the inte-
grand bound in (2) as
FIG. 5: Similar to Fig. 4 but with detector frequency range
(10−4 , 0.1)Hz in stead of (10−5 , 1)Hz. This plot is for LISA. flow = max(10−5 , fmin ), (36)
fup = min(1, fmax ), (37)
Besides the instrument noise mentioned above, there
is more confusion noise due to the galaxy binaries which for space-based detectors.
 6

TABLE I: Less accurate (M > 1%) NR simulations found in this work.

Here, we list the parameters for each simulation, including mass ratio q,
lowest frequency M fmin , highest frequency M fmax , and initial spin con-
figuration. Additionally, we provide the maximum mismatch between
the highest resolution simulation and the second highest resolution sim-
ulation max M. Here max means the maximum value in the total mass
range shown in Fig. 4. The subscriptions ‘LIGO’, ‘LISA’, ‘Taiji’ and
‘Tianqin’ are for corresponding detectors.

SXS ID q χ1 χ2 M fmin M fmax max MLIGO max MLISA max MTaiji max MTianqin
1415 1.50 (0.00,0.00,0.50) (0.00,-0.00,0.50) 0.0017 0.1789 0.0756 0.0978 0.1005 0.1077
0627 1.91 (-0.51,0.44,-0.35) (0.19,-0.01,-0.06) 0.0083 0.1626 0.0709 0.0650 0.0660 0.0491
1413 1.41 (-0.00,-0.00,0.50) (-0.00,-0.00,0.40) 0.0017 0.1660 0.0570 0.0732 0.0755 0.0805
1414 1.83 (-0.00,-0.00,-0.50) (0.00,-0.00,0.40) 0.0017 0.1636 0.0527 0.0695 0.0715 0.0749
1390 1.42 (0.15,0.44,-0.16) (-0.02,0.34,0.10) 0.0017 0.1659 0.0510 0.0648 0.0672 0.0714
1393 1.79 (-0.37,-0.33,-0.00) (-0.27,-0.39,0.11) 0.0017 0.1784 0.0490 0.0647 0.0667 0.0694
1392 1.51 (-0.40,0.23,0.17) (0.35,-0.13,-0.25) 0.0017 0.1795 0.0478 0.0625 0.0642 0.0679
1389 1.63 (-0.29,0.20,-0.30) (-0.01,0.42,0.16) 0.0017 0.1771 0.0460 0.0594 0.0613 0.0651
1391 1.83 (-0.15,0.29,-0.33) (-0.33,-0.29,-0.03) 0.0017 0.1813 0.0440 0.0566 0.0583 0.0619
1412 1.63 (-0.00,-0.00,0.40) (-0.00,0.00,-0.30) 0.0017 0.1822 0.0421 0.0562 0.0578 0.0606
1416 1.78 (0.00,-0.00,-0.40) (-0.00,0.00,-0.40) 0.0017 0.1825 0.0392 0.0524 0.0539 0.0561
1926 4.00 (0.76,0.26,0.04) (0.00,-0.14,0.79) 0.0065 0.1943 0.0266 0.0350 0.0337 0.0335
2000 4.00 (-0.40,0.69,0.08) (0.45,0.65,-0.11) 0.0066 0.1902 0.0171 0.0206 0.0204 0.0204
1992 4.00 (-0.61,0.07,-0.51) (-0.27,0.75,-0.05) 0.0062 0.1733 0.0162 0.0192 0.0195 0.0194
2044 4.00 (0.74,-0.29,0.11) (0.14,-0.60,0.52) 0.0067 0.1890 0.0148 0.0151 0.0149 0.0140
1991 4.00 (-0.26,-0.51,-0.56) (-0.07,0.06,0.79) 0.0061 0.1724 0.0136 0.0201 0.0219 0.0221
2038 4.00 (-0.80,-0.05,0.05) (-0.01,-0.08,-0.39) 0.0065 0.1920 0.0135 0.0249 0.0244 0.0246
2054 4.00 (0.66,-0.45,0.08) (0.38,-0.31,0.63) 0.0065 0.1887 0.0135 0.0199 0.0190 0.0192
2074 4.00 (-0.66,0.44,0.07) (-0.74,0.28,0.10) 0.0066 0.1548 0.0127 0.0214 0.0210 0.0211
1987 4.00 (0.38,0.43,-0.55) (0.54,0.58,0.04) 0.0062 0.1659 0.0119 0.0160 0.0177 0.0182
1110 7.00 (-0.00,-0.00,0.00) (-0.00,-0.00,-0.00) 0.0023 0.1724 0.0106 0.0510 0.0423 0.0427
1928 4.00 (-0.33,0.72,0.07) (0.62,0.48,-0.13) 0.0065 0.1898 0.0104 0.0114 0.0115 0.0113
1978 4.00 (0.50,0.26,0.57) (-0.77,-0.20,0.03) 0.0070 0.1438 0.0092 0.0171 0.0168 0.0179
1135 1.00 (-0.00,-0.00,-0.44) (-0.00,0.00,-0.44) 0.0073 0.1355 0.0089 0.0134 0.0137 0.0097
1623 3.93 (0.02,0.55,0.43) (-0.56,-0.33,-0.45) 0.0066 0.1863 0.0089 0.0187 0.0184 0.0176
1993 4.00 (-0.06,-0.58,-0.54) (-0.24,-0.76,0.02) 0.0062 0.1439 0.0087 0.0109 0.0116 0.0121
1994 4.00 (0.58,0.16,-0.53) (0.11,-0.79,-0.08) 0.0062 0.1938 0.0085 0.0121 0.0118 0.0108
1981 4.00 (0.26,-0.55,-0.52) (-0.35,-0.72,-0.02) 0.0062 0.1460 0.0081 0.0120 0.0114 0.0101
1156 4.39 (-0.16,0.21,0.38) (0.53,-0.55,0.11) 0.0041 0.1776 0.0079 0.0085 0.0088 0.0106
1629 3.46 (0.54,0.15,-0.45) (-0.23,0.08,-0.73) 0.0059 0.2096 0.0077 0.0089 0.0097 0.0104
1923 4.00 (-0.79,0.04,0.09) (-0.75,0.28,0.02) 0.0066 0.1622 0.0074 0.0188 0.0184 0.0169
2011 4.00 (0.79,-0.09,0.03) (0.37,0.69,0.18) 0.0063 0.2168 0.0064 0.0292 0.0260 0.0243
1863 3.63 (-0.45,0.30,-0.58) (0.33,0.33,0.43) 0.0060 0.1654 0.0063 0.0114 0.0133 0.0137
1997 4.00 (-0.76,-0.24,0.04) (-0.00,0.14,0.79) 0.0063 0.1663 0.0060 0.0114 0.0104 0.0101
1983 4.00 (-0.47,0.35,-0.55) (-0.52,-0.59,0.14) 0.0062 0.1721 0.0058 0.0130 0.0136 0.0144
2005 4.00 (0.36,0.71,0.07) (0.48,0.64,0.06) 0.0065 0.1290 0.0057 0.0136 0.0130 0.0119
2081 4.00 (-0.36,0.71,0.06) (0.62,0.49,-0.14) 0.0065 0.1904 0.0055 0.0104 0.0105 0.0112
2048 4.00 (0.80,-0.02,0.02) (-0.26,0.41,0.64) 0.0065 0.1924 0.0054 0.0092 0.0100 0.0102
1579 3.44 (0.21,0.46,-0.38) (0.18,0.48,-0.59) 0.0062 0.1837 0.0053 0.0137 0.0145 0.0153
1986 4.00 (-0.39,0.45,-0.53) (0.11,0.03,0.79) 0.0063 0.1412 0.0052 0.0123 0.0120 0.0108
2007 4.00 (0.77,0.22,0.04) (0.00,0.15,-0.79) 0.0063 0.1887 0.0051 0.0123 0.0124 0.0136
1979 4.00 (-0.53,0.00,0.60) (-0.03,-0.12,-0.79) 0.0067 0.1697 0.0050 0.0133 0.0132 0.0137
2043 4.00 (-0.70,-0.39,0.06) (-0.50,0.34,0.52) 0.0063 0.1638 0.0050 0.0107 0.0104 0.0105
1975 4.00 (0.45,0.27,0.61) (0.04,-0.13,0.79) 0.0071 0.1899 0.0049 0.0181 0.0190 0.0200
1917 4.00 (0.38,0.71,0.01) (-0.68,0.36,0.20) 0.0066 0.1862 0.0046 0.0101 0.0105 0.0113
1972 4.00 (-0.50,0.20,0.59) (0.80,0.00,-0.06) 0.0068 0.2032 0.0044 0.0094 0.0099 0.0103
1974 4.00 (-0.34,0.43,0.59) (-0.00,-0.00,-0.00) 0.0068 0.2039 0.0044 0.0162 0.0177 0.0190
0147 1.00 (0.40,0.29,-0.00) (-0.40,-0.29,-0.00) 0.0107 0.1616 0.0043 0.0104 0.0099 0.0093
2015 4.00 (0.57,0.56,0.03) (0.04,-0.07,0.39) 0.0065 0.1731 0.0042 0.0110 0.0107 0.0101
0469 1.00 (-0.16,0.78,0.03) (0.04,-0.01,0.40) 0.0059 0.1842 0.0039 0.0111 0.0120 0.0132
1927 4.00 (0.52,-0.61,0.01) (0.03,0.79,0.10) 0.0063 0.1738 0.0037 0.0213 0.0189 0.0169
2034 4.00 (-0.79,-0.07,0.06) (0.39,0.07,-0.03) 0.0066 0.1860 0.0036 0.0121 0.0124 0.0133
2010 4.00 (0.78,0.16,0.02) (0.23,0.75,0.16) 0.0065 0.1819 0.0033 0.0112 0.0101 0.0114
to next page
 7

