引力波天文系列文章将介绍引力波能观测到什么,如何观测,以及基本的数据分析方法。侧重于引力波的空间探测和分析,但核心的测量原理、仪器响应以及数据分析方法基本都是通用的。
本文介绍引力波数据的分析方法。
数据分析基础
上面推导的频域响应是严格的,但只针对特定的时刻:
s ( t ) ∣ t o r b i t = F − 1 { s ~ ( f ) ∣ t o r b i t } s(t)|_{t_{\rm orbit}} = \mathcal{F}^{-1}\{ \tilde{s}(f)|_{t_{\rm orbit}}\}
s ( t ) ∣ t o r b i t = F − 1 { s ~ ( f ) ∣ t o r b i t }
只有t = t o r b i t t=t_{\rm orbit} t = t o r b i t 的值才对应实际观测。随着t o r b i t t_{\rm orbit} t o r b i t 改变,探测器位置改变,s ~ ( f ) \tilde{s}(f) s ~ ( f ) 改变。
实际观测数据s ˉ ( t o r b i t ) \bar{s}(t_{\rm orbit}) s ˉ ( t o r b i t ) 是一系列不同位置处s ( t ) s(t) s ( t ) 在对应时刻取值所组成的序列,对它做傅里叶变换得到的是?怎么得到前面推导的s ~ ( f ) \tilde{s}(f) s ~ ( f ) ?
注意:对于双星绕转或并和等系统,s ( t ) s(t) s ( t ) 是有傅里叶变换的。但探测器噪声n ( t ) n(t) n ( t ) 以及随机背景信号s ( t ) s(t) s ( t ) 都是随机过程,属于能量信号,严格讲是没有傅里叶变换的(积分发散),只有功率谱,这也是为什么会引入特征应变h c ( f ) h_c(f) h c ( f ) 。这里只是个记号,不求严谨。
d ( t ) = s ( t ) + n ( t ) ; s ( t ) ≡ F P ( t ) ∗ h P ( t ) d ~ ( f ) = s ~ ( f ) + n ~ ( f ) ; s ~ ( f ) = F P ( f ) h ~ P ( f ) \begin{aligned}
d(t) = s(t) + n(t); ~~~ s(t) \equiv F^P(t) * h_P(t)\\
\tilde{d}(f) = \tilde{s}(f) + \tilde{n}(f); ~~~ \tilde{s}(f) = F^P(f) \tilde{h}_P(f)
\end{aligned} d ( t ) = s ( t ) + n ( t ) ; s ( t ) ≡ F P ( t ) ∗ h P ( t ) d ~ ( f ) = s ~ ( f ) + n ~ ( f ) ; s ~ ( f ) = F P ( f ) h ~ P ( f )
信噪比
d ( t ) = F P ( t ) ∗ h P ( t ) + n ( t ) d ~ ( f ) = F P ( f ) h ~ P ( f ) + n ~ ( f ) \begin{aligned}
d(t) = F^P(t) * h_P(t) + n(t)\\
\tilde{d}(f) = F^P(f) \tilde{h}_P(f) + \tilde{n}(f)
\end{aligned} d ( t ) = F P ( t ) ∗ h P ( t ) + n ( t ) d ~ ( f ) = F P ( f ) h ~ P ( f ) + n ~ ( f )
在描述引力波信号时,不同偏振模式不能简单合并,有各自不同的转移函数,而且一般地F P F^P F P 是复值函数。但不同引力波信号的功率(能量)可以直接相加:
S h ( f ) = ∑ P ∣ h ~ P ( f ) ∣ 2 ↔ S h ( f ) = 1 2 ∑ P ∣ h ~ P ( f ) ∣ 2 \begin{aligned}
S_h(f) = \sum_P |\tilde{h}_P(f)|^2 \leftrightarrow S_h(f) = \frac{1}{2}\sum_P |\tilde{h}_P(f)|^2
\end{aligned} S h ( f ) = P ∑ ∣ h ~ P ( f ) ∣ 2 ↔ S h ( f ) = 2 1 P ∑ ∣ h ~ P ( f ) ∣ 2
对各态历经的平稳随机过程,有两种等价的自(互)相关 :
一种是系综平均:E { d I ( t ) d J ( t + τ ) } ⟨ d I ( t ) d J ( t + τ ) ⟩ E\{ d_I(t)d_J(t+\tau) \} ~~~ \big\langle d_I(t)d_J(t+\tau)\big\rangle E { d I ( t ) d J ( t + τ ) } ⟨ d I ( t ) d J ( t + τ ) ⟩
一种是时间平均:lim T → ∞ 1 T ∫ − T / 2 T / 2 d I ( t ) d J ( t + τ ) d t d I ( t ) ⋆ d J ( t + τ ) \lim_{T\rightarrow\infty} \frac{1}{T}\int_{-T/2}^{T/2} d_I(t)d_J(t+\tau) dt ~~~~ d_I(t) \star d_J(t+\tau) lim T → ∞ T 1 ∫ − T / 2 T / 2 d I ( t ) d J ( t + τ ) d t d I ( t ) ⋆ d J ( t + τ )
相比函数的互相关∫ − ∞ ∞ f ( t ) g ( t + τ ) d t \int_{-\infty}^{\infty} f(t) g(t+\tau) dt ∫ − ∞ ∞ f ( t ) g ( t + τ ) d t ,多了对时间的平均??为什么
不同于一般互相关定义,这里互相关是包含时间平均的,具体参考Creighton & Anderson. 2011.
