Harnessing the potential of PyStoch: detecting continuous gravitational waves from interesting supernova remnant targets

TL;DR

The paper presents a hybrid approach for continuous gravitational wave searches that uses the PyStoch GW radiometer as a fast first-pass to identify CW candidates from known supernova remnants, followed by targeted CW pipelines for confirmation. It demonstrates the method on simulated spinning-down NS signals and applies it to O3 data in the 20–1726 Hz band for CasA, Vela Jr., G347, and SN1987A, finding no significant candidates and setting 95% CL upper limits on the strain amplitude, with the CasA limit reaching $1.13 \times 10^{-25}$ at $201.57$ Hz. A key innovation is the frequency-bin combination strategy, which aggregates adjacent bins to accommodate spin-down–induced frequency evolution, coupled with a p-value based significance assessment and Bayesian ULs. The work highlights a fast, computationally efficient screening workflow that can guide follow-up with coherent CW analyses, enabling tighter constraints on neutron star emission models for current and future GW detectors.

Abstract

Detecting continuous gravitational waves (CWs) is challenging due to their weak amplitude and high computational demands, especially with poorly constrained source parameters. Stochastic gravitational-wave background (SGWB) searches using cross-correlation techniques can identify unresolved astrophysical sources, including CWs, at lower computational cost, albeit with reduced sensitivity. This motivates a hybrid approach where SGWB algorithms act as a first-pass filter to identify CW candidates for follow-up with dedicated CW pipelines. We evaluated the discovery potential of the SGWB analysis tool PyStoch for detecting CWs, using simulated signals from spinning down NSs. We then applied the method to data from the third LIGO-Virgo-KAGRA observing run (O3), covering the (20-1726) Hz frequency band, and targeting four supernova remnants: Vela Jr., G347.3-0.5, Cassiopeia A, and the NS associated with the 1987A supernova remnant. If necessary, significant candidates are followed up using the 5-vector Resampling and Band-Sampled Data Frequency-Hough techniques. However, since no interesting candidates were identified in the real O3 analysis, we set 95\% confidence-level upper limits on the CW strain amplitude $h_0$. The most stringent limit was obtained for Cassiopeia A, and is $h_0 = 1.13 \times 10^{-25}$ at $201.57$ Hz with a frequency resolution of $1/32$ Hz. As for the other targets, the best upper limits have been set with the same frequency resolution, and correspond to $h_0 = 1.20 \times 10^{-25} $ at $202.16$ Hz for G347.3-0.5, $1.20 \times 10^{-25}$ at $217.81$ Hz for Vela Jr., and $1.47 \times 10^{-25}$ at $186.41$ Hz for the NS in the 1987A supernova remnant.

Figures (5)

• Figure 1: Flowchart of the PyStoch search for CWs targeting SN1987a, Vela Jr., G347 and CasA. First, PyStoch processes cross-correlated and folded data from LIGO Hanford and Livingston, producing narrow-band SNR maps across frequencies from 20 to 1726 Hz, with a default resolution of 1/32 Hz. Next, using the frequency bin combination strategy [Eq. (\ref{['comb']})], candidates are identified when the following conditions are simultaneously satisfied: SNR $>$ 4.5, i.e. $p$ value $<$ 10% (Sec. \ref{['subsec: uls']}), and a frequency distribution $\Delta f$ consistent with theoretical expectations (Sec. \ref{['sec:model']}). Promising candidates undergo further scrutiny via 5-vec Resampling Francesco) and BSD frequency-Hough methods FH. In the absence of confirmed candidates, 95% confidence-level ULs on the strain amplitude are computed (Sec. \ref{['subsec: uls']}).
• Figure 2: SNR versus frequency for a dataset processed with PyStoch containing simulated Gaussian noise with an injected CW signal. The dataset corresponds to an observation time $T_{\text{obs}} = 14$ months and a frequency resolution of $\delta f_{\text{def}} = 1/32$ Hz. The red triangles represent the dataset without bin combination (i.e. $N=0$ and $\delta f_{\text{comb}} = \delta f_{\text{def}}$), while the blue triangles correspond to the dataset after the bin combination performed with $N=5$, i.e. $\delta f_{\text{comb}} = 11/32$ Hz. The fake CW signal has $h_0 = 2.2 \times 10^{-25}$, $f_0 = 150$ Hz (dashed line), and is spread over 11 default frequency bins ($\Delta f_0 = 11/32$ Hz) due to its spin-down parameters, i.e. $\dot{f}_0 = -10^{-8}$ Hz/s and $\ddot{f}_0 = 10^{-17}$ Hz/s$^2$.
• Figure 3: SNR versus frequency for a dataset processed with PyStoch, containing pure simulated Gaussian noise with $T_{\text{obs}} = 14$ months and frequency resolution of $\delta f_{\text{def}} = 1/32$ Hz. The red triangles represent the dataset without bin combination (i.e. $N=0$ and $\delta f_{\text{comb}} = \delta f_{\text{def}}$), while the blue triangles correspond to the dataset after the bin combination performed with $N=30$, i.e. $\delta f_{\text{comb}} = 61/32$ Hz.
• Figure 4: Best ( $\delta f_{\text{comb}}=$ 1/32 Hz, top) and worst ($\delta f_{\text{comb}}=$ 47/32 Hz, bottom) 95% confidence-level ULs between 20 and 1726 Hz, computed with PyStoch for each target: SN1987a (red), Vela Jr. (green), G347 (orange) and CasA (blue).
• Figure 5: PyStoch best ULs computed targeting CasA (red), frequency-Hough ULs computed targeting the Galactic Center (orange), both between 20 and 1726 Hz, and Resampling ULs computed targeting Scorpius-X1 at selected frequencies for both Livingston (blue triangles) and Hanford (blue circles). We remark that this is an order-of-magnitude comparison of quantities that are not significantly affected by changes in the sky position.