TABLE I continue
SXS ID q χ1 χ2 M fmin M fmax max MLIGO max MLISA max MTaiji max MTianqin
1973 4.00 (0.29,0.47,0.58) (-0.08,0.07,-0.79) 0.0067 0.1823 0.0029 0.0132 0.0138 0.0150
1614 2.68 (0.20,0.03,0.71) (-0.11,-0.07,0.03) 0.0063 0.1896 0.0028 0.0118 0.0121 0.0134
1713 3.97 (0.05,-0.47,0.30) (0.68,-0.16,-0.34) 0.0066 0.1842 0.0028 0.0092 0.0092 0.0100
1741 2.77 (0.58,-0.51,-0.06) (-0.01,-0.05,-0.45) 0.0061 0.1964 0.0027 0.0092 0.0098 0.0105
2079 4.00 (-0.39,-0.69,0.04) (-0.31,0.73,-0.12) 0.0066 0.1948 0.0027 0.0158 0.0131 0.0132
1209 2.00 (0.06,-0.01,0.85) (-0.19,0.83,0.01) 0.0062 0.1864 0.0026 0.0093 0.0097 0.0107
2004 4.00 (-0.27,-0.75,0.02) (-0.22,0.77,-0.09) 0.0063 0.2052 0.0026 0.0111 0.0107 0.0094
2064 4.00 (-0.44,-0.67,0.00) (0.24,-0.61,-0.46) 0.0063 0.1698 0.0023 0.0102 0.0094 0.0089
0705 2.00 (-0.03,-0.04,0.80) (0.76,-0.26,0.02) 0.0062 0.1925 0.0022 0.0105 0.0113 0.0125
1095 2.00 (0.22,0.77,0.02) (-0.09,0.04,-0.79) 0.0057 0.1743 0.0022 0.0124 0.0119 0.0100
1659 3.47 (-0.07,0.58,0.54) (-0.04,0.17,0.43) 0.0066 0.1842 0.0022 0.0111 0.0110 0.0113
2058 4.00 (-0.18,0.78,0.03) (0.35,-0.33,-0.64) 0.0062 0.1803 0.0022 0.0103 0.0097 0.0086
1591 3.59 (0.31,-0.28,0.50) (0.48,-0.11,0.32) 0.0066 0.1906 0.0020 0.0123 0.0121 0.0130
1399 1.58 (-0.29,-0.20,-0.23) (-0.37,0.03,0.20) 0.0028 0.2196 0.0019 0.0108 0.0118 0.0125
0708 2.00 (0.76,-0.23,0.04) (-0.06,-0.10,0.79) 0.0061 0.1620 0.0018 0.0108 0.0105 0.0094
0968 2.00 (0.07,0.80,-0.01) (-0.60,0.51,0.10) 0.0059 0.1737 0.0018 0.0088 0.0093 0.0102
0888 2.00 (-0.61,-0.51,0.03) (-0.20,-0.42,0.65) 0.0061 0.1805 0.0017 0.0127 0.0123 0.0102
1839 3.76 (0.25,-0.33,0.50) (0.18,-0.54,-0.37) 0.0066 0.1715 0.0017 0.0106 0.0104 0.0112
0835 2.00 (-0.48,-0.64,0.02) (0.00,-0.00,0.00) 0.0059 0.1907 0.0016 0.0097 0.0108 0.0118
0900 2.00 (-0.15,0.79,0.04) (-0.30,0.38,0.64) 0.0061 0.1952 0.0016 0.0086 0.0096 0.0103
1532 3.02 (-0.59,-0.29,0.39) (0.17,0.12,-0.31) 0.0065 0.1981 0.0015 0.0116 0.0114 0.0106
1668 3.43 (0.38,0.13,-0.66) (-0.43,-0.63,0.07) 0.0061 0.1434 0.0015 0.0103 0.0094 0.0082
1929 4.00 (0.43,-0.67,0.08) (0.65,-0.45,0.15) 0.0067 0.1835 0.0014 0.0122 0.0114 0.0097
0733 2.00 (0.35,-0.19,-0.02) (-0.11,0.79,0.06) 0.0060 0.2009 0.0012 0.0108 0.0105 0.0093
1656 3.40 (-0.37,0.19,0.59) (-0.06,-0.08,-0.12) 0.0066 0.1810 0.0012 0.0118 0.0124 0.0134
0664 1.33 (-0.79,-0.12,0.03) (-0.79,-0.10,0.03) 0.0056 0.1881 0.0011 0.0094 0.0100 0.0109
1006 1.03 (0.64,0.21,-0.35) (-0.48,0.18,0.50) 0.0059 0.1814 0.0011 0.0112 0.0109 0.0099
1557 2.94 (0.69,-0.07,0.20) (-0.04,0.79,-0.13) 0.0062 0.1887 0.0011 0.0137 0.0135 0.0136
1696 2.63 (0.67,0.33,0.09) (-0.12,0.16,0.27) 0.0062 0.1709 0.0011 0.0113 0.0108 0.0091
1770 2.55 (0.42,0.28,0.58) (-0.33,0.63,-0.34) 0.0063 0.1912 0.0011 0.0112 0.0116 0.0127
1787 3.23 (0.59,0.37,0.27) (0.14,0.30,-0.67) 0.0063 0.1954 0.0011 0.0122 0.0119 0.0108
0834 1.00 (-0.56,-0.57,0.03) (-0.00,0.00,-0.00) 0.0057 0.2074 0.0010 0.0099 0.0103 0.0114
0907 1.00 (-0.73,-0.33,-0.02) (0.53,-0.05,0.60) 0.0059 0.2097 0.0010 0.0117 0.0125 0.0135
1206 1.00 (0.62,-0.58,-0.05) (0.18,0.83,0.08) 0.0057 0.1797 0.0010 0.0099 0.0102 0.0112
0905 1.00 (0.65,-0.46,0.02) (0.49,-0.11,0.62) 0.0059 0.1925 0.0009 0.0110 0.0108 0.0116
0916 1.00 (-0.77,-0.21,0.01) (-0.56,-0.57,0.08) 0.0057 0.1979 0.0009 0.0097 0.0101 0.0110
0966 2.00 (-0.71,-0.37,0.06) (-0.68,0.42,-0.06) 0.0060 0.2054 0.0009 0.0109 0.0115 0.0125
1149 3.00 (0.00,-0.00,0.70) (-0.00,-0.00,0.60) 0.0063 0.1798 0.0009 0.0132 0.0130 0.0142
1523 2.93 (0.49,-0.26,0.46) (0.41,0.30,0.40) 0.0065 0.1963 0.0009 0.0100 0.0107 0.0117
0750 2.00 (-0.28,-0.48,0.57) (0.07,-0.05,-0.80) 0.0060 0.1768 0.0008 0.0109 0.0111 0.0118
1000 1.21 (0.31,0.63,0.34) (-0.60,-0.02,0.48) 0.0060 0.1853 0.0008 0.0105 0.0103 0.0093
1086 1.07 (-0.33,-0.35,0.63) (0.59,0.18,0.16) 0.0060 0.1990 0.0008 0.0112 0.0115 0.0126
1197 2.00 (-0.78,-0.34,-0.04) (0.65,-0.54,0.10) 0.0059 0.1903 0.0008 0.0098 0.0102 0.0111
1199 2.00 (0.68,-0.51,0.04) (0.10,0.08,-0.84) 0.0057 0.2148 0.0008 0.0093 0.0092 0.0101
1849 2.70 (0.54,-0.00,0.53) (-0.41,0.31,0.34) 0.0063 0.1963 0.0008 0.0106 0.0103 0.0094
2131 2.00 (0.00,0.00,0.85) (0.00,-0.00,0.85) 0.0060 0.1824 0.0008 0.0133 0.0142 0.0156
0601 1.06 (-0.50,0.07,0.59) (0.02,0.04,0.66) 0.0061 0.1974 0.0007 0.0091 0.0095 0.0105
0635 1.00 (0.67,-0.44,0.03) (-0.06,-0.04,0.80) 0.0059 0.1930 0.0007 0.0106 0.0111 0.0119
0170 1.00 (-0.00,-0.00,0.44) (0.00,0.00,0.44) 0.0071 0.1582 0.0006 0.0110 0.0109 0.0121
0323 1.22 (0.00,-0.00,0.33) (-0.00,-0.00,-0.44) 0.0066 0.1553 0.0006 0.0097 0.0095 0.0105
0781 2.00 (0.79,-0.14,0.03) (0.05,0.10,-0.79) 0.0057 0.2059 0.0006 0.0118 0.0116 0.0108
1071 1.07 (-0.13,0.20,0.66) (0.33,-0.56,0.39) 0.0060 0.1995 0.0006 0.0124 0.0133 0.0145
1716 2.24 (-0.29,0.36,0.53) (-0.55,-0.08,0.46) 0.0062 0.1936 0.0006 0.0095 0.0102 0.0112
0256 2.00 (-0.00,0.00,0.60) (-0.00,0.00,0.60) 0.0057 0.1609 0.0005 0.0116 0.0121 0.0133
0351 1.00 (-0.20,0.77,0.03) (0.08,-0.01,0.80) 0.0060 0.1906 0.0005 0.0115 0.0113 0.0108
0936 2.00 (-0.68,-0.42,-0.01) (0.79,0.08,0.04) 0.0060 0.1985 0.0005 0.0104 0.0110 0.0120
0948 2.00 (0.03,0.01,0.80) (-0.42,0.38,-0.56) 0.0061 0.1913 0.0005 0.0095 0.0094 0.0101
1014 1.69 (-0.64,0.10,0.33) (-0.54,0.09,0.46) 0.0061 0.2025 0.0005 0.0123 0.0121 0.0128
1632 3.01 (0.55,-0.51,0.25) (-0.63,-0.13,-0.33) 0.0062 0.1970 0.0005 0.0133 0.0136 0.0147
1718 2.31 (-0.30,0.33,0.61) (-0.38,0.64,-0.09) 0.0062 0.1832 0.0005 0.0108 0.0106 0.0105
to next page
 8

