Deep exploration for continuous gravitational waves at 171–172 Hz in LIGO second

observing run data

Karl Wette,1, 2, ∗ Liam Dunn,3, 2 Patrick Clearwater,4, 3, 2 and Andrew Melatos3, 2
1
Centre for Gravitational Astrophysics, Australian National University, Canberra ACT 2601, Australia
2
ARC Centre of Excellence for Gravitational Wave Discovery (OzGrav), Hawthorn VIC 3122, Australia
3
School of Physics, University of Melbourne, Parkville VIC 3010, Australia
4
Gravitational Wave Data Centre, Swinburne University of Technology, Hawthorn VIC 3122, Australia
(Dated: March 21, 2022)
We pursue a novel strategy towards a first detection of continuous gravitational waves from
rapidly-rotating deformed neutron stars. Computational power is focused on a narrow region of
signal parameter space selected by a strategically-chosen benchmark. We search data from the
2nd observing run of the LIGO Observatory with an optimised analysis run on graphics processing
units. While no continuous waves are detected, the search achieves a sensitivity to gravitational
arXiv:2103.12976v2 [gr-qc] 18 Mar 2022

wave strain of h0 = 1.01 × 10−25 at 90% confidence, 24% to 69% better than past searches of the
same parameter space. Constraints on neutron star deformity are within theoretical maxima, thus
a detection by this search was not inconceivable.

I. INTRODUCTION in searches for continuous waves from known pulsars [15–

17]. While the signal parameters are known, the tar-
Neutron stars, the dense remnants of exploded stars, geted pulsars may not radiate detectable continuous
are of particular interest in gravitational wave astron- waves, or else their gravitational-wave and electromag-
omy. Two orbiting neutron stars inevitably collide and netic frequencies may differ [18]. Breadth-first strate-
merge, generating a characteristic chirp signal in gravi- gies, which prioritise parameter-space coverage at the
tational waves detectable by the LIGO [1] and Virgo [2] expense of sensitivity, are used in all-sky surveys for
observatories. The first such detection, in gravitational electromagnetically-quiet neutron stars [19–21]. While
waves [3] and electromagnetic radiation [4], has enlight- the expansive parameter space may contain signals, the
ened neutron stars physics, heavy element production in analysis method may be insufficiently sensitive to detect
the Universe, and cosmology. them. Searches targeting compact remnants of super-
A single rapidly-rotating neutron star may also emit novae [22–25], low-mass X-ray binaries [26–29], and the
gravitational waves, provided it is non-axisymmetric. Galactic Centre [30] where some parameters may be un-
The expected continuous wave signal is long-lived, de- known, adopt intermediate strategies.
creases in frequency over time as rotational energy is ra- In this paper, we pursue the novel combination of an
diated in gravitational waves, and is modulated by the all-sky survey for continuous waves with a depth-first
relative motion between neutron star and detector. The strategy. The search range of gravitational-wave frequen-
neutron star non-axisymmetry might arise from defor- cies is limited to 1 Hz, and a single benchmark is used
mations due to e.g. its magnetic field, accretion of mat- to select all other search parameters. We apply an opti-
ter from a companion star, or normal oscillation modes; mised analysis method to data from the 2nd observing
see [5, 6] for recent reviews. While theoretical and obser- run (O2) of LIGO1 [31], and utilise graphics process-
vational predictions of the non-axisymmetry exist [7–11], ing units (GPUs) to maximise computational efficiency.
typical non-axisymmetries of Galactic neutron stars are While no continuous wave signals are detected, our search
unknown [12, 13]. is the most sensitive yet performed in the O2 data over
Continuous wave signals are expected to be marginally the chosen parameter space, improving by 24–69% over
detectable by contemporary detectors; analysis of year- previous searches.
long datasets, at significant computational cost, is likely
required for a first detection [14]. It is uncertain what
sensitivity is required, and what region of signal param- II. SEMI-COHERENT ANALYSIS
eter space should be explored, in order to maximise de-
tection prospects. Moreover, due to finite computational The gravitational-wave strain of a continuous wave sig-
resources, search sensitivity and parameter-space cover- nal is written as four amplitudes Ai multiplying four os-
age cannot be maximised simultaneously. This mandates cillatory basis functions [32]. The Ai are functions of
the use of strategies which balance these objectives.
Depth-first strategies, which prioritise sensitivity at
the expense of parameter-space coverage, are employed
1 The Virgo detector joined O2 for only the last ∼ 9% of the run
time. Due to the limited data available, which would not have
noticeably improved sensitivity, Virgo O2 data was not used in
∗ karl.wette@anu.edu.au this search.
 2