⟨ x ( t ) x ( t + τ ) ⟩ = r x ( τ ) , ⟨ x ~ ( f ) x ~ ∗ ( f ′ ) ⟩ = S ( f ) δ ( f − f ′ ) \langle x(t)x(t+\tau)\rangle = r^x(\tau), ~~~~ \langle \tilde{x}(f)\tilde{x}^*(f')\rangle = S(f)\delta(f-f')
⟨ x ( t ) x ( t + τ ) ⟩ = r x ( τ ) , ⟨ x ~ ( f ) x ~ ∗ ( f ′ ) ⟩ = S ( f ) δ ( f − f ′ )
⟨ x 2 ( t ) ⟩ = r x ( 0 ) = ∫ S ( f ) d f , ⟨ ∣ x ~ ( f ) ∣ 2 ⟩ = S ( f ) δ ( 0 ) \langle x^2(t)\rangle = r^x(0) = \int S(f)df, ~~~~ \langle |\tilde{x}(f)|^2\rangle = S(f)\delta(0)
⟨ x 2 ( t ) ⟩ = r x ( 0 ) = ∫ S ( f ) d f , ⟨ ∣ x ~ ( f ) ∣ 2 ⟩ = S ( f ) δ ( 0 )
通过探测信号自相关或互相关计算(互)功率谱,与
自相关r d ( t 1 , t 2 ) = E { d ( t 1 ) d ( t 2 ) } ; r s ( t 1 , t 2 ) = E { s ( t 1 ) s ( t 2 ) } r^d(t_1, t_2) = E\{ d(t_1) d(t_2) \}; ~~ r^s(t_1, t_2) = E\{ s(t_1) s(t_2) \} r d ( t 1 , t 2 ) = E { d ( t 1 ) d ( t 2 ) } ; r s ( t 1 , t 2 ) = E { s ( t 1 ) s ( t 2 ) }
互相关r I J d ( t 1 , t 2 ) = E { d I ( t 1 ) d J ( t 2 ) } ; r I J s ( t 1 , t 2 ) = E { s I ( t 1 ) s J ( t 2 ) } r^d_{IJ}(t_1, t_2) = E\{ d_I(t_1) d_J(t_2) \}; ~~ r^s_{IJ}(t_1, t_2) = E\{ s_I(t_1) s_J(t_2) \} r I J d ( t 1 , t 2 ) = E { d I ( t 1 ) d J ( t 2 ) } ; r I J s ( t 1 , t 2 ) = E { s I ( t 1 ) s J ( t 2 ) }
其中期望对应系综平均
对平稳随机过程自相关函数只与时间差有关,不随具体计算的时间点变化。
对各态历经过程,系综平均等于时间平均
自相关r s ( t ) = r s ( t 1 − t 2 ) = ⟨ s ( t ) s ( t ) ⟩ , S ( f ) = s ~ ( f ) s ~ ∗ ( f ) = ∣ s ~ ( f ) ∣ 2 r^s(t) = r^s(t_1-t_2) = \left\langle s(t) s(t) \right\rangle,~~~ S(f) = \tilde{s}(f)\tilde{s}^*(f) = |\tilde{s}(f)|^2 r s ( t ) = r s ( t 1 − t 2 ) = ⟨ s ( t ) s ( t ) ⟩ , S ( f ) = s ~ ( f ) s ~ ∗ ( f ) = ∣ s ~ ( f ) ∣ 2
互相关r I J s ( t ) = r I J s ( t 1 − t 2 ) = ⟨ s I ( t ) ⋆ s J ( t ) ⟩ , S I J ( f ) = s ~ I ( f ) s ~ J ∗ ( f ) r^s_{IJ}(t) = r^s_{IJ}(t_1 - t_2) = \left\langle s_I(t) \star s_J(t) \right\rangle,~~~ S_{IJ}(f) = \tilde{s}_I(f)\tilde{s}_J^*(f) r I J s ( t ) = r I J s ( t 1 − t 2 ) = ⟨ s I ( t ) ⋆ s J ( t ) ⟩ , S I J ( f ) = s ~ I ( f ) s ~ J ∗ ( f )
r s ( t ) = ∫ S ( f ) e i 2 π f t d f , ⟨ s 2 ( t ) ⟩ = r s ( 0 ) = ∫ N ( f ) d f r^s(t) = \int S(f) e^{i 2 \pi f t} df,~~~ \left\langle s^2(t) \right\rangle = r^s(0) = \int N(f) df
r s ( t ) = ∫ S ( f ) e i 2 π f t d f , ⟨ s 2 ( t ) ⟩ = r s ( 0 ) = ∫ N ( f ) d f
⟨ s ( t ) ⋆ s ( t ) ⟩ ⇆ ⟨ s ~ ( f ) s ~ ∗ ( f ) ⟩ ⇆ S ( f ) \left\langle s(t) \star s(t) \right\rangle \leftrightarrows \left\langle \tilde{s}(f)\tilde{s}^*(f) \right\rangle \leftrightarrows S(f)
⟨ s ( t ) ⋆ s ( t ) ⟩ ⇆ ⟨ s ~ ( f ) s ~ ∗ ( f ) ⟩ ⇆ S ( f )
r I J s ( t ) = ∫ S I J ( f ) e i 2 π f t d f , ⟨ s I ( t ) s J ( t ) ⟩ = r I J s ( 0 ) = ∫ S I J ( f ) d f r^s_{IJ}(t) = \int S_{IJ}(f) e^{i 2 \pi f t} df,~~~ \left\langle s_I(t)s_J(t) \right\rangle = r^s_{IJ}(0) = \int S_{IJ}(f) df
r I J s ( t ) = ∫ S I J ( f ) e i 2 π f t d f , ⟨ s I ( t ) s J ( t ) ⟩ = r I J s ( 0 ) = ∫ S I J ( f ) d f
对于随机背景如果不做时间延迟,相关就是信号相乘并对时间积分,由于是随机过程,需要求系综平均,即多次测量求平均。噪声不相关求平均为0。
⟨ d I ( t ) d J ( t ) ⟩ = ⟨ s I ( t ) s J ( t ) ⟩ \left\langle d_I(t) d_J(t) \right\rangle = \left\langle s_I(t) s_J(t) \right\rangle
⟨ d I ( t ) d J ( t ) ⟩ = ⟨ s I ( t ) s J ( t ) ⟩