TABLE I continue
SXS ID q χ1 χ2 M fmin M fmax max MLIGO max MLISA max MTaiji max MTianqin
2006 4.00 (-0.49,-0.63,-0.02) (0.75,-0.23,0.14) 0.0063 0.1810 0.0005 0.0111 0.0093 0.0085
0065 8.00 (-0.00,-0.00,0.50) (0.00,0.00,0.00) 0.0067 0.1400 0.0004 0.0115 0.0111 0.0094
0324 1.22 (-0.00,-0.00,0.33) (-0.00,-0.00,-0.44) 0.0088 0.1277 0.0004 0.0101 0.0098 0.0098
0374 2.00 (-0.26,0.47,0.59) (0.00,-0.00,0.00) 0.0062 0.1868 0.0004 0.0099 0.0103 0.0113
0383 1.75 (-0.31,0.74,0.04) (0.10,0.01,0.79) 0.0060 0.1942 0.0004 0.0108 0.0107 0.0113
0476 1.00 (-0.17,0.37,0.44) (0.04,0.00,0.80) 0.0061 0.1891 0.0004 0.0142 0.0148 0.0162
0662 1.33 (-0.68,-0.41,0.03) (-0.02,0.09,0.79) 0.0060 0.1899 0.0004 0.0099 0.0103 0.0112
0688 1.67 (-0.65,-0.46,0.03) (-0.03,0.10,0.79) 0.0061 0.1686 0.0004 0.0101 0.0098 0.0090
0772 2.00 (-0.45,0.66,0.08) (-0.14,0.79,0.01) 0.0060 0.1960 0.0004 0.0130 0.0137 0.0148
0845 2.00 (0.74,-0.30,0.04) (-0.04,-0.05,0.40) 0.0061 0.2014 0.0004 0.0130 0.0138 0.0148
0941 2.00 (-0.03,0.03,0.80) (-0.05,-0.57,-0.56) 0.0062 0.1785 0.0004 0.0094 0.0100 0.0111
0988 2.00 (-0.04,0.03,0.80) (-0.20,-0.77,0.02) 0.0062 0.1619 0.0004 0.0090 0.0096 0.0106
1090 1.59 (-0.30,-0.33,0.48) (-0.30,-0.34,0.52) 0.0061 0.1891 0.0004 0.0117 0.0126 0.0139
1529 3.14 (0.33,-0.59,0.13) (-0.38,0.50,0.48) 0.0063 0.1636 0.0004 0.0102 0.0095 0.0079
1642 3.29 (-0.30,-0.54,0.26) (0.72,-0.15,0.04) 0.0066 0.1904 0.0004 0.0137 0.0132 0.0112
1676 3.25 (0.11,0.18,0.44) (0.31,-0.13,0.22) 0.0066 0.2032 0.0004 0.0108 0.0106 0.0107
1692 2.88 (-0.51,0.31,-0.02) (-0.34,-0.41,-0.07) 0.0060 0.1886 0.0004 0.0111 0.0108 0.0094
0333 2.00 (0.00,0.00,0.80) (-0.00,0.00,0.80) 0.0063 0.1786 0.0003 0.0120 0.0129 0.0143
0348 1.19 (-0.21,0.45,0.60) (0.06,0.01,0.76) 0.0061 0.1781 0.0003 0.0116 0.0114 0.0120
0478 1.32 (-0.23,0.63,0.14) (0.08,-0.00,0.78) 0.0060 0.1859 0.0003 0.0134 0.0131 0.0118
0571 1.09 (0.00,0.08,-0.02) (-0.00,0.00,-0.29) 0.0057 0.1986 0.0003 0.0091 0.0094 0.0104
0575 1.20 (-0.00,0.01,0.39) (0.00,-0.00,0.14) 0.0059 0.1901 0.0003 0.0105 0.0111 0.0123
0691 1.67 (-0.73,-0.31,0.06) (0.02,0.80,-0.07) 0.0059 0.1792 0.0003 0.0104 0.0101 0.0088
0745 2.00 (-0.16,-0.52,0.59) (-0.00,0.00,0.00) 0.0062 0.2040 0.0003 0.0087 0.0094 0.0103
0830 2.00 (0.70,-0.38,0.06) (-0.65,-0.46,-0.08) 0.0059 0.1802 0.0003 0.0123 0.0118 0.0099
0859 1.00 (-0.01,0.04,0.80) (-0.34,-0.21,0.01) 0.0060 0.1943 0.0003 0.0128 0.0126 0.0128
0991 2.00 (-0.67,-0.44,0.03) (-0.32,-0.72,0.12) 0.0060 0.2002 0.0003 0.0121 0.0131 0.0142
1011 1.53 (0.51,0.31,0.31) (0.33,0.48,0.48) 0.0060 0.1857 0.0003 0.0126 0.0132 0.0145
1020 1.24 (0.37,0.38,0.49) (0.53,-0.27,0.52) 0.0060 0.1910 0.0003 0.0095 0.0100 0.0108
1023 1.22 (-0.59,-0.02,0.36) (0.21,-0.63,-0.37) 0.0057 0.1880 0.0003 0.0089 0.0093 0.0102
1063 1.78 (-0.47,0.28,-0.29) (-0.29,-0.41,0.58) 0.0059 0.1965 0.0003 0.0097 0.0103 0.0113
1070 1.20 (-0.15,-0.43,0.64) (-0.42,-0.41,-0.49) 0.0059 0.1691 0.0003 0.0107 0.0111 0.0122
1571 3.44 (-0.28,-0.35,0.64) (0.48,-0.51,-0.10) 0.0068 0.1777 0.0003 0.0087 0.0093 0.0102
1616 2.87 (0.53,0.30,0.42) (-0.38,0.02,-0.54) 0.0062 0.2089 0.0003 0.0128 0.0137 0.0149
1709 3.44 (-0.09,0.20,0.29) (0.14,0.47,-0.60) 0.0065 0.1871 0.0003 0.0101 0.0098 0.0097
1930 4.00 (0.09,0.79,0.04) (-0.17,0.07,-0.78) 0.0065 0.2064 0.0003 0.0103 0.0089 0.0081
2161 3.00 (0.00,-0.00,0.60) (0.00,0.00,0.00) 0.0057 0.1913 0.0003 0.0118 0.0116 0.0125
0178 1.00 (0.00,0.00,0.99) (-0.00,-0.00,0.99) 0.0056 0.1941 0.0002 0.0144 0.0151 0.0165
0372 1.50 (0.00,-0.00,0.80) (-0.00,0.00,-0.40) 0.0060 0.1851 0.0002 0.0088 0.0093 0.0102
0395 1.00 (-0.10,0.42,-0.42) (0.04,-0.01,0.80) 0.0059 0.2164 0.0002 0.0095 0.0101 0.0110
0408 2.00 (-0.29,0.53,0.02) (0.08,0.01,0.80) 0.0061 0.1909 0.0002 0.0113 0.0123 0.0135
0505 1.85 (-0.26,0.51,0.09) (0.08,0.01,0.79) 0.0061 0.2000 0.0002 0.0090 0.0095 0.0104
0655 1.33 (0.61,-0.52,0.03) (0.00,0.00,-0.00) 0.0057 0.2062 0.0002 0.0113 0.0119 0.0130
0679 1.67 (-0.04,-0.04,0.80) (0.79,-0.14,0.03) 0.0062 0.1587 0.0002 0.0118 0.0124 0.0137
0774 2.00 (0.79,-0.12,0.02) (-0.00,-0.00,0.00) 0.0059 0.1851 0.0002 0.0107 0.0104 0.0089
0777 2.00 (0.80,0.05,0.07) (0.75,-0.29,0.01) 0.0059 0.2001 0.0002 0.0127 0.0125 0.0120
0870 1.00 (0.04,-0.01,0.80) (-0.02,0.40,0.01) 0.0060 0.1930 0.0002 0.0103 0.0106 0.0117
0957 2.00 (-0.73,-0.33,0.00) (0.58,-0.07,-0.54) 0.0059 0.1765 0.0002 0.0110 0.0105 0.0090
0963 1.00 (0.58,-0.55,-0.00) (-0.57,0.56,0.00) 0.0057 0.2067 0.0002 0.0108 0.0105 0.0097
1084 1.76 (0.48,-0.16,0.48) (-0.73,0.10,0.30) 0.0061 0.1841 0.0002 0.0131 0.0128 0.0116
1196 1.00 (0.64,-0.55,0.06) (0.64,-0.55,0.06) 0.0057 0.1681 0.0002 0.0110 0.0115 0.0126
1406 1.60 (-0.29,0.29,0.24) (-0.38,-0.01,0.15) 0.0028 0.1984 0.0002 0.0103 0.0110 0.0120
1495 1.00 (-0.00,-0.00,0.78) (0.00,0.00,0.53) 0.0061 0.1920 0.0002 0.0124 0.0132 0.0144
1518 2.08 (0.23,-0.66,0.08) (-0.60,-0.17,0.15) 0.0059 0.1989 0.0002 0.0109 0.0113 0.0123
1645 2.29 (-0.33,-0.38,0.45) (0.00,-0.64,0.46) 0.0062 0.1917 0.0002 0.0115 0.0113 0.0113
1852 3.03 (-0.45,0.25,0.46) (-0.15,0.71,-0.25) 0.0063 0.1921 0.0002 0.0096 0.0099 0.0108
1860 3.42 (-0.35,-0.20,0.64) (0.67,0.33,-0.24) 0.0067 0.1779 0.0002 0.0106 0.0112 0.0123
2097 1.00 (-0.00,0.00,0.30) (0.00,-0.00,-0.00) 0.0051 0.2019 0.0002 0.0098 0.0105 0.0116
2125 2.00 (0.00,-0.00,0.30) (0.00,-0.00,0.30) 0.0056 0.2023 0.0002 0.0105 0.0104 0.0115
0155 1.00 (-0.00,-0.00,0.80) (0.00,0.00,0.80) 0.0055 0.1886 0.0001 0.0123 0.0130 0.0143
to next page
 9