the strain amplitude h0 , and three angles determining Parameter Value

the neutron star orientation and the initial signal phase. Minimum frequency fmin 171.0 Hz
The basis functions are functions of the phase φ(t). For Maximum frequency fmax 172.0 Hz
an isolated neutron star, φ(t) is a function of: the sky Minimum spin-down f˙min −2.7 × 10−13 Hz s−1
position, given by its right ascension α and declination Maximum spin-down f˙max −3.8 × 10−18 Hz s−1
δ; and the gravitational-wave frequency f , and its first Number of segments N 26
time derivative (or spin-down) f˙, as observed at the So- Segment time-span T 858 194 s (9.9 d)
lar System barycenter at a given reference time. Very Maximum per-segment mismatch µ̃max 0.097
Maximum full-span mismatch µ̂max 0.27
young (. 1000 yr old) neutron stars require higher-order
frequency derivatives, and those with binary companions
TABLE I. Parameters of the search.
require additional orbital parameters. We do not tar-
get these sources, as the consequent increase in compu-
tational burden is not justified by their expected abun-
dances, which are comparable to isolated neutron stars.
The computational cost of continuous wave searches
of large parameter spaces and year-long datasets us-
ing fully phase-coherent matched filtering would be pro-
hibitive. We therefore employ a semi-coherent analysis
method [33–35]. The data are first partitioned into N
segments of time-span T . Within each segment, the data
are filtered against a bank of signal templates. Each
template computes the F-statistic 2F, the fully-coherent
matched filter analytically maximised over Ai [32], which
represents a signal with parameters (α0 , δ 0 , f 0 , f˙0 ). The
template bank is constructed using a parameter-space
metric [36, 37] to ensure any signal is recovered within
a prescribed maximum loss in signal power (relative to FIG. 1. Gravitational-wave frequency f versus spin-down
a perfect match), known as the maximum mismatch and f˙. Dots/crosses represent 2289 pulsars [41] with f˙ < 0.
given by µ̃max . Optimal lattices [38, 39] are used to re- Crosses indicate a subset of 289 pulsars with f > 15 Hz and
duce the overlap between nearby templates, and thereby f˙ & −10−12 Hz s−1 . Gravitational-wave emission at twice the
minimise the template bank and the cost of computing neutron star rotation frequency is assumed. The horizontal
extent of the dashed/dotted lines denotes the range of fmin
2F.
considered during the search setup; at a given fmin , the dashed
Continuous wave signal templates are then constructed area denotes the range of f˙min considered, and the dotted line
spanning the whole dataset, with parameters (α, δ, f, f˙) denotes f˙max .
drawn from a second template bank with maximum mis-
match µ̂max . For each full-span template, and for each
segment, the per-segment template is selected whose more sensitive searches, where h0 is smaller, are pre-
frequency evolution f (t) = d[φ(t)/2π]/dt most closely ferred. In the second factor,
matches the full-span template, as determined by the
parameter-space metric [34]. Then, for each full-span q
template, we compute the detection statistic 2F̂: the σS2 H + σS2 L
σh = , (2)
mean of the N values of 2F corresponding to the N best- SH + SL
match per-segment templates. This technique permits
better sensitivity than a fully-coherent analysis given lim- SH and SL are the noise power spectral densities, har-
ited computational resources [40]. The search setup is monically averaged over a 1-Hz band, of O2 data from
determined by the parameters N , T , µ̃max , and µ̂max . the LIGO detectors at Hanford, WA, and Livingston, LA
respectively; and σSH and σSL are the respective standard
deviations of the 1800 power spectral density bins in the
III. PARAMETER SPACE same 1-Hz band. The exponent of σh was chosen empiri-
cally to favour 1-Hz bands where the power spectrum has
Table I lists the parameters of the search. They max- minimal variation over frequency, and does not contain
imise the benchmark any prominent instrumental artefacts [44]. In the third
factor, ρ is the density of observed pulsars [41] within an
−1/4 Earth-centred sphere with radius d, the distance out to
h−2
0 × σh × ρ1/3 . (1)
which the search is sensitive, given by [45]
In the first factor, h0 is the estimated sensitivity [42, 43] r
of a semi-coherent search of the LIGO O2 data with the 5GIzz
given parameters. The negative exponent denotes that d= h−1
0 , (3)
8c3 τ
 3

