We present a comprehensive study on how well gravitational-wave signals of binary black holes with nonzero eccentricities are recovered by BayesWave, a Bayesian algorithm used by the LIGO-Virgo Collaboration for unmodeled reconstructions of signal waveforms and parameters. We used two different waveform models to produce simulated signals of binary black holes with eccentric orbits and embed them in samples of simulated noise of design-sensitivity Advanced LIGO detectors. We studied the network overlaps and point estimates of central moments of signal waveforms recovered by BayesWave as a function of $e$, the eccentricity of the binary at 8 Hz orbital frequency. BayesWave recovers signals of near-circular ($e\lesssim0.2$) and highly eccentric ($e\gtrsim0.7$) binaries with network overlaps similar to that of circular ($e=0$) ones, however it produces lower network overlaps for binaries with $e\in[0.2,0.7]$. Estimation errors on central frequencies and bandwidths (measured relative to bandwidths) are nearly independent from $e$, while estimation errors on central times and durations (measured relative to durations) increase and decrease with $e$ above $e\gtrsim0.5$, respectively. We also tested how BayesWave performs when reconstructions are carried out using generalized wavelets with linear frequency evolution (chirplets) instead of sine-Gaussian wavelets. We have found that network overlaps improve by $\sim 10-20$ percent when chirplets are used, and the improvement is the highest at low ($e<0.5$) eccentricities. There is however no significant change in the estimation errors of central moments when the chirplet base is used.

中文翻译：

---

来自非零偏心率的双黑洞引力波信号的贝叶斯重建

我们全面研究了贝叶斯波如何很好地恢复具有非零偏心率的双黑洞的引力波信号，贝叶斯波是一种贝叶斯算法，LIGO-Virgo Collaboration 使用它对信号波形和参数进行未建模重建。我们使用了两种不同的波形模型来生成具有偏心轨道的双黑洞的模拟信号，并将它们嵌入到具有设计灵敏度的高级 LIGO 探测器的模拟噪声样本中。我们研究了由 BayesWave 恢复的信号波形中心矩的网络重叠和点估计作为 $e$ 的函数，即二进制在 8 Hz 轨道频率下的偏心率。BayesWave 恢复近圆形 ($e\lesssim0.2$) 和高度偏心 ($e\gtrsim0.7$) 二进制文件的信号，其网络重叠类似于圆形 ($e=0$) 二进制文件，然而，它为 $e\in[0.2,0.7]$ 的二进制文件产生较低的网络重叠。中心频率和带宽的估计误差（相对于带宽测量）几乎与 $e$ 无关，而中心时间和持续时间（相对于持续时间测量）的估计误差随着 $e$ 高于 $e\gtrsim0.5$ 而增加和减少， 分别。我们还测试了当使用具有线性频率演化的广义小波（chirplets）而不是正弦高斯小波进行重建时 BayesWave 的表现。我们发现，当使用 chirplets 时，网络重叠提高了 $\sim 10-20$%，并且在低 ($e<0.5$) 离心率时改进最高。然而，当使用 chirplet 基时，中心矩的估计误差没有显着变化。