TABLE I continue
SXS ID q χ1 χ2 M fmin M fmax max MLIGO max MLISA max MTaiji max MTianqin
0179 1.50 (-0.00,-0.00,0.99) (0.13,0.05,0.14) 0.0056 0.1677 0.0001 0.0126 0.0130 0.0144
0328 1.00 (0.00,0.00,0.80) (-0.00,-0.00,0.80) 0.0062 0.1871 0.0001 0.0100 0.0105 0.0115
0415 1.00 (0.00,0.00,-0.00) (-0.00,0.00,-0.40) 0.0056 0.1978 0.0001 0.0089 0.0092 0.0102
0495 1.38 (-0.03,0.13,0.01) (0.01,-0.00,0.40) 0.0059 0.1862 0.0001 0.0094 0.0100 0.0111
0564 1.69 (-0.01,0.01,0.27) (0.00,0.00,0.61) 0.0061 0.1912 0.0001 0.0089 0.0091 0.0101
0681 1.67 (0.06,-0.01,0.80) (-0.28,0.75,0.03) 0.0062 0.1897 0.0001 0.0122 0.0129 0.0142
0694 1.67 (-0.07,0.80,0.03) (0.10,-0.02,0.79) 0.0060 0.1938 0.0001 0.0119 0.0118 0.0124
0706 2.00 (-0.02,0.05,0.80) (-0.40,-0.69,0.03) 0.0063 0.1693 0.0001 0.0111 0.0116 0.0128
0752 2.00 (-0.39,0.42,0.56) (-0.48,-0.63,0.12) 0.0062 0.1830 0.0001 0.0107 0.0105 0.0109
0854 2.00 (0.70,-0.39,0.04) (0.36,-0.18,0.02) 0.0059 0.1927 0.0001 0.0138 0.0135 0.0121
0915 2.00 (0.71,-0.36,0.08) (-0.30,-0.74,-0.07) 0.0059 0.1865 0.0001 0.0108 0.0104 0.0091
0964 2.00 (0.70,-0.38,0.01) (-0.73,0.33,-0.01) 0.0060 0.1770 0.0001 0.0116 0.0112 0.0093
1044 1.77 (0.66,0.08,-0.25) (-0.03,-0.74,0.26) 0.0057 0.2069 0.0001 0.0090 0.0095 0.0104
1068 1.46 (0.08,0.02,0.19) (-0.63,0.02,0.43) 0.0060 0.1909 0.0001 0.0095 0.0100 0.0111
1194 2.00 (0.75,-0.39,0.07) (-0.68,-0.49,-0.09) 0.0059 0.1840 0.0001 0.0104 0.0100 0.0083
1477 1.00 (-0.00,-0.00,0.80) (0.00,0.00,0.80) 0.0062 0.1879 0.0001 0.0094 0.0100 0.0110
1521 3.07 (-0.38,0.26,0.39) (0.36,0.44,0.28) 0.0065 0.1469 0.0001 0.0097 0.0103 0.0114
1747 2.66 (-0.21,-0.00,0.70) (-0.09,0.16,-0.39) 0.0065 0.1829 0.0001 0.0119 0.0122 0.0135
1893 2.62 (0.30,0.51,0.51) (-0.31,-0.17,0.71) 0.0065 0.1882 0.0001 0.0136 0.0134 0.0128
2156 3.00 (-0.00,0.00,0.40) (-0.00,0.00,-0.60) 0.0059 0.1788 0.0001 0.0103 0.0101 0.0106
0255 2.00 (0.00,-0.00,0.60) (0.00,-0.00,-0.00) 0.0056 0.1788 0.0000 0.0110 0.0114 0.0125
0418 1.00 (0.00,0.00,-0.00) (-0.00,-0.00,0.40) 0.0059 0.1858 0.0000 0.0090 0.0094 0.0105
0553 1.07 (-0.01,0.03,0.69) (0.00,0.00,0.46) 0.0061 0.1864 0.0000 0.0091 0.0096 0.0106
0581 1.68 (-0.21,0.55,0.50) (0.01,0.00,0.07) 0.0061 0.1932 0.0000 0.0117 0.0119 0.0132
0607 1.50 (-0.04,0.20,0.22) (-0.12,0.37,0.22) 0.0060 0.1865 0.0000 0.0098 0.0103 0.0112
2101 1.00 (-0.00,0.00,0.60) (0.00,-0.00,0.00) 0.0052 0.2245 0.0000 0.0104 0.0111 0.0123