where G is the gravitational constant, Izz = 1038 kg m2 quadratic scaling of computational cost with frequency;
is the principal moment of inertia of a typical neutron f˙min from the ranges shown in Figure 1 at a given fmin ; N
star, c is the speed of light, and τ = −f /(4f˙min ) is the from 2 to 134, the maximum number of segments with
characteristic spin-down timescale assuming energy loss T = 2 days in the O2 data; and µ̃max and µ̂max from
only in gravitational waves. As ρ ∝ d−3 ∝ h30 in the worst 0.05 to 1.0. (Fixed parameters are fmax = fmin + 1 Hz,
case, the exponents of h0 and ρ in Eq. (1) are chosen so and f˙max as shown in Figure 1.) In the second mode
that h−20 ρ
1/3
∝ h−1
0 and a smaller h0 is preferred. (chosen for 20% of trials) parameters are sampled from
The use of ρ in Eq. (1) is motivated by the hypothesis ranges 90–110% of the last accepted parameters, in or-
that neutron stars which predominately radiate gravita- der to improve upon them. In the third mode (chosen
tional waves are found in similar regions of the Galaxy, for 10% of trials) only µ̃max and µ̂max are sampled, from
and in overlapping regions of the f –f˙ plane (Figure 1) 0.05 up to the last accepted values of µ̃max and µ̂max , for
as observed pulsars. We may therefore use the observed the purpose of absorbing any remaining computational
density of pulsars as a prior on the possible density of allowance.
gravitational-wave emitting neutron stars. While it may For sampled trial parameters to be acceptable, cer-
be only approximately true, this hypothesis is useful in tain criteria must be satisfied. The estimated sensitiv-
guiding a detection strategy. There is no evidence that ity [42, 43] must improve upon the last accepted trial.
gravitational-wave emitting neutron stars only occupy The estimated computational cost [35] must be within
special regions of the Galaxy. Simulations of Galactic 10–100% of a prescribed budget; for the second and third
neutron stars [12, 13] indicate that, while electromag- modes, the cost must also be greater than the last ac-
netic emission leads to higher spin-down rates and hence cepted trial. At least 20,000 iterations are performed,
lower neutron star rotation frequencies than gravitational until the estimated cost is ≥ 99% of the budget. Through
radiation, nevertheless the two populations overlap in the experimentation we found that the Monte Carlo process
f –f˙ plane. is robust to fine-tuning of the sampling modes. Aside
Assuming gravitational-wave emission at twice the from fmin , the parameters (Table I) do not rail against
neutron star rotation frequency, the number of pulsars their sampling limits.
in Figure 1 with frequencies 166.5–176.5 Hz, bracketing
the search frequency range (Table I), is 0.2 per Hz. The
expected number of Galactic neutron stars (& 108 ; [46]) IV. IMPLEMENTATION
is, however, much larger than the number of observed
pulsars (∼ 3 × 103 ; [41]). An optimistic estimate (ignor- LIGO O2 data, starting at UTC 2016 November 30
ing e.g. selection effects) of the number of neutron stars 17:31:57 (GPS 1164562334), are partitioned into 16 seg-
within the search frequency range is therefore ∼ 7 × 103 . ments, followed by a gap where no usable data are
More pessimistic estimates, which account for the dis- present, followed by a further 10 segments, ending at
tance out to which the search is sensitive, are outlined in UTC 2017 August 25 21:59:34 (GPS 1187733592); all
the Discussion. segments and gaps are of time-span T . The data are
Millisecond pulsars 2 , with f & 100 Hz, are hypoth- further divided into 12626 blocks [48] of 1800 s duration,
esised to have spun up by accretion of matter from a and then Fourier transformed.
companion star; the same emission mechanism could also Computational efficiency was optimised using GPUs.
have built a non-axisymmetric neutron star [7]. A plau- For this search, times to compute the F-statistic [49] and
sible explanation for the maximum observed millisecond the semi-coherent 2F̂ are reduced by factors of ∼ 240
pulsar spin frequency is that spin-up due to accretion is and ∼ 4.2 respectively, relative to non-GPU processors.
balanced by spin-down due to gravitational waves [47]. Computation of 2F dominates the total analysis time.
If true, this suggests that at frequencies where one finds The analysis ran for ∼ 5800 days on the OzSTAR su-
millisecond pulsars which are still undergoing accretion, percomputer using NVIDIA P100 type GPUs; a total of
one might also expect millisecond pulsars where accretion 4.3 × 1016 templates were analysed.
has ceased and which may be spinning down dominantly
though gravitational waves.
Parameters are chosen to maximise Eq. (1) through a V. CANDIDATES
Monte Carlo process. Trial values are drawn according
to three sampling modes. In the first mode (chosen for
70% of trials), parameters are sampled from large initial Figure 2 plots the detection statistic 2F̂ of the top 105
ranges: fmin from 25 to 500 Hz, set by the limited sen- candidates as a function of sky position. Clear outliers
sitivity of LIGO O2 data at low frequencies, and by the are visible with maximum 2F ≈ 14.7, at α separated
by ∼ 12.1h , and at δ ∼ ±66.1◦ . The presence of out-
liers of similar strength at opposing points suggests an
instrumental artefact. Figure 3 plots, as a function of
2 We note that, as shown in Figure 1, the setup of the search did frequency, the single-detector 2F̂ of the top 105 candi-
not consider the most rapidly-spinning millisecond pulsars with dates and the noise power spectrum of the Hanford and
f > 500 Hz, due to computational restrictions. Livingston detectors individually, and the multi-detector
 4

Data Method h0 Sensitivity f˙min / Hz s−1 Ref.

O2 F-statistic 3.40 × 10−25 −1.0 × 10−8 [19]
O2 SkyHough 2.27 × 10−25 −1.0 × 10−8 [19]
O1 PowerFlux 2.10 × 10−25 −1.0 × 10−8 [50]
O2 FrequencyHough 1.72 × 10−25 −1.0 × 10−8 [19]
O2 Einstein@Home 1.33 × 10−25 ∗
−2.6 × 10−9 [20]
O2 This search 1.07 × 10−25 −2.7 × 10−13
O2 This search 1.01 × 10−25 ∗
−2.7 × 10−13

TABLE II. Estimated sensitivity h0 and minimum spin-down

f˙min of this search, and of previous searches of data from the
FIG. 2. Detection statistic 2F̂ of the top 105 candidates 1st (O1) and 2nd (O2) LIGO observing runs which cover the
versus right ascension α and declination δ. Areas with no same parameter space. All sensitivities are at 95% confidence,
candidates are shaded grey. except those asterisked which are at 90% confidence.

