New method to search for continuous gravitational waves from unknown neutron stars in binary systems

Nonaxisymmetric, fast-spinning unknown neutron stars in binary systems may emit continuous gravitational waves (nearly monochromatic long-duration signals) which can be detected by ground-based detectors like LIGO and Virgo. In this paper we present a new pipeline, called BinarySkyHough, that can carry out all-sky searches for neutron stars in binary systems by exploiting the usage of graphics processing units. We give a detailed explanation of this new pipeline, and we present simulations which allow us to estimate the sensitivity of the new pipeline, which is approximately twice as sensitive as the best active pipeline with a comparable computational cost.

Figure 1

Gravitational-wave frequency and absolute value of the gravitational-wave first frequency derivative for pulsars with gravitational-wave frequency greater than 10 Hz. Black dots indicate the pulsars which are part of a binary system. Data taken from [15] and downloaded with [16].

Figure 2

Eccentricity of pulsars in binary systems as a function of their rotational frequency. Data taken from [15] and downloaded with [16].

Figure 3

The points show the projected semimajor axis amplitude and orbital period of known pulsars in binary systems. The four lines show different combinations of assumed companion mass and inclination angle for a $1.4M_{\odot }$ neutron star, following Eq. (16). Data taken from [15] and downloaded with [16].

Figure 4

Flowchart showing the steps of the BinarySkyHough algorithm for one search band. Red text shows the steps performed with GPU kernels, and dotted boxes represent the post-processing stages (different vetoes and follow-up procedures are not part of the BinarySkyHough pipeline). Option A shows the steps followed when the data is divided in $N$ different datasets, and option B shows the steps when the data is not divided. The main difference is that option B does not apply the coincidences step in the postprocessing stage.

Figure 5

An example showing the PHMD scheme for the Hough pipelines. The dotted line shows the type of tracks which the BinarySkyHough pipeline uses to combine PHMDs at different timestamps, while the dashed lines show some tracks given by different spin-down/up parameters, which are used by the SkyHough pipeline.

Figure 6

The filled areas mark the regions in binary parameter space where the observed binary modulation is less than one frequency bin, for $T_c=900\mathrm{s}$. Data taken from [15] and downloaded with [16].

Figure 7

Number of binary templates for two searches covering different binary parameter spaces: For “HighP”, $P$ ranges from 10 to 30 days and $a_p$ from 9 to 25 seconds, while for “LowP” $P$ ranges from 0.1 to 0.15 days and $a_p$ from 0.05 to 0.08 seconds. For each frequency, two different mismatch parameters are chosen (shown with circles): 0.1 and 0.4 for 73.6 Hz; 1.6 and 6.4 for 436.9 Hz; 6.4, and 25.6 for 1200 Hz. The coherent time is 900 s, and the observation time is 11178584 s (the duration of the O1 run).

Figure 8

Maximum coherent time allowed by Eq. (38) for four different choices of binary parameters. The black horizontal lines mark 1800 s and 900 s.

Figure 9

Maximum eccentricity allowed by Eq. (43) for three different choices of binary parameters, with a constant choice of $T_c=900\mathrm{s}$. The third and fourth traces overlap.

Figure 10

Efficiency versus sensitivity depth at different parts of the binary parameter space and two different frequencies. All injections have been analyzed with option B ($\mathrm{H}1+\mathrm{L}1$ data), $N_C=0.05$, $m_x=0.4$ and $P_F=1$.

Figure 11

Efficiency versus sensitivity depth for different mismatch configurations. All injections have been analyzed at the binary space 2, with option B ($\mathrm{H}1+\mathrm{L}1$ data) and $N_C=0.05$.

Figure 12

Efficiency versus sensitivity depth for different run configurations. $N_C$ shows the fraction of total templates which are passed to the second step of the search. All injections have been analyzed at the binary space 2, with $m_x=0.4$ and $P_F=1$.

Figure 13

Figures showing the parameter estimation of the new method for simulations in the binary space 2, with option B and $N_C=0.05$. Top left: frequency. Top right: angular frequency $\mathrm{\Omega }$. Bottom left: $a_p$. Bottom right: $t_{\mathrm{asc}}$. The left vertical axis shows the mean absolute error of detected signals in number of bins, while the right vertical axis shows the mean absolute error of detected signals in the correspondent units.

Figure 14

Estimated $h_0$ sensitivity at 95% confidence to random polarized signals with Advanced LIGO O1 data compared to the SkyHough all-sky search for isolated NS for O1 data and to the TwoSpect pipeline.