⟨ d I ( t ) d J ( t ) d I ( t ′ ) d J ( t ′ ) ⟩ = ⟨ s I ( t ) s I ( t ′ ) s J ( t ) s J ( t ′ ) ⟩ + ⟨ n I ( t ) n I ( t ′ ) n J ( t ) n J ( t ′ ) ⟩ + ⟨ s I ( t ) s I ( t ′ ) n J ( t ) n J ( t ′ ) ⟩ + ⟨ n I ( t ) n I ( t ′ ) s J ( t ) s J ( t ′ ) ⟩ \begin{aligned}
\left\langle d_I(t) d_J(t) d_I(t') d_J(t') \right\rangle =& \left\langle s_I(t)s_I(t') s_J(t)s_J(t') \right\rangle + \left\langle n_I(t) n_I(t') n_J(t)n_J(t') \right\rangle\\
&+ \left\langle s_I(t)s_I(t') n_J(t)n_J(t') \right\rangle + \left\langle n_I(t)n_I(t') s_J(t)s_J(t') \right\rangle
\end{aligned} ⟨ d I ( t ) d J ( t ) d I ( t ′ ) d J ( t ′ ) ⟩ = ⟨ s I ( t ) s I ( t ′ ) s J ( t ) s J ( t ′ ) ⟩ + ⟨ n I ( t ) n I ( t ′ ) n J ( t ) n J ( t ′ ) ⟩ + ⟨ s I ( t ) s I ( t ′ ) n J ( t ) n J ( t ′ ) ⟩ + ⟨ n I ( t ) n I ( t ′ ) s J ( t ) s J ( t ′ ) ⟩
⟨ s I ( t ) s I ( t ′ ) s J ( t ) s J ( t ′ ) ⟩ ↔ ⟨ s I ( t ) s J ( t ) ⟩ ⟨ s I ( t ′ ) s J ( t ′ ) ⟩ ↔ ⟨ s I ( t ) s I ( t ′ ) ⟩ ⟨ s J ( t ) s J ( t ′ ) ⟩ \left\langle s_I(t)s_I(t') s_J(t)s_J(t') \right\rangle \leftrightarrow \left\langle s_I(t) s_J(t) \right\rangle\left\langle s_I(t') s_J(t') \right\rangle \leftrightarrow \left\langle s_I(t) s_I(t') \right\rangle\left\langle s_J(t) s_J(t') \right\rangle
⟨ s I ( t ) s I ( t ′ ) s J ( t ) s J ( t ′ ) ⟩ ↔ ⟨ s I ( t ) s J ( t ) ⟩ ⟨ s I ( t ′ ) s J ( t ′ ) ⟩ ↔ ⟨ s I ( t ) s I ( t ′ ) ⟩ ⟨ s J ( t ) s J ( t ′ ) ⟩
⟨ d I ( t ) d J ( t ) d I ( t ′ ) d J ( t ′ ) ⟩ − ⟨ s I ( t ) s J ( t ) ⟩ ⟨ s I ( t ′ ) s J ( t ′ ) ⟩ = ⟨ n I ( t ) n I ( t ′ ) n J ( t ) n J ( t ′ ) ⟩ + ⟨ s I ( t ) s I ( t ′ ) n J ( t ) n J ( t ′ ) ⟩ + ⟨ n I ( t ) n I ( t ′ ) s J ( t ) s J ( t ′ ) ⟩ = ⟨ n I ( t ) n I ( t ′ ) ⟩ ⟨ n J ( t ) n J ( t ′ ) ⟩ + ⟨ s I ( t ) s I ( t ′ ) ⟩ ⟨ n J ( t ) n J ( t ′ ) ⟩ + ⟨ n I ( t ) n I ( t ′ ) ⟩ ⟨ s J ( t ) s J ( t ′ ) ⟩ \begin{aligned}
&\left\langle d_I(t) d_J(t) d_I(t') d_J(t') \right\rangle - \left\langle s_I(t) s_J(t) \right\rangle\left\langle s_I(t') s_J(t') \right\rangle \\
&=\left\langle n_I(t) n_I(t') n_J(t)n_J(t') \right\rangle+ \left\langle s_I(t)s_I(t') n_J(t)n_J(t') \right\rangle + \left\langle n_I(t)n_I(t') s_J(t)s_J(t') \right\rangle\\
&=\left\langle n_I(t) n_I(t')\right\rangle \left\langle n_J(t)n_J(t') \right\rangle + \left\langle s_I(t)s_I(t')\right\rangle \left\langle n_J(t)n_J(t') \right\rangle + \left\langle n_I(t)n_I(t')\right\rangle \left\langle s_J(t)s_J(t') \right\rangle\\
\end{aligned} ⟨ d I ( t ) d J ( t ) d I ( t ′ ) d J ( t ′ ) ⟩ − ⟨ s I ( t ) s J ( t ) ⟩ ⟨ s I ( t ′ ) s J ( t ′ ) ⟩ = ⟨ n I ( t ) n I ( t ′ ) n J ( t ) n J ( t ′ ) ⟩ + ⟨ s I ( t ) s I ( t ′ ) n J ( t ) n J ( t ′ ) ⟩ + ⟨ n I ( t ) n I ( t ′ ) s J ( t ) s J ( t ′ ) ⟩ = ⟨ n I ( t ) n I ( t ′ ) ⟩ ⟨ n J ( t ) n J ( t ′ ) ⟩ + ⟨ s I ( t ) s I ( t ′ ) ⟩ ⟨ n J ( t ) n J ( t ′ ) ⟩ + ⟨ n I ( t ) n I ( t ′ ) ⟩ ⟨ s J ( t ) s J ( t ′ ) ⟩
在弱信号假设下⟨ s ( t ) s ( t ′ ) ⟩ ≪ ⟨ n ( t ) n ( t ′ ) ⟩ \left\langle s(t)s(t')\right\rangle \ll \left\langle n(t)n(t') \right\rangle ⟨ s ( t ) s ( t ′ ) ⟩ ≪ ⟨ n ( t ) n ( t ′ ) ⟩
⟨ d I ( t ) d J ( t ) d I ( t ′ ) d J ( t ′ ) ⟩ − ⟨ d I ( t ) d J ( t ) ⟩ 2 ≈ ⟨ n I ( t ) n I ( t ′ ) ⟩ ⟨ n J ( t ) n J ( t ′ ) ⟩ \left\langle d_I(t) d_J(t) d_I(t') d_J(t') \right\rangle - \left\langle d_I(t) d_J(t) \right\rangle^2 \approx \left\langle n_I(t) n_I(t') \right\rangle \left\langle n_J(t)n_J(t') \right\rangle