2F [51] and harmonically-averaged noise power spectrum

of both detectors. A feature at f ≈ 171.4276 Hz appears
more prominently at Hanford than at Livingston, incon-
sistent with an astrophysical signal but consistent with
an instrumental artefact.
We perform a multi-stage follow-up search of a
small region around the outlier, at negligible cost:
α ∈ ±[5.894h , 6.188h ], δ ∈ ∓[63.517◦ , 68.599◦ ], f ∈
[171.4271, 171.4281] Hz. Each follow-up search stage
halves N , and doubles T , relative to the previous stage
(or to the initial search), in order to improve sensitivity.
The final stage, which searches a single 160-day segment,
yields a top candidate with multi-detector 2F = 95.1
and single-detector 2F = 113.3 and 13.33 for Hanford
FIG. 3. Detection statistic 2F̂ of the top 105 candidates and Livingston respectively. We note that the multi-
(dark-coloured dots, left vertical axis) and the noise power detector 2F is less than one of the single-detector 2F,
spectrum of the O2 data (light-coloured lines, right vertical
violating a long-established consistency criterion [52] for
axis) versus frequency, for all LIGO data (left, black), and for
data from the Hanford (middle, red) and Livingston (right, an astrophysical signal. We conclude that the outlier is
green) detectors. an instrumental artefact, and exclude candidates within
the follow-up parameter space from further analysis.
Figure 4 plots the histogram H(2F̂) of the largest 2F̂,
after removal of the outlier, from 222 partitions of the
parameter space. Each partition is composed of many
disjoint patches evenly distributed across the sky; we ex-
pect the largest 2F̂ from the partitions to be statistically
independent. We fit H(2F̂) with a smooth probabil-
ity density p(2F̂) using Gaussian kernel density estima-
tion [53, 54]. As no obvious outliers remain, we consider
p(2F̂) to well-approximate the distribution of 2F in each
partition in the absence of a signal. It follows from ex-
treme value statistics that the probability density P (2F̂)
of the largest 2F̂ expected from the entire search, in the
absence of a signal, is [55]
"Z #222
2F̂
FIG. 4. Expected distribution of the largest detection statis- d 0 0
P (2F̂) = d(2F̂ )p(2F̂ ) . (4)
tic 2F̂. Black solid line: histogram H(2F̂) of the largest 2F̂, d(2F̂) 0
after removal of the outlier, from 222 parameter-space par-
titions. Black dashed line: probability density p(2F̂) fit to
H(2F̂). Black dotted line: probability density P (2F̂) of the The largest found 2F̂ = 11.1, after removal of the out-
largest 2F̂ from the entire search. Vertical red line: largest
lier, is indicated in Figure 4 and is consistent with the
candidate 2F̂. expected distribution P (2F̂). We conclude that no con-
tinuous wave signals are detected.
 5