NR waveforms have a critical limitation that they are ulation, the merger part is also the least accurate part of
some short (due to the computational cost) and mainly the waveform. In the frequency domain, when the black
focus on the merger phase. Especially for the gravita- hole mass increases, the merger part moves from right
tional waves emitted by supermassive black hole binaries, to left. Note that the most sensitive range of the detec-
the majority of the evolution occurs in the inspiral phase. tor locates at the center. For relative small mass BBHs,
Therefore, simply calculating the accuracy of NR wave- the merger part waveform locates at the right side of the
forms will lose the important inspiral phase, which will aforementioned sensitive frequency range. When black
affect the results of the accuracy of the waveforms. In fu- hole mass increases, the merger part falls into the sen-
ture work, we plan to use the PN(Post-Newtonian)-NR sitive frequency range. Consequently the mismatch fac-
waveform models including SEOBNR, SEOBNRE and tor increases. When the black hole mass increases more,
others to investigate the waveform template accuracy for the merger part waveform leaves the sensitive frequency
space-based detectors. range. So the mismatch factor decreases consequently.
The corresponding mismatch factors between the high- Comparing to the result for LIGO, we find that the
est resolution simulation and the second highest resolu- numerical relativity accuracy for space-based detectors
tion simulation for LISA, Taiji and Tianqin are shown is comparable to that for ground-based detectors. That
in Fig. 4. Similar to the situation for LIGO, most NR is to say if the accuracy requirement is similar to that of
simulations admit accuracy better than 99%. A few NR LIGO, the current numerical relativity simulation results
simulations have less accuracy. We list these less accurate can satisfy the need of space-based detectors.
simulations in Tab. I. Considering that the frequency range of space-based
From Fig. 4, we can see SXS:BBH:1131 has very large detector may not reach (10−5 , 1)Hz, we have also calcu-
mismatch factor. This means ones must take caution lated the mismatch factor with replacing (36) and (37)
when using SXS:BBH:1131 result. For other simulations with
listed in Tab. I, ones also have to note the specific accu- flow = max(10−4 , fmin ), (38)
racy requirement when using those simulation results.
fup = min(0.1, fmax ). (39)
For all lines of Fig. 4, there is a typical behavior that
the line increases along with the black hole mass and then The results are almost the same as Fig. 4. Since the
decreases. We can understand this fact as follows. Due to results for LISA, Taiji and Tianqin are similar to each
the numerical error accumulation, the merger part of the other, we only plot LISA as the example in Fig. 5.
waveform corresponds to the least accurate part of the The frequency range of numerical relativity waveform
waveform. Due to the resolution requirement of the sim- shown in Fig. 1 is the most optimal one. We can see
 10

FIG. 6: Similar to Fig. 4 but with flow = max(10−5 , 1.2fmin )Hz (left panel) and flow = max(10−5 , 1.5fmin )Hz (right panel)
instead of flow = max(10−5 , fmin )Hz. Like Fig. 5, we again use LISA as the example.