⟨ d I ( t ) d J ( t ) d I ( t ′ ) d J ( t ′ ) ⟩ − ⟨ d I ( t ) d J ( t ) ⟩ 2 ≈ ⟨ n I ( t ) n I ( t ′ ) ⟩ ⟨ n J ( t ) n J ( t ′ ) ⟩
考虑到引力波信号与噪声不相关:
r ( t ) = d ⋆ d = s ⋆ s + n ⋆ n ; r I J ( t ) = d I ⋆ d J = s I ⋆ s J + n I ⋆ n J r(t) = d \star d = s \star s + n \star n; ~~~ r_{IJ}(t) = d_I \star d_J = s_I \star s_J + n_I \star n_J
r ( t ) = d ⋆ d = s ⋆ s + n ⋆ n ; r I J ( t ) = d I ⋆ d J = s I ⋆ s J + n I ⋆ n J
如果进一步假设不同探测器噪声不相关:
r ( t ) = s ⋆ s + n ⋆ n ; r I J ( t ) = s I ⋆ s J r(t) = s \star s + n \star n; ~~~ r_{IJ}(t) = s_I \star s_J
r ( t ) = s ⋆ s + n ⋆ n ; r I J ( t ) = s I ⋆ s J
⟨ d 1 d 2 ⟩ = ⟨ s 1 s 2 ⟩ + ⟨ n 1 n 2 ⟩ + ⟨ s 1 n 2 ⟩ + ⟨ n 1 s 2 ⟩ \left\langle d_1 d_2 \right\rangle = \left\langle s_1 s_2\right\rangle + \left\langle n_1 n_2\right\rangle + \left\langle s_1 n_2\right\rangle + \left\langle n_1 s_2\right\rangle
⟨ d 1 d 2 ⟩ = ⟨ s 1 s 2 ⟩ + ⟨ n 1 n 2 ⟩ + ⟨ s 1 n 2 ⟩ + ⟨ n 1 s 2 ⟩
轨道平均 v.s. 方位平均
地基探测器:地球自转+公转
空间探测器:探测器自转+公转,卫星三角在探测器平面内自转,
信号功率谱描述
S ( f ) = ∣ s ~ ( f ) ∣ 2 = s ~ ( f ) s ~ ∗ ( f ) , S h ( f ) = 1 2 ? ? ∑ P ∣ h ~ P ( f ) ∣ 2 \begin{aligned}
S(f) = |\tilde{s}(f)|^2 = \tilde{s}(f)\tilde{s}^*(f), ~~~S_h(f) = \textcolor{red}{\frac{1}{2}??} \sum_P |\tilde{h}_P(f)|^2
\end{aligned} S ( f ) = ∣ s ~ ( f ) ∣ 2 = s ~ ( f ) s ~ ∗ ( f ) , S h ( f ) = 2 1 ? ? P ∑ ∣ h ~ P ( f ) ∣ 2
其中,S G W ( f ) S_{\rm \tiny GW}(f) S G W ( f ) 为引力波信号原始功率为不同偏振模式信号功率之和。
由此可定义引力波的功率响应函数:
R ( f ) = S ( f ) S h ( f ) = s ~ ( f ) s ~ ∗ ( f ) 1 2 ? ? ∑ P ∣ h ~ P ( f ) ∣ 2 \boxed{\mathcal{R}(f) = \frac{S(f)}{S_h(f)} = \frac{\tilde{s}(f)\tilde{s}^*(f)}{\textcolor{red}{\frac{1}{2}??} \sum_P |\tilde{h}_P(f)|^2}}
R ( f ) = S h ( f ) S ( f ) = 2 1 ? ? ∑ P ∣ h ~ P ( f ) ∣ 2 s ~ ( f ) s ~ ∗ ( f )
通常说的引力波响应函数指的就是功率转移函数(power transfer function)
这里要有一个R(f)的曲线图,覆盖从PTA到地基观测的所有波段,参考Romano & Cornish 2017 P74
显然这里s ~ ( f ) , S ( f ) , R ( f ) \tilde{s}(f), S(f), \mathcal{R}(f) s ~ ( f ) , S ( f ) , R ( f ) 都是依赖方位角及偏振角的函数。一般情况下,难以化简,但若对偏振角求平均
多探测器联合
( F + F × ) = R 2 ψ ( D + D × ) = ( cos 2 ψ − sin 2 ψ sin 2 ψ cos 2 ψ ) ( D + D × ) \begin{aligned}
\begin{pmatrix}
F^{+}\\
F^{\times}
\end{pmatrix} =
\mathcal{R}_{2\psi}
\begin{pmatrix}
D^{+} \\
D^{\times}
\end{pmatrix} =
\begin{pmatrix}
\cos 2\psi & -\sin 2\psi \\
\sin 2\psi & \cos 2\psi
\end{pmatrix}
\begin{pmatrix}
D^{+} \\
D^{\times}
\end{pmatrix}
\end{aligned} ( F + F × ) = R 2 ψ ( D + D × ) = ( cos 2 ψ sin 2 ψ − sin 2 ψ cos 2 ψ ) ( D + D × )
F + F + ∗ = cos 2 2 ψ D + D + ∗ − 1 2 sin 4 ψ ( D + D × ∗ + D + ∗ D × ) + sin 2 2 ψ D × D × ∗ F × F × ∗ = sin 2 2 ψ D + D + ∗ + 1 2 sin 4 ψ ( D + D × ∗ + D + ∗ D × ) + cos 2 2 ψ D × D × ∗ F + F × ∗ = 1 2 sin 4 ψ ( D + D + ∗ − D × D × ∗ ) + cos 2 2 ψ D + D × ∗ − sin 2 2 ψ D + ∗ D × F + ∗ F × = 1 2 sin 4 ψ ( D + D + ∗ − D × D × ∗ ) + cos 2 2 ψ D + ∗ D × − sin 2 2 ψ D + D × ∗ \begin{aligned}
&\small F^{+} {F^{+}}^* = \cos^2 2\psi D^{+}{D^{+}}^* - \frac{1}{2}\sin 4\psi (D^{+} {D^{\times}}^* + {D^{+}}^*D^{\times} ) + \sin^2 2\psi D^{\times} {D^{\times}}^*\\