VI. SENSITIVITY in the naive estimate, or ∼ 0.2–13 in the pessimistic es-

timate. Given the many assumptions and simplicity of
Table II lists the estimated sensitivity h0 = the above calculations, at best this suggests substantial
1.01 × 10−25 achieved by the search. This is a statis- uncertainty in the number of electromagnetically-quiet
tical statement at 90% confidence: were a large number neutron stars this search is sensitive to.
of continuous wave signals present within the parameter Neutron star non-axisymmetry is characterised by the
space, with the given h0 and other amplitude parameters equatorial ellipticity . The minimum  to which this
chosen at random, we would have detected 90% of them. search is sensitive at 90% confidence and d = 320 pc [62],
This statement is validated by performing 500 searches of
c4 h0 d
small regions of the parameter space, each containing a = = 1.04 × 10−6 , (5)
simulated signal as described above, and confirming the 4π 2 GIzz f 2
expected 90% detection rate. For comparison with pre- is within conservative maximum values attainable by the-
vious searches, we have also estimated the sensitivity at oretical models [9]. A detection of continuous waves by
95% confidence, which gives h0 = 1.07 × 10−25 . this search was certainly possible, therefore, based on
current knowledge of neutron star physics.
The sensitivity achieved by this search confirms the ad-
VII. DISCUSSION vantages of a depth-first strategy for all-sky continuous
wave surveys. Such a strategy, in concert with comple-
Table II lists the sensitivity h0 achieved by previous mentary breadth-first surveys of wide parameter spaces,
searches for continuous waves; they covered parameter should continue to be pursued. There is ample scope to
spaces including, and significantly larger than, the pa- refine the benchmark of Eq. (1), perhaps by including
rameter space of this search. Our search improves in a more informed distribution of Galactic neutron stars
sensitivity by 24–69% over previous searches, and is the spinning down through electromagnetic and gravitational
most sensitive exploration of this parameter space yet waves. An insightful choice of benchmark could be piv-
performed. otal to a first detection of continuous gravitational waves.
While optimisation techniques are used to select the
most sensitive continuous wave search, when the pa-
rameter space is specified a priori [40, 56–58], optimi- ACKNOWLEDGMENTS
sation of the parameter space and search setup simul-
taneously is uncommon [59]. Our approach realises a We thank Hannah Middleton for helpful comments on
simple astrophysically-motivated benchmark [Eq. (1)] as the manuscript. This research was supported by the Aus-
optimised choices for both parameter space and search tralian Research Council Centre of Excellence for Gravi-
setup. It makes explicit the prior assumptions used to tational Wave Discovery (OzGrav) through project num-
construct the parameter space, which then permits those ber CE170100004. It used data, software and/or web
assumptions to be refined by improved understanding of tools obtained from the Gravitational Wave Open Science
neutron star physics and Galactic neutron star popula- Center (https://www.gw-openscience.org/), a service
tions. of LIGO Laboratory, the LIGO Scientific Collaboration
The minimum spin-down f˙min (Table I) is smaller and the Virgo Collaboration. LIGO Laboratory and
than used in the previous searches listed in Table II. Advanced LIGO are funded by the United States Na-
While [60, 61] also focus on small spin-downs, our choice tional Science Foundation (NSF) as well as the Science
of f˙min was not made a priori, but arose as a conse- and Technology Facilities Council (STFC) of the United
quence of Eq. (1). The chosen f˙min and Eq. (3) imply Kingdom, the Max-Planck-Society (MPS), and the State
that the search is sensitive to gravitationally-radiating of Niedersachsen/Germany for support of the construc-
neutron stars within a distance d = 320 pc. Of the tion of Advanced LIGO and construction and operation
289 pulsars indicated in Figure 1, only 10 are within of the GEO600 detector. Additional support for Ad-
this distance. Nevertheless, only a small fraction (∼ vanced LIGO was provided by the Australian Research
3 × 103 /108 ) of Galactic neutron stars are observed as Council. Virgo is funded, through the European Gravi-
pulsars, and therefore one might naively expect ∼ 3 × 105 tational Observatory (EGO), by the French Centre Na-
electromagnetically-quiet neutron stars within this dis- tional de Recherche Scientifique (CNRS), the Italian Is-
tance. A more pessimistic count may be derived starting tituto Nazionale della Fisica Nucleare (INFN) and the
from the modelled volume density of neutron stars in Dutch Nikhef, with contributions by institutions from
the solar neighbourhood [46] of ∼ 1–5×10−4 pc−3 ; this Belgium, Germany, Greece, Hungary, Ireland, Japan,
suggests only ∼ 3 × 103 –2 × 105 neutron stars within a Monaco, Poland, Portugal, Spain. It used the soft-
volume with radius d = 320 pc. Taken together with ware packages LALSuite [63], Octave [64], OctApps [65],
the estimated number of pulsars within the searched fre- Python [66], NumPy [67], Matplotlib [68], and SciPy [69].
quency band as a fraction of the number of observed pul- The search was performed on the OzSTAR national facil-
sars (∼ 0.2/3 × 103 ), the number of neutron stars within ity at Swinburne University of Technology. The OzSTAR
the searched band and sensitive volume might be ∼ 20 program receives funding in part from the Astronomy
 6

National Collaborative Research Infrastructure Strategy (NCRIS) allocation provided by the Australian Govern-
ment. Document number LIGO-P2000536.