FIG. 7: Similar to Fig. 4 but with fup = min(1, 0.8fmax )Hz (left panel) and fup = min(1, 0.5fmax )Hz (right panel) instead of
fup = min(1, fmax )Hz. Like Fig. 5, we again use LISA as the example.

choice, we have also considered

flow = max(10−5 , 1.2fmin), (40)

fup = min(1, fmax ), (41)

and

flow = max(10−5 , 1.5fmin), (42)

fup = min(1, fmax ). (43)

Similar to Fig. 5, we once again use LISA as example and

plot the results in Fig. 6 for these two frequency range
choices. As ones expected, when we consider shorter in-
spiral part, the waveform accuracy becomes higher. So
FIG. 8: Similar to Fig. 4 but with flow =
we can see several lines above 10−2 in the left panel of
max(10−5 , 1.5fmin )Hz and fup = min(1, 0.8fmax )Hz fre- Fig. 6 fall down below 10−2 in the right panel.
quency choice instead of (36) and (37). Like Fig. 5, we again Regarding to high frequency side, we check how the
use LISA as the example. cutting frequency affects the waveform accuracy. For
comparison we have compared the results plotted in
Fig. 4 to frequency choices

clear unphysical oscillation near the low frequency fmin . flow = max(10−5 , fmin ), (44)
In order to check the influence of such frequency range fup = min(1, 0.8fmax), (45)
 11

and development without intermediate analytical steps, using

techniques such as reduced order modeling.
flow = max(10−5 , fmin ), (46) Several coordinated efforts have been undertaken
fup = min(1, 0.5fmax ). (47) to produce numerical relativity simulations of binary
black hole mergers for gravitational wave applications.
The result is shown in Fig. 7. As ones expected, the These include the Numerical Injection Analysis (NINJA)
high frequency side affects large black hole mass systems project [51], the collaboration between Numerical Rela-
more. But in all, the influence is small. tivity and Analytical Relativity (NRAR), and the wave-
And more we have also considered conservative fre- form catalogs released by the SXS collaboration and
quency range choice on both low and high frequency side Georgia Tech.
In this work, we use numerical simulations of binary
flow = max(10−5 , 1.5fmin), (48) black hole mergers performed by the SXS Collaboration
fup = min(1, 0.8fmax). (49) using the Spectral Einstein Code (SpEC). The SXS cat-
alog has been used to construct SEOBNRE waveform
The result is plotted in Fig. 8. In a short summary, model [28–31] and other waveform models. The accu-
the different frequency choices roughly result in similar racy of numerical relativity waveform is very important
waveform accuracy. to gravitational wave astronomy study.
In previous works, the accuracy issue of numerical rel-
ativity waveform has been well studied for ground based
IV. SUMMARY AND CONCLUSION
detectors. In the current paper, we focus on space-based
detectors. We have systematically investigated the effect
One of the most challenging and fascinating problems of the waveform frequency range, the detector sensitiv-
in gravitational physics is to understand the dynamics ity detail, the BBH’s black hole mass and others on the
of binary black hole mergers in the strong-field regime. waveform accuracy issue.
In this regime, the components of the binary move at
Each waveform of SXS catalog has been investigated.
relativistic speeds and the spacetime curvature becomes
Special attention is payed to matching factor calculation
highly nonlinear, making analytical approximations in-
between highest and second highest resolution used in the
adequate. The only reliable way to obtain precise solu-
numerical simulations. Our calculation results indicate
tions to Einstein’s field equations in this regime is to use
that the numerical relativity waveforms are as accurate
numerical relativity, which involves solving the full non-
as 99% with respect to space-based detectors including
linear equations on high-performance computers. This
LISA, Taiji and Tianqin. Such accuracy level is compara-
breakthrough was achieved in 2005 after decades of ef-
ble to the one with respect to LIGO. If only the accuracy
forts [22].
requirement for space-based detectors is similar to that
Numerical relativity simulations of binary black hole
of ground-based ones, the current numerical relativity
mergers are essential for modeling the gravitational wave
waveforms are valid for waveform modelling.
signals emitted by these systems during their late inspi-
ral, merger, and ringdown phases. These signals are used
to infer the properties of the source systems and to test
general relativity in extreme conditions. All binary black
Acknowledgments
hole detections made by LIGO and Virgo have been an-
alyzed using waveform models that incorporate numer-
ical relativity data. The most prominent examples of This work was supported in part by the National
these models are the effective-one-body and phenomeno- Key Research and Development Program of China Grant
logical waveform models. Numerical relativity also plays No. 2021YFC2203001 and in part by the NSFC
a key role in validating these models and testing their (No. 11920101003, No. 12021003 and No. 12005016).
accuracy and robustness. Moreover, numerical relativity Z. Cao was supported by “the Interdiscipline Research
waveforms can be directly used for parameter estima- Funds of Beijing Normal University” and CAS Project
tion, template bank construction, and waveform family for Young Scientists in Basic Research YSBR-006.

[1] The LIGO Scientific Collaboration, the Virgo Collab- tion, the KAGRA Collaboration, R. Abbott, et al. Open
oration, the KAGRA Collaboration, R. Abbott, et al. data from the third observing run of LIGO, Virgo, KA-
GWTC-3: Compact Binary Coalescences Observed by GRA and GEO. arXiv e-prints, page arXiv:2302.03676,
LIGO and Virgo During the Second Part of the Third February 2023.
Observing Run. arXiv e-prints, page arXiv:2111.03606, [3] M. Coleman Miller and Nicols Yunes. The new frontier
November 2021. of gravitational waves. Nature, 568(7753):469–476, Apr
[2] The LIGO Scientific Collaboration, the Virgo Collabora- 2019.
 12