&\small F^{\times} {F^{\times}}^* = \sin^2 2\psi D^{+}{D^{+}}^* + \frac{1}{2}\sin 4\psi (D^{+} {D^{\times}}^* + {D^{+}}^*D^{\times} ) + \cos^2 2\psi D^{\times} {D^{\times}}^*\\
&\small F^{+} {F^{\times}}^* = \frac{1}{2}\sin 4\psi \left( D^{+}{D^{+}}^* - D^{\times}{D^{\times}}^* \right) + \cos^2 2\psi D^{+} {D^{\times}}^* - \sin^2 2\psi {D^{+}}^* D^{\times}\\
&\small {F^{+}}^* F^{\times} = \frac{1}{2}\sin 4\psi \left( D^{+}{D^{+}}^* - D^{\times}{D^{\times}}^* \right) + \cos^2 2\psi {D^{+}}^* D^{\times} - \sin^2 2\psi D^{+} {D^{\times}}^*\\
% &\small {F^{+}}^* F^{\times} + {F^{+}}^* F^{\times} = \frac{1}{2}\sin 4\psi \left( D^{+}{D^{+}}^* - D^{\times}{D^{\times}}^* \right) + \cos 4\psi \left({D^{+}}^* D^{\times} + D^{+} {D^{\times}}^*\right)
\end{aligned} F + F + ∗ = cos 2 2 ψ D + D + ∗ − 2 1 sin 4 ψ ( D + D × ∗ + D + ∗ D × ) + sin 2 2 ψ D × D × ∗ F × F × ∗ = sin 2 2 ψ D + D + ∗ + 2 1 sin 4 ψ ( D + D × ∗ + D + ∗ D × ) + cos 2 2 ψ D × D × ∗ F + F × ∗ = 2 1 sin 4 ψ ( D + D + ∗ − D × D × ∗ ) + cos 2 2 ψ D + D × ∗ − sin 2 2 ψ D + ∗ D × F + ∗ F × = 2 1 sin 4 ψ ( D + D + ∗ − D × D × ∗ ) + cos 2 2 ψ D + ∗ D × − sin 2 2 ψ D + D × ∗
对于长周期信号,随着轨道运动
对轨道/探测器方位求平均⟨ D + D × ∗ ⟩ = ⟨ D + ∗ D × ⟩ \left\langle D^{+} {D^{\times}}^* \right\rangle = \left\langle {D^{+}}^* D^{\times} \right\rangle ⟨ D + D × ∗ ⟩ = ⟨ D + ∗ D × ⟩ ,从而有:
⟨ F + F + ∗ ⟩ = cos 2 2 ψ ⟨ D + D + ∗ ⟩ − sin 4 ψ ⟨ D + D × ∗ ⟩ + sin 2 2 ψ ⟨ D × D × ∗ ⟩ ⟨ F × F × ∗ ⟩ = sin 2 2 ψ ⟨ D + D + ∗ ⟩ + sin 4 ψ ⟨ D + D × ∗ ⟩ + cos 2 2 ψ ⟨ D × D × ∗ ⟩ ⟨ F + F × ∗ ⟩ = ⟨ F + ∗ F × ⟩ = 1 2 sin 4 ψ ( ⟨ D + D + ∗ ⟩ − ⟨ D × D × ∗ ⟩ ) + cos 4 ψ ⟨ D + D × ∗ ⟩ \begin{aligned}
&\small \left\langle F^{+} {F^{+}}^*\right\rangle = \cos^2 2\psi \left\langle D^{+}{D^{+}}^*\right\rangle - \sin 4\psi \left\langle D^{+} {D^{\times}}^*\right\rangle + \sin^2 2\psi \left\langle D^{\times} {D^{\times}}^*\right\rangle\\
&\small \left\langle F^{\times} {F^{\times}}^*\right\rangle = \sin^2 2\psi \left\langle D^{+}{D^{+}}^*\right\rangle + \sin 4\psi \left\langle D^{+} {D^{\times}}^*\right\rangle + \cos^2 2\psi \left\langle D^{\times} {D^{\times}}^*\right\rangle\\
&\small \left\langle F^{+} {F^{\times}}^*\right\rangle = \left\langle {F^{+}}^* F^{\times}\right\rangle = \frac{1}{2}\sin 4\psi \left( \left\langle D^{+}{D^{+}}^*\right\rangle - \left\langle D^{\times}{D^{\times}}^*\right\rangle \right) + \cos 4\psi \left\langle D^{+} {D^{\times}}^*\right\rangle
\end{aligned} ⟨ F + F + ∗ ⟩ = cos 2 2 ψ ⟨ D + D + ∗ ⟩ − sin 4 ψ ⟨ D + D × ∗ ⟩ + sin 2 2 ψ ⟨ D × D × ∗ ⟩ ⟨ F × F × ∗ ⟩ = sin 2 2 ψ ⟨ D + D + ∗ ⟩ + sin 4 ψ ⟨ D + D × ∗ ⟩ + cos 2 2 ψ ⟨ D × D × ∗ ⟩ ⟨ F + F × ∗ ⟩ = ⟨ F + ∗ F × ⟩ = 2 1 sin 4 ψ ( ⟨ D + D + ∗ ⟩ − ⟨ D × D × ∗ ⟩ ) + cos 4 ψ ⟨ D + D × ∗ ⟩
对长周期的单频信号(双白矮星)
进一步的,如果对偏振角求平均有:
⟨ F P F P ′ ∗ ⟩ ‾ = ⟨ ∣ D + ∣ 2 ⟩ + ⟨ ∣ D × ∣ 2 ⟩ 2 δ P P ′ \overline{ \left\langle F^P {F^{P'}}^* \right\rangle } = {\small \frac{ \left\langle {\left|D^{+}\right|}^2\right\rangle + \left\langle {\left|D^{\times}\right|}^2\right\rangle}{2} } \delta_{PP'}
⟨ F P F P ′ ∗ ⟩ = 2 ⟨ ∣ D + ∣ 2 ⟩ + ⟨ ∣ D × ∣ 2 ⟩ δ P P ′