[1] J. Aasi et al. (LIGO Scientific Collaboration), Advanced tional Waves from Known Pulsars at Two Harmonics
LIGO, Classical and Quantum Gravity 32, 074001 in 2015-2017 LIGO Data, Astrophysical Journal 879, 10
(2015), arXiv:1411.4547 [gr-qc]. (2019), arXiv:1902.08507 [astro-ph.HE].
[2] F. Acernese et al., Status of Advanced Virgo, in European [16] B. P. Abbott et al. (LIGO Scientific Collaboration and
Physical Journal Web of Conferences, Vol. 182 (2018) p. Virgo Collaboration and others), Narrow-band search for
02003. gravitational waves from known pulsars using the sec-
[3] B. P. Abbott et al. (LIGO Scientific Collaboration ond LIGO observing run, Physical Review D 99, 122002
and Virgo Collaboration), GW170817: Observation of (2019).
Gravitational Waves from a Binary Neutron Star In- [17] R. Abbott et al. (LIGO Scientific Collaboration and
spiral, Physical Review Letters 119, 161101 (2017), Virgo Collaboration and others), Gravitational-wave
arXiv:1710.05832 [gr-qc]. Constraints on the Equatorial Ellipticity of Millisecond
[4] B. P. Abbott et al. (LIGO Scientific Collaboration and Pulsars, Astrophysical Journal 902, L21 (2020).
Virgo Collaboration and others), Multi-messenger Ob- [18] D. I. Jones and N. Andersson, Gravitational waves from
servations of a Binary Neutron Star Merger, Astrophys- freely precessing neutron stars, Monthly Notices of the
ical Journal 848, L12 (2017), arXiv:1710.05833 [astro- Royal Astronomical Society 331, 203 (2002), arXiv:gr-
ph.HE]. qc/0106094.
[5] K. Riles, Recent searches for continuous gravitational [19] B. P. Abbott et al. (LIGO Scientific Collaboration and
waves, Modern Physics Letters A 32, 1730035-685 Virgo Collaboration and others), All-sky search for con-
(2017), arXiv:1712.05897 [gr-qc]. tinuous gravitational waves from isolated neutron stars
[6] M. Sieniawska and M. Bejger, Continuous Gravita- using Advanced LIGO O2 data, Physical Review D 100,
tional Waves from Neutron Stars: Current Status and 024004 (2019), arXiv:1903.01901 [astro-ph.HE].
Prospects, Universe 5, 217 (2019), arXiv:1909.12600 [20] B. Steltner, M. A. Papa, H.-B. Eggenstein, B. Allen,
[astro-ph.HE]. V. Dergachev, R. Prix, B. Machenschalk, S. Walsh, S. J.
[7] A. Melatos and D. J. B. Payne, Gravitational Radia- Zhu, O. Behnke, and S. Kwang, Einstein@Home All-
tion from an Accreting Millisecond Pulsar with a Mag- sky Search for Continuous Gravitational Waves in LIGO
netically Confined Mountain, Astrophysical Journal 623, O2 Public Data, Astrophysical Journal 909, 79 (2021),
1044 (2005), arXiv:astro-ph/0503287. arXiv:2009.12260 [astro-ph.HE].
[8] A. Mastrano and A. Melatos, Updated gravitational- [21] P. B. Covas and A. M. Sintes, First All-Sky Search
wave upper limits on the internal magnetic field strength for Continuous Gravitational-Wave Signals from Un-
of recycled pulsars, Monthly Notices of the Royal Astro- known Neutron Stars in Binary Systems Using Advanced
nomical Society 421, 760 (2012), arXiv:1112.1542 [astro- LIGO Data, Physical Review Letters 124, 191102 (2020),
ph.HE]. arXiv:2001.08411 [gr-qc].
[9] N. K. Johnson-McDaniel and B. J. Owen, Maximum elas- [22] B. P. Abbott et al. (LIGO Scientific Collaboration and
tic deformations of relativistic stars, Physical Review D Virgo Collaboration), Searches for Continuous Gravita-
88, 044004 (2013), arXiv:1208.5227 [astro-ph.SR]. tional Waves from 15 Supernova Remnants and Fomal-
[10] G. Woan, M. D. Pitkin, B. Haskell, D. I. Jones, and haut b with Advanced LIGO, Astrophysical Journal 875,
P. D. Lasky, Evidence for a Minimum Ellipticity in Mil- 122 (2019).
lisecond Pulsars, Astrophysical Journal 863, L40 (2018), [23] L. Lindblom and B. J. Owen, Directed searches for
arXiv:1806.02822 [astro-ph.HE]. continuous gravitational waves from twelve supernova
[11] F. Gittins, N. Andersson, and D. I. Jones, Modelling neu- remnants in data from Advanced LIGO’s second ob-
tron star mountains, Monthly Notices of the Royal As- serving run, Physical Review D 101, 083023 (2020),
tronomical Society 500, 5570 (2021), arXiv:2009.12794 arXiv:2003.00072 [gr-qc].
[astro-ph.HE]. [24] M. Millhouse, L. Strang, and A. Melatos, Search for
[12] C. Palomba, Simulation of a population of isolated neu- gravitational waves from 12 young supernova remnants
tron stars evolving through the emission of gravitational with a hidden Markov model in Advanced LIGO’s second
waves, Monthly Notices of the Royal Astronomical Soci- observing run, Physical Review D 102, 083025 (2020),
ety 359, 1150 (2005), arXiv:astro-ph/0503046. arXiv:2003.08588 [gr-qc].
[13] L. Wade, X. Siemens, D. L. Kaplan, B. Knispel, and [25] M. A. Papa, J. Ming, E. V. Gotthelf, B. Allen, R. Prix,
B. Allen, Continuous gravitational waves from isolated V. Dergachev, H.-B. Eggenstein, A. Singh, and S. J.
Galactic neutron stars in the advanced detector era, Zhu, Search for Continuous Gravitational Waves from the
Physical Review D 86, 124011 (2012), arXiv:1209.2971 Central Compact Objects in Supernova Remnants Cas-
[gr-qc]. siopeia A, Vela Jr., and G347.3-0.5, Astrophysical Jour-
[14] P. R. Brady, T. Creighton, C. Cutler, and B. F. Schutz, nal 897, 22 (2020), arXiv:2005.06544 [astro-ph.HE].
Searching for periodic sources with LIGO, Physical Re- [26] B. P. Abbott et al. (LIGO Scientific Collaboration and
view D 57, 2101 (1998), arXiv:gr-qc/9702050. Virgo Collaboration and others), Upper Limits on Grav-
[15] B. P. Abbott et al. (LIGO Scientific Collaboration and itational Waves from Scorpius X-1 from a Model-based
Virgo Collaboration and others), Searches for Gravita- Cross-correlation Search in Advanced LIGO Data, Astro-
 7