[4] Pau Amaro-Seoane et al. Astrophysics with the Laser advanced virgo. Phys. Rev. D, 101:084002, Apr 2020.
Interferometer Space Antenna. Living Reviews in Rela- [18] Frans Pretorius. Evolution of binary black-hole space-
tivity, 26(1):2, December 2023. times. Phys. Rev. Lett., 95:121101, Sep 2005.
[5] Piotr Jaranowski and Andrzej Królak. Gravitational- [19] M. Campanelli, C. O. Lousto, P. Marronetti, and Y. Zlo-
Wave Data Analysis. Formalism and Sample Appli- chower. Accurate evolutions of orbiting black-hole bina-
cations: The Gaussian Case. arXiv e-prints, page ries without excision. Phys. Rev. Lett., 96:111101, Mar
arXiv:0711.1115, November 2007. 2006.
[6] Andrzej Królak. Principles of Gravitational-Wave Data [20] John G. Baker, Joan Centrella, Dae-Il Choi, Michael
Analysis. In Handbook of Gravitational Wave Astronomy, Koppitz, and James van Meter. Gravitational-wave ex-
page 43. 2021. traction from an inspiraling configuration of merging
[7] Lorenzo Speri, Nikolaos Karnesis, Arianna I. Renzini, black holes. Phys. Rev. Lett., 96:111102, Mar 2006.
and Jonathan R. Gair. A roadmap of gravitational wave [21] Zhoujian Cao, Hwei-Jang Yo, and Jui-Ping Yu. Rein-
data analysis. Nature Astronomy, 6:1356–1363, Decem- vestigation of moving punctured black holes with a new
ber 2022. code. Phys. Rev. D, 78:124011, Dec 2008.
[8] Nelson Christensen and Renate Meyer. Parameter es- [22] Tianyu Zhao, Zhoujian Cao, Chun-Yu Lin, and Hwei-
timation with gravitational waves. Rev. Mod. Phys., Jang Yo. Numerical Relativity for Gravitational Wave
94:025001, Apr 2022. Source Modelling, pages 1–30. Springer Singapore, Sin-
[9] Eric Chassande-Mottin, Eric Lebigot, Hugo Magaldi, Eve gapore, 2020.
Chase, Archana Pai, Gayathri V, and Gabriele Vedovato. [23] Curt Cutler and Éanna E. Flanagan. Gravitational waves
Wavelet graphs for the direct detection of gravitational from merging compact binaries: How accurately can one
waves. arXiv e-prints, page arXiv:1710.09256, October extract the binary’s parameters from the inspiral wave-
2017. form? Phys. Rev. D, 49:2658–2697, Mar 1994.
[10] P. Bacon, V. Gayathri, E. Chassande-Mottin, A. Pai, [24] A. Buonanno and T. Damour. Effective one-body ap-
F. Salemi, and G. Vedovato. Driving unmodeled proach to general relativistic two-body dynamics. Phys.
gravitational-wave transient searches using astrophysical Rev. D, 59:084006, Mar 1999.
information. Phys. Rev. D, 98:024028, Jul 2018. [25] Alessandra Buonanno, Yi Pan, John G. Baker, Joan
[11] Elena Cuoco, Jade Powell, Marco Cavaglià, Kendall Centrella, Bernard J. Kelly, Sean T. McWilliams, and
Ackley, Michal Bejger, Chayan Chatterjee, Michael James R. van Meter. Approaching faithful templates for
Coughlin, Scott Coughlin, Paul Easter, Reed Essick, nonspinning binary black holes using the effective-one-
Hunter Gabbard, Timothy Gebhard, Shaon Ghosh, body approach. Phys. Rev. D, 76:104049, Nov 2007.
Leila Haegel, Alberto Iess, David Keitel, Zsuzsa Marka, [26] Alejandro Bohé, Lijing Shao, Andrea Taracchini,
Szabolcs Marka, Filip Morawski, Tri Nguyen, Rich Alessandra Buonanno, Stanislav Babak, Ian W. Harry,
Ormiston, Michael Puerrer, Massimiliano Razzano, Kai Ian Hinder, Serguei Ossokine, Michael Pürrer, Vivien
Staats, Gabriele Vajente, and Daniel Williams. Enhanc- Raymond, Tony Chu, Heather Fong, Prayush Kumar,
ing Gravitational-Wave Science with Machine Learning. Harald P. Pfeiffer, Michael Boyle, Daniel A. Hemberger,
arXiv e-prints, page arXiv:2005.03745, May 2020. Lawrence E. Kidder, Geoffrey Lovelace, Mark A. Scheel,
[12] Marlin B. Schäfer, Ondřej Zelenka, Alexander H. Nitz, and Béla Szilágyi. Improved effective-one-body model of
He Wang, Shichao Wu, Zong-Kuan Guo, Zhoujian Cao, spinning, nonprecessing binary black holes for the era of
Zhixiang Ren, Paraskevi Nousi, Nikolaos Stergioulas, gravitational-wave astrophysics with advanced detectors.
Panagiotis Iosif, Alexandra E. Koloniari, Anastasios Phys. Rev. D, 95:044028, Feb 2017.
Tefas, Nikolaos Passalis, Francesco Salemi, Gabriele [27] Danilo Chiaramello and Alessandro Nagar. Faithful
Vedovato, Sergey Klimenko, Tanmaya Mishra, Bernd analytical effective-one-body waveform model for spin-
Brügmann, Elena Cuoco, E. A. Huerta, Chris Messenger, aligned, moderately eccentric, coalescing black hole bi-
and Frank Ohme. First machine learning gravitational- naries. Phys. Rev. D, 101:101501, May 2020.
wave search mock data challenge. Phys. Rev. D , [28] Zhoujian Cao and Wen-Biao Han. Waveform model for
107(2):023021, January 2023. an eccentric binary black hole based on the effective-
[13] He Wang, Shichao Wu, Zhoujian Cao, Xiaolin Liu, and one-body-numerical-relativity formalism. Phys. Rev. D,
Jian-Yang Zhu. Gravitational-wave signal recognition of 96:044028, Aug 2017.
ligo data by deep learning. Phys. Rev. D, 101:104003, [29] Xiaolin Liu, Zhoujian Cao, and Lijing Shao. Validat-
May 2020. ing the effective-one-body numerical-relativity waveform
[14] Heming Xia, Lijing Shao, Junjie Zhao, and Zhoujian Cao. models for spin-aligned binary black holes along eccentric
Improved deep learning techniques in gravitational-wave orbits. Phys. Rev. D, 101:044049, Feb 2020.
data analysis. Phys. Rev. D, 103:024040, Jan 2021. [30] Xiaolin Liu, Zhoujian Cao, and Zong-Hong Zhu. A
[15] CunLiang Ma, Wei Wang, He Wang, and Zhoujian Cao. higher-multipole gravitational waveform model for an
Ensemble of deep convolutional neural networks for real- eccentric binary black holes based on the effective-
time gravitational wave signal recognition. Phys. Rev. D, one-body-numerical-relativity formalism. Classical and
105:083013, Apr 2022. Quantum Gravity, 39(3):035009, February 2022.
[16] Cunliang Ma, Wei Wang, He Wang, and Zhoujian [31] Xiaolin Liu, Zhoujian Cao, and Lijing Shao. Up-
Cao. Artificial intelligence model for gravitational wave graded waveform model of eccentric binary black hole
search based on the waveform envelope. Phys. Rev. D, based on effective-one-body-numerical-relativity for spin-
107:063029, Mar 2023. aligned binary black holes. International Journal of Mod-
[17] B. P. Abbott et al. Optically targeted search for gravita- ern Physics D, 32:2350015, Feb 2023.
tional waves emitted by core-collapse supernovae during [32] Geraint Pratten, Sascha Husa, Cecilio Garcı́a-Quirós,
the first and second observing runs of advanced ligo and Marta Colleoni, Antoni Ramos-Buades, Héctor Estellés,
 13