⟨ F + F + ∗ ⟩ ‾ = ⟨ F × F × ∗ ⟩ ‾ = ⟨ ∣ D + ∣ 2 ⟩ + ⟨ ∣ D × ∣ 2 ⟩ 2 \overline{ \left\langle F^+ {F^+}^* \right\rangle } = \overline{ \left\langle F^\times {F^\times}^* \right\rangle } = {\small \frac{ \left\langle {\left|D^{+}\right|}^2\right\rangle + \left\langle {\left|D^{\times}\right|}^2\right\rangle}{2} }
⟨ F + F + ∗ ⟩ = ⟨ F × F × ∗ ⟩ = 2 ⟨ ∣ D + ∣ 2 ⟩ + ⟨ ∣ D × ∣ 2 ⟩
⟨ F + F × ∗ ⟩ ‾ = ⟨ F × F + ∗ ⟩ ‾ = 0 \overline{ \left\langle F^+ {F^\times}^* \right\rangle } = \overline{ \left\langle F^\times {F^+}^* \right\rangle } = 0
⟨ F + F × ∗ ⟩ = ⟨ F × F + ∗ ⟩ = 0
对方位角和偏振角平均后所有偏振交叉项都为零,且两种偏振模式的(平均)功率响应相同。
s ~ ( f ) s ~ ∗ ( f ) = ∑ P ⟨ F P F P ′ ∗ ⟩ ‾ ∣ h ~ P ( f ) ∣ 2 \tilde{s}(f)\tilde{s}^*(f) = \sum_P \overline{ \left\langle F^P {F^{P'}}^* \right\rangle } |\tilde{h}_P(f)|^2
s ~ ( f ) s ~ ∗ ( f ) = P ∑ ⟨ F P F P ′ ∗ ⟩ ∣ h ~ P ( f ) ∣ 2
由于对偏振角以及方位角平均后有⟨ F + F + ∗ ⟩ ‾ = ⟨ F × F × ∗ ⟩ ‾ \overline{ \left\langle F^+ {F^+}^* \right\rangle } = \overline{ \left\langle F^\times {F^\times}^* \right\rangle } ⟨ F + F + ∗ ⟩ = ⟨ F × F × ∗ ⟩ ,从而响应函数:
R ( f , k ^ ) = s ~ ( f ) s ~ ∗ ( f ) ∑ P ∣ h ~ P ( f ) ∣ 2 = ⟨ F + F + ∗ ⟩ ‾ = ⟨ F × F × ∗ ⟩ ‾ = ⟨ ∣ D + ∣ 2 ⟩ + ⟨ ∣ D × ∣ 2 ⟩ 2 \mathcal{R}(f, \hat{k}) = \frac{\tilde{s}(f)\tilde{s}^*(f)}{\sum_P |\tilde{h}_P(f)|^2} = \overline{\left\langle F^+ {F^+}^*\right\rangle} = \overline{\left\langle F^\times {F^\times}^*\right\rangle} = {\small \frac{ \left\langle {\left|D^{+}\right|}^2\right\rangle + \left\langle {\left|D^{\times}\right|}^2\right\rangle}{2} }
R ( f , k ^ ) = ∑ P ∣ h ~ P ( f ) ∣ 2 s ~ ( f ) s ~ ∗ ( f ) = ⟨ F + F + ∗ ⟩ = ⟨ F × F × ∗ ⟩ = 2 ⟨ ∣ D + ∣ 2 ⟩ + ⟨ ∣ D × ∣ 2 ⟩
R ( f , k ^ ) = s ~ ( f ) s ~ ∗ ( f ) 1 2 ? ? ∑ P ∣ h ~ P ( f ) ∣ 2 = ∑ P ⟨ F P F P ∗ ⟩ ‾ = ⟨ ∣ D + ∣ 2 ⟩ + ⟨ ∣ D × ∣ 2 ⟩ \mathcal{R}(f, \hat{k}) = \frac{\tilde{s}(f)\tilde{s}^*(f)}{\textcolor{red}{\frac{1}{2}??} \sum_P |\tilde{h}_P(f)|^2} = \sum_P \overline{\left\langle F^P {F^P}^*\right\rangle} = \left\langle {\left|D^{+}\right|}^2\right\rangle + \left\langle {\left|D^{\times}\right|}^2\right\rangle
R ( f , k ^ ) = 2 1 ? ? ∑ P ∣ h ~ P ( f ) ∣ 2 s ~ ( f ) s ~ ∗ ( f ) = P ∑ ⟨ F P F P ∗ ⟩ = ⟨ ∣ ∣ ∣ D + ∣ ∣ ∣ 2 ⟩ + ⟨ ∣ ∣ ∣ D × ∣ ∣ ∣ 2 ⟩
引力波频谱描述 :S h , Ω , h c S_h, Ω, h_c S h , Ω , h c
检测与参数推断
最佳滤波/匹配滤波
信号探测能力/灵敏度曲线
Fisher信息矩阵
MCMC
探测器联合观测
联合分析 ():多台探测器数据合并,最佳信噪比及对应的参数限制
利用所有探测器的数据纳入同一统计量,计算最佳匹配模板。“综合孔径”(Aperture Synthesis)
对于模板确定的波源,如致密双星并和系统,联合观测的信噪比就是各探测器之和,Fisher信息矩阵也是各探测器之和?。
ρ 2 = ∑ I ρ I 2 , I i j = ∑ I I i j I \rho^2 = \sum_I \rho_I^2, \mathcal{I}_{ij} = \sum_I \mathcal{I}^I_{ij}
ρ 2 = I ∑ ρ I 2 , I i j = I ∑ I i j I
N个(全同)探测器的网络相比单探测器,最佳匹配滤波的信噪比提升为N \sqrt{N} N 倍
定义
S t o t = ∑ λ i S ( i ) ∑ i λ i \mathbb{S}^{\rm tot} = \frac{\sum \lambda_i \mathbb{S}^{(i)}}{\sum_i \lambda_i}
S t o t = ∑ i λ i ∑ λ i S ( i )
其中 λ i \lambda_i λ i 的取值需要使得整体信噪比 μ S t o t σ S t o t \frac{\mu_{\mathbb{S}_{\rm tot}}}{\sigma_{\mathbb{S}_{\rm tot}}} σ S t o t μ S t o t 最大化。
μ S ( i ) = T ( i ) ∫ d f S h ( f ) Γ I J ( f ) , σ S ( i ) = T ( i ) ∫ d f N I ( i ) ( f ) N J ( i ) ( f ) Q ~ ( f ) Q ~ ∗ ( f ) \mu_{\mathbb{S}^{(i)}} = T^{(i)}\int df S_h(f) \Gamma_{IJ}(f), ~~~ \sigma_{\mathbb{S}^{(i)}} = T^{(i)}\int df N_I^{(i)}(f)N_J^{(i)}(f) \tilde{Q}(f)\tilde{Q}^*(f)
μ S ( i ) = T ( i ) ∫ d f S h ( f ) Γ I J ( f ) , σ S ( i ) = T ( i ) ∫ d f N I ( i ) ( f ) N J ( i ) ( f ) Q ~ ( f ) Q ~ ∗ ( f )