physical Journal 847, 47 (2017). Review D 85, 042003 (2012), arXiv:1111.5650 [gr-qc].
[27] B. P. Abbott et al. (LIGO Scientific Collaboration and [43] C. Dreissigacker, R. Prix, and K. Wette, Fast and accu-
Virgo Collaboration and others), Search for gravitational rate sensitivity estimation for continuous-gravitational-
waves from Scorpius X-1 in the second Advanced LIGO wave searches, Physical Review D 98, 084058 (2018).
observing run with an improved hidden Markov model, [44] P. B. Covas, A. Effler, E. Goetz, P. M. Meyers, A. Neun-
Physical Review D 100, 122002 (2019). zert, M. Oliver, B. L. Pearlstone, V. J. Roma, R. M. S.
[28] H. Middleton, P. Clearwater, A. Melatos, and L. Dunn, Schofield, V. B. Adya, et al., Identification and mitiga-
Search for gravitational waves from five low mass x-ray tion of narrow spectral artifacts that degrade searches for
binaries in the second Advanced LIGO observing run persistent gravitational waves in the first two observing
with an improved hidden Markov model, Physical Review runs of Advanced LIGO, Physical Review D 97, 082002
D 102, 023006 (2020), arXiv:2006.06907 [astro-ph.HE]. (2018), arXiv:1801.07204 [astro-ph.IM].
[29] Y. Zhang, M. A. Papa, B. Krishnan, and A. L. Watts, [45] K. Wette et al., Searching for gravitational waves from
Search for Continuous Gravitational Waves from Scor- Cassiopeia A with LIGO, Classical and Quantum Gravity
pius X-1 in LIGO O2 Data, Astrophysical Journal Let- 25, 235011 (2008), arXiv:0802.3332 [gr-qc].
ters 906, L14 (2021), arXiv:2011.04414 [astro-ph.HE]. [46] N. Sartore, E. Ripamonti, A. Treves, and R. Turolla,
[30] O. J. Piccinni, P. Astone, S. D’Antonio, S. Frasca, Galactic neutron stars. I. Space and velocity distributions
G. Intini, I. La Rosa, P. Leaci, S. Mastrogiovanni, in the disk and in the halo, Astronomy and Astrophysics
A. Miller, and C. Palomba, Directed search for contin- 510, A23 (2010), arXiv:0908.3182 [astro-ph.GA].
uous gravitational-wave signals from the Galactic Center [47] L. Bildsten, Gravitational Radiation and Rotation of Ac-
in the Advanced LIGO second observing run, Physical creting Neutron Stars, Astrophysical Journal 501, L89
Review D 101, 082004 (2020), arXiv:1910.05097 [gr-qc]. (1998), arXiv:astro-ph/9804325.
[31] R. Abbott et al. (LIGO Scientific Collaboration and [48] E. Goetz (LIGO Scientific Collaboration), Segments used
Virgo Collaboration), Open data from the first and sec- for creating standard SFTs in O2 data, Tech. Rep.
ond observing runs of Advanced LIGO and Advanced T1900085-v1 (LIGO, 2019).
Virgo, SoftwareX 13, 100658 (2021). [49] L. Dunn et al., in preparation (2021).
[32] P. Jaranowski, A. Królak, and B. F. Schutz, Data anal- [50] V. Dergachev and M. A. Papa, Sensitivity Improvements
ysis of gravitational-wave signals from spinning neutron in the Search for Periodic Gravitational Waves Using O1
stars: The signal and its detection, Physical Review D LIGO Data, Physical Review Letters 123, 101101 (2019),
58, 063001 (1998), arXiv:gr-qc/9804014. arXiv:1902.05530 [gr-qc].
[33] P. R. Brady and T. Creighton, Searching for periodic [51] C. Cutler and B. F. Schutz, Generalized F-statistic:
sources with LIGO. II. Hierarchical searches, Physical Multiple detectors and multiple gravitational wave pul-
Review D 61, 082001 (2000), arXiv:gr-qc/9812014. sars, Physical Review D 72, 063006 (2005), arXiv:gr-
[34] K. Wette, Parameter-space metric for all-sky semicoher- qc/0504011.
ent searches for gravitational-wave pulsars, Physical Re- [52] B. Abbott et al. (LIGO Scientific Collaboration), Ein-
view D 92, 082003 (2015), arXiv:1508.02372 [gr-qc]. stein@Home search for periodic gravitational waves in
[35] K. Wette, S. Walsh, R. Prix, and M. A. Papa, Imple- LIGO S4 data, Physical Review D 79, 022001 (2009),
menting a semicoherent search for continuous gravita- arXiv:0804.1747 [gr-qc].
tional waves using optimally constructed template banks, [53] E. Parzen, On Estimation of a Probability Density Func-
Physical Review D 97, 123016 (2018), arXiv:1804.03392 tion and Mode, Annals of Mathematical Statistics 33,
[astro-ph.IM]. 1065 (1962).
[36] K. Wette and R. Prix, Flat parameter-space metric for [54] M. Rosenblatt, Remarks on Some Nonparametric Esti-
all-sky searches for gravitational-wave pulsars, Physical mates of a Density Function, Annals of Mathematical
Review D 88, 123005 (2013), arXiv:1310.5587 [gr-qc]. Statistics 27, 832 (1956).
[37] K. Wette, Empirically extending the range of va- [55] J. Abadie et al. (LIGO Scientific Collaboration), First
lidity of parameter-space metrics for all-sky searches Search for Gravitational Waves from the Youngest
for gravitational-wave pulsars, Physical Review D 94, Known Neutron Star, Astrophysical Journal 722, 1504
122002 (2016), arXiv:1607.00241 [gr-qc]. (2010), arXiv:1006.2535 [gr-qc].
[38] R. Prix, Template-based searches for gravitational [56] C. Messenger, H. J. Bulten, S. G. Crowder, V. Dergachev,
waves: efficient lattice covering of flat parameter D. K. Galloway, E. Goetz, R. J. G. Jonker, P. D. Lasky,
spaces, Classical and Quantum Gravity 24, S481 (2007), G. D. Meadors, A. Melatos, S. Premachandra, K. Riles,
arXiv:0707.0428 [gr-qc]. L. Sammut, E. H. Thrane, J. T. Whelan, and Y. Zhang,
[39] K. Wette, Lattice template placement for coherent all-sky Gravitational waves from Scorpius X-1: A comparison
searches for gravitational-wave pulsars, Physical Review of search methods and prospects for detection with ad-
D 90, 122010 (2014), arXiv:1410.6882 [gr-qc]. vanced detectors, Physical Review D 92, 023006 (2015),
[40] R. Prix and M. Shaltev, Search for continuous grav- arXiv:1504.05889 [gr-qc].
itational waves: Optimal StackSlide method at fixed [57] S. Walsh, M. Pitkin, M. Oliver, S. D’Antonio, V. Der-
computing cost, Physical Review D 85, 084010 (2012), gachev, A. Królak, P. Astone, M. Bejger, M. Di Gio-
arXiv:1201.4321 [gr-qc]. vanni, O. Dorosh, S. Frasca, P. Leaci, S. Mastrogio-
[41] R. N. Manchester, G. B. Hobbs, A. Teoh, and vanni, A. Miller, C. Palomba, M. A. Papa, O. J. Pic-
M. Hobbs, The Australia Telescope National Facility Pul- cinni, K. Riles, O. Sauter, and A. M. Sintes, Comparison
sar Catalogue, Astronomical Journal 129, 1993 (2005), of methods for the detection of gravitational waves from
arXiv:astro-ph/0412641 [astro-ph]. unknown neutron stars, Physical Review D 94, 124010
[42] K. Wette, Estimating the sensitivity of wide-parameter- (2016), arXiv:1606.00660 [gr-qc].
space searches for gravitational-wave pulsars, Physical
 8