and Rafel Jaume. Setting the cornerstone for a family domain templates and simulated signals in the detection
of models for gravitational waves from compact binaries: of gravitational waves from coalescing compact binaries.
The dominant harmonic for nonprecessing quasicircular Classical and Quantum Gravity, 27(8):084020, 2010.
black holes. Phys. Rev. D , 102(6):064001, September [42] D Shoemaker (LIGO Scientific Collaboration). 2010
2020. advanced ligo anticipated sensitivity curves ligo doc-
[33] Cecilio Garcı́a-Quirós, Marta Colleoni, Sascha Husa, ument t0900288-v3. URL https://dcc.ligo.org/cgi-
Héctor Estellés, Geraint Pratten, Antoni Ramos-Buades, bin/DocDB/ShowDocument?docid=2974, 2010.
Maite Mateu-Lucena, and Rafel Jaume. Multimode [43] M. Armano et al. Charge-induced force noise on free-
frequency-domain model for the gravitational wave sig- falling test masses: Results from lisa pathfinder. Phys.
nal from nonprecessing black-hole binaries. Phys. Rev. Rev. Lett., 118:171101, Apr 2017.
D, 102:064002, Sep 2020. [44] Wen-Hong Ruan, Chang Liu, Zong-Kuan Guo, Yue-
[34] Jonathan Blackman, Scott E. Field, Mark A. Scheel, Liang Wu, and Rong-Gen Cai. The lisa-taiji network.
Chad R. Galley, Christian D. Ott, Michael Boyle, Nature Astronomy, 4(2):108–109, Feb 2020.
Lawrence E. Kidder, Harald P. Pfeiffer, and Béla Szilágyi. [45] Jun Luo, Li-Sheng Chen, Hui-Zong Duan, Yun-Gui
Numerical relativity waveform surrogate model for gener- Gong, Shoucun Hu, Jianghui Ji, Qi Liu, Jianwei Mei,
ically precessing binary black hole mergers. Phys. Rev. Vadim Milyukov, Mikhail Sazhin, et al. Tianqin: a space-
D, 96:024058, Jul 2017. borne gravitational wave detector. Classical and Quan-
[35] Vijay Varma, Scott E. Field, Mark A. Scheel, Jonathan tum Gravity, 33(3):035010, 2016.
Blackman, Davide Gerosa, Leo C. Stein, Lawrence E. [46] Jun Luo, Yan-Zheng Bai, Lin Cai, Bin Cao, Wei-Ming
Kidder, and Harald P. Pfeiffer. Surrogate models for Chen, Yu Chen, De-Cong Cheng, Yan-Wei Ding, Hui-
precessing binary black hole simulations with unequal Zong Duan, Xingyu Gou, Chao-Zheng Gu, De-Feng Gu,
masses. Phys. Rev. Res., 1:033015, Oct 2019. Zi-Qi He, Shuang Hu, Yuexin Hu, Xiang-Qing Huang,
[36] Tousif Islam, Vijay Varma, Jackie Lodman, Scott E. Qinghua Jiang, Yuan-Ze Jiang, Hong-Gang Li, Hong-Yin
Field, Gaurav Khanna, Mark A. Scheel, Harald P. Pfeif- Li, Jia Li, Ming Li, Zhu Li, Zhu-Xi Li, Yu-Rong Liang,
fer, Davide Gerosa, and Lawrence E. Kidder. Eccentric Fang-Jie Liao, Yan-Chong Liu, Li Liu, Pei-Bo Liu, Xuhui
binary black hole surrogate models for the gravitational Liu, Yuan Liu, Xiong-Fei Lu, Yan Luo, Jianwei Mei,
waveform and remnant properties: Comparable mass, Min Ming, Shao-Bo Qu, Ding-Yin Tan, Mi Tang, Liang-
nonspinning case. Phys. Rev. D, 103:064022, Mar 2021. Cheng Tu, Cheng-Rui Wang, Fengbin Wang, Guan-Fang
[37] Michael Boyle, Daniel Hemberger, Dante A. B. Iozzo, Wang, Jian Wang, Lijiao Wang, Xudong Wang, Ran Wei,
Geoffrey Lovelace, Serguei Ossokine, Harald P. Pfeif- Shu-Chao Wu, Chun-Yu Xiao, Meng-Zhe Xie, Xiao-Shi
fer, Mark A. Scheel, Leo C. Stein, Charles J. Wood- Xu, Liang Yang, Ming-Lin Yang, Shan-Qing Yang, Hsien-
ford, Aaron B. Zimmerman, Nousha Afshari, Kevin Chi Yeh, Jian-Bo Yu, Lihua Zhang, Meng-Hao Zhao, and
Barkett, Jonathan Blackman, Katerina Chatziioannou, Ze-Bing Zhou. The first round result from the TianQin-1
Tony Chu, Nicholas Demos, Nils Deppe, Scott E. Field, satellite. Classical and Quantum Gravity, 37(18):185013,
Nils L. Fischer, Evan Foley, Heather Fong, Alyssa Gar- aug 2020.
cia, Matthew Giesler, Francois Hebert, Ian Hinder, Reza [47] Alexandre Toubiana, Sylvain Marsat, Stanislav Babak,
Katebi, Haroon Khan, Lawrence E. Kidder, Prayush Ku- John Baker, and Tito Dal Canton. Parameter estimation
mar, Kevin Kuper, Halston Lim, Maria Okounkova, Tere- of stellar-mass black hole binaries with lisa. Phys. Rev.
sita Ramirez, Samuel Rodriguez, Hannes R. Rüter, Pa- D, 102:124037, Dec 2020.
tricia Schmidt, Bela Szilagyi, Saul A. Teukolsky, Vijay [48] Travis Robson, Neil J Cornish, and Chang Liu. The con-
Varma, and Marissa Walker. The SXS collaboration cata- struction and use of LISA sensitivity curves. Classical
log of binary black hole simulations. Classical and Quan- and Quantum Gravity, 36(10):105011, apr 2019.
tum Gravity, 36(19):195006, October 2019. [49] Wen-Hong Ruan, Zong-Kuan Guo, Rong-Gen Cai, and
[38] LVK collaboration. Pycbc software. Yuan-Zhong Zhang. Taiji program: Gravitational-wave
https://pycbc.org/. sources. International Journal of Modern Physics A,
[39] Caltech-Cornell-CITA. binary black hole simulation re- 35(17):2050075, June 2020.
sults. http://www.black-holes.org/waveforms. [50] Neil Cornish and Travis Robson. Galactic binary science
[40] Tony Chu, Heather Fong, Prayush Kumar, Harald P with the new lisa design. Journal of Physics: Conference
Pfeiffer, Michael Boyle, Daniel A Hemberger, Lawrence E Series, 840(1):012024, may 2017.
Kidder, Mark A Scheel, and Bela Szilagyi. On the ac- [51] J. Aasi et al. The NINJA-2 project: detecting and charac-
curacy and precision of numerical waveforms: Effect of terizing gravitational waveforms modelled using numeri-
waveform extraction methodology. Classical and Quan- cal binary black hole simulations. Classical and Quantum
tum Gravity, 33(16):165001, 2016. Gravity, 31(11):115004, June 2014.
[41] DJA McKechan, C Robinson, and Bangalore Surya-
narayana Sathyaprakash. A tapering window for time-