⟨ ( ∑ λ i S ( i ) ) 2 ⟩ → ∑ λ i 2 ⟨ S ( i ) 2 ⟩ \left\langle \left(\sum \lambda_i \mathbb{S}^{(i)} \right)^2\right\rangle \rightarrow \sum \lambda^2_i \left\langle { \mathbb{S}^{(i)}}^2\right\rangle
⟨ ( ∑ λ i S ( i ) ) 2 ⟩ → ∑ λ i 2 ⟨ S ( i ) 2 ⟩
μ S t o t = ∑ λ i μ S ( i ) ∑ i λ i , σ S t o t = ∑ λ i 2 σ S ( i ) 2 ∑ i λ i \mu_{\mathbb{S}^{\rm tot}} = \frac{\sum \lambda_i \mu_{\mathbb{S}^{(i)}} }{\sum_i \lambda_i}, ~~~ \sigma_{\mathbb{S}^{\rm tot}} = \frac{\sqrt{\sum \lambda^2_i \sigma^2_{\mathbb{S}^{(i)}} } }{\sum_i \lambda_i}
μ S t o t = ∑ i λ i ∑ λ i μ S ( i ) , σ S t o t = ∑ i λ i ∑ λ i 2 σ S ( i ) 2
ρ = μ S t o t σ S t o t = ∑ i λ i μ S ( i ) ∑ i λ i 2 σ S ( i ) 2 = ⟨ λ ∣ σ − 2 ⟩ ⟨ λ ∣ λ ⟩ \rho = \frac{ \mu_{\mathbb{S}^{\rm tot}} }{ \sigma_{\mathbb{S}^{\rm tot}} } = \frac{ \sum_i \lambda_i \mu_{\mathbb{S}^{(i)}} }{ \sqrt{\sum_i \lambda^2_i \sigma^2_{\mathbb{S}^{(i)}} } } = \frac{\left\langle \lambda \big| \sigma^{-2} \right\rangle}{\sqrt{\left\langle \lambda \big| \lambda \right\rangle}}
ρ = σ S t o t μ S t o t = ∑ i λ i 2 σ S ( i ) 2 ∑ i λ i μ S ( i ) = ⟨ λ ∣ ∣ ∣ λ ⟩ ⟨ λ ∣ ∣ ∣ σ − 2 ⟩
其中⟨ a ∣ b ⟩ ≡ ∑ i a i b i σ i 2 T ( i ) \left\langle a \big| b \right\rangle \equiv \sum_i a_i b_i \sigma_i^2 T^{(i)} ⟨ a ∣ ∣ ∣ b ⟩ ≡ ∑ i a i b i σ i 2 T ( i ) ,当λ i = σ i − 2 \lambda_i = \sigma_i^{-2} λ i = σ i − 2 时总信噪比最大:
ρ = ∑ i T ( i ) μ σ i 2 = ∑ i ρ I J ( i ) 2 \rho = \sqrt{\sum_i \frac{ T^{(i)}\mu }{ \sigma^2_i}} = \sqrt{\sum_i {\rho_{\tiny IJ}^{(i)}}^2}
ρ = i ∑ σ i 2 T ( i ) μ = i ∑ ρ I J ( i ) 2
三角定位:
直观定性的,两个探测器的空间定位是天球面上的一个环带,三台探测器时两个环带交叉,对应天球面上两个不同方位的小区域。
具体定量的,从Fisher参数推断角度理解,Fisher信息矩阵是参数协方差矩阵的倒数,总参数协方差为
Σ i j = ∏ Σ i j I ∑ Σ i j I \Sigma_{ij} = \frac{\prod \Sigma^I_{ij}}{\sum \Sigma^I_{ij}}
Σ i j = ∑ Σ i j I ∏ Σ i j I
在参数空间中参数等概率曲面为椭球,参数观测误差越大,对应方向的轴越长。
协方差是相乘关系,因此只要不同探测器的参数不存在兼并,即参数长轴方向有区别,联合分析就可以显著提升对参数误差的限制。
一致性分析 (Coincidence Analysis):独立分析各探测数据,对比确认探测器中一致的候选事件。
相关分析 :相干coherence v.s. 相关correlation
上面讨论的都是数据与模板匹配,这里是数据与数据匹配
∫ − T / 2 T / 2 d t 1 ∫ − T / 2 T / 2 d t 2 d I ( t 1 ) d J ( t 2 ) Q ( t 1 − t 2 ) \int_{-T/2}^{T/2} dt_1 \int_{-T/2}^{T/2} dt_2 ~ d_I(t_1) d_J(t_2) Q(t_1 - t_2)
∫ − T / 2 T / 2 d t 1 ∫ − T / 2 T / 2 d t 2 d I ( t 1 ) d J ( t 2 ) Q ( t 1 − t 2 )
参考随机噪声,最佳滤波信噪比:
ρ I J = ⟨ R ⟩ T ∫ d f γ I J 2 ( f ) S h 2 ( f ) N I ( f ) N J ( f ) \rho_{\tiny IJ} = \langle \mathcal{R} \rangle \sqrt{T} \sqrt{\int df \frac{ \gamma_{IJ}^2(f) S^2_h(f)}{N_I(f) N_J(f)} }
ρ I J = ⟨ R ⟩ T ∫ d f N I ( f ) N J ( f ) γ I J 2 ( f ) S h 2 ( f )
S h ( f ) = f ∣ h ~ c ( f ) ∣ 2 = 1 ⟨ R ⟩ f ∣ s ~ c ( f ) ∣ 2 S_h(f) = f |\tilde{h}_c(f)|^2 = \frac{1}{\langle \mathcal{R} \rangle} f |\tilde{s}_c(f)|^2
S h ( f ) = f ∣ h ~ c ( f ) ∣ 2 = ⟨ R ⟩ 1 f ∣ s ~ c ( f ) ∣ 2
ρ I J = T ∫ d f f 2 γ I J 2 ( f ) ∣ s ~ c ( f ) ∣ 2 N I ( f ) ∣ s ~ c ( f ) ∣ 2 N J ( f ) \rho_{\tiny IJ} = \sqrt{T} \sqrt{\int df ~ f^2 \gamma_{IJ}^2(f) \frac{ |\tilde{s}_c(f)|^2}{N_I(f)} \frac{ |\tilde{s}_c(f)|^2}{N_J(f)} }
ρ I J = T ∫ d f f 2 γ I J 2 ( f ) N I ( f ) ∣ s ~ c ( f ) ∣ 2 N J ( f ) ∣ s ~ c ( f ) ∣ 2
f 2 f^2 f 2 项开根号后与T \sqrt{T} T 也抵消不了,量纲不对???