[58] S. Walsh, K. Wette, M. A. Papa, and R. Prix, Opti- [64] J. W. Eaton, D. Bateman, S. Hauberg, and R. Wehbring,
mizing the choice of analysis method for all-sky searches GNU Octave manual: a high-level interactive language
for continuous gravitational waves with Einstein@Home, for numerical computations (2020).
Physical Review D 99, 082004 (2019). [65] K. Wette, R. Prix, D. Keitel, M. Pitkin, C. Dreissigacker,
[59] J. Ming, B. Krishnan, M. A. Papa, C. Aulbert, and J. T. Whelan, and P. Leaci, OctApps: a library of Octave
H. Fehrmann, Optimal directed searches for continu- functions for continuous gravitational-wave data analy-
ous gravitational waves, Physical Review D 93, 064011 sis, Journal of Open Source Software 3, 707 (2018).
(2016), arXiv:1510.03417 [gr-qc]. [66] Python Software Foundation, Python Language Refer-
[60] V. Dergachev and M. A. Papa, Results from the First ence (2020).
All-Sky Search for Continuous Gravitational Waves from [67] C. R. Harris, K. J. Millman, S. J. van der Walt, R. Gom-
Small-Ellipticity Sources, Physical Review Letters 125, mers, P. Virtanen, D. Cournapeau, E. Wieser, J. Taylor,
171101 (2020), arXiv:2004.08334 [gr-qc]. S. Berg, N. J. Smith, et al., Array programming with
[61] V. Dergachev and M. A. Papa, Results from high- NumPy, Nature 585, 357–362 (2020).
frequency all-sky search for continuous gravita- [68] J. D. Hunter, Matplotlib: A 2D graphics environment,
tional waves from small-ellipticity sources (2020), Computing in Science & Engineering 9, 90 (2007).
arXiv:2012.04232 [gr-qc]. [69] P. Virtanen, R. Gommers, T. E. Oliphant, M. Haber-
[62] S. Bonazzola and E. Gourgoulhon, Gravitational waves land, T. Reddy, D. Cournapeau, E. Burovski, P. Peter-
from pulsars: emission by the magnetic field induced son, W. Weckesser, J. Bright, et al., SciPy 1.0: Funda-
distortion, Astronomy & Astrophysics 312, 675 (1996), mental Algorithms for Scientific Computing in Python,
arXiv:astro-ph/9602107. Nature Methods 17, 261 (2020).
[63] LIGO Scientific Collaboration, LIGO Algorithm Library
- LALSuite, Free software (GPL) (2018).