这里是对角向积分过后的结果,对于点源而言是不需要的!!
这里的相关分析和前面的联合分析的区别和联系,信噪比会比匹配滤波显著低吗???
同样需要知道S h ( f ) S_h(f) S h ( f ) 的先验,以确定最佳滤波;重叠约化函数
对比匹配滤波的信噪比
ρ I J = ∫ d f ∣ s ~ ( f ) ∣ 2 N ( f ) , ∣ s ~ ( f ) ∣ 2 = R ( f ) S h ( f ) \rho_{\tiny IJ} = \sqrt{\int df \frac{ |\tilde{s}(f)|^2}{N(f)} }, |\tilde{s}(f)|^2 = \mathcal{R}(f)S_h(f)
ρ I J = ∫ d f N ( f ) ∣ s ~ ( f ) ∣ 2 , ∣ s ~ ( f ) ∣ 2 = R ( f ) S h ( f )
随机过程的相关函数
对各态历经的平稳随机过程,有两种等价的自(互)相关 :
一种是系综平均:E { n I ( t ) n J ( t + τ ) } , ⟨ n I ( t ) n J ( t + τ ) ⟩ E\{ n_I(t)n_J(t+\tau) \}, \big\langle n_I(t)n_J(t+\tau)\big\rangle E { n I ( t ) n J ( t + τ ) } , ⟨ n I ( t ) n J ( t + τ ) ⟩
一种是时间平均:lim T → ∞ 1 T ∫ − T / 2 T / 2 n I ( t ) n J ( t + τ ) d t \lim_{T\rightarrow\infty} \frac{1}{T}\int_{-T/2}^{T/2} n_I(t)n_J(t+\tau) dt lim T → ∞ T 1 ∫ − T / 2 T / 2 n I ( t ) n J ( t + τ ) d t
注意,相比确定性函数的互相关∫ − ∞ ∞ f ( t ) g ( t + τ ) d t \int_{-\infty}^{\infty} f(t) g(t+\tau) dt ∫ − ∞ ∞ f ( t ) g ( t + τ ) d t ,这里随机过程互相关多了对时间平均。从各态历经角度理解,系综平均等于时间平均,这里的1 T \frac{1}{T} T 1 来自对时间平均;而直观地,不同于通常函数,对于随机过程,当T → ∞ T\rightarrow\infty T → ∞ 积分是发散的,除以T T T 整体结果才收敛。随机过程是功率信号,能量积分发散,需要除以时间得到功率;相对的,对确定性的能量信号,能量有限,反而不能除以(无限长的)时间。
考虑到随机过程能量发散,并不存在傅里叶变换,但形式上可记为n ~ ( f ) \tilde{n}(f) n ~ ( f ) 。
随机过程能量发散(功率信号),n ~ ( f ) \tilde{n}(f) n ~ ( f ) 是发散的,但功率谱密度有限
⟨ n ~ ( f ) n ~ ∗ ( f ′ ) ⟩ = δ ( f − f ′ ) S n ( f ) \left\langle \tilde{n}(f) \tilde{n}^*(f') \right\rangle = \delta(f-f')S_n(f)
⟨ n ~ ( f ) n ~ ∗ ( f ′ ) ⟩ = δ ( f − f ′ ) S n ( f )
⟨ n ~ ( f ) n ~ ∗ ( f ) ⟩ = δ ( 0 ) S n ( f ) → ∞ \left\langle \tilde{n}(f) \tilde{n}^*(f) \right\rangle = \delta(0)S_n(f) \rightarrow \infty
⟨ n ~ ( f ) n ~ ∗ ( f ) ⟩ = δ ( 0 ) S n ( f ) → ∞
对于确定信号???
信噪比与误报率
假设目标信号服从正态分布,对于信噪比阈值,在统计上有:
ρ t h r e s h o l d = 2 ( e r f c i n v ( 2 α F A R ) − e r f c i n v [ 2 ( 1 − β F N R ) ] ) \rho_{\rm threshold} = \sqrt{2}\Big({\rm erfcinv}(2\alpha_{\rm FAR})-{\rm erfcinv}[2(1-\beta_{\rm FNR})]\Big)
ρ t h r e s h o l d = 2 ( e r f c i n v ( 2 α F A R ) − e r f c i n v [ 2 ( 1 − β F N R ) ] )
这里α F A R , β F N R \alpha_{\rm FAR},\beta_{\rm FNR} α F A R , β F N R 分别为预设的误报率(False alarm rate)和漏报率(False negative rate),1 − β F N R 1-\beta_{\rm FNR} 1 − β F N R 对应于信号探测率。信号探测的误报和漏报率都取千分之一,则对应信噪比阈值约6.2,引力波观测中信噪比阈值通常取6 ~ 8。
对于随机背景,多探测器互相关,通常取信噪比阈值2,对应95%以上的置信度(误报率5%以下),单探测器自相关,信噪比阈值则约定为5。
特征应变强度
δ \delta δ 函数与s i n c \rm sinc s i n c 函数
δ ( x ) = lim ε → ∞ 1 ∣ ε ∣ π e − x 2 ε 2 = lim ε → ∞ 1 2 ε e − ∣ x ∣ ε = lim ε → 0 sin ( x ε ) π x = lim ε → 0 sin ( π x ε ) π x = lim ε → 0 1 ε s i n c ( x ε ) = lim ε → ∞ 1 ε r e c t ( x ε ) = lim ε → 0 ε r e c t ( ε x ) \begin{aligned}
\delta(x) &= \lim_{\varepsilon\rightarrow\infty} \frac{1}{|\varepsilon|\sqrt{\pi} } e^{-\frac{x^2}{\varepsilon^2}}
= \lim_{\varepsilon\rightarrow\infty} \frac{1}{2 \varepsilon} e^{-\frac{|x|}{\varepsilon}}
= \lim_{\varepsilon\rightarrow0} \frac{\sin\left(\frac{x}{\varepsilon}\right)}{\pi x} % <!-- https://functions.wolfram.com/GeneralizedFunctions/DiracDelta/09/0006/ -->
= \lim_{\varepsilon\rightarrow 0} \frac{\sin\left(\frac{\pi x}{\varepsilon}\right)}{\pi x}\\ % <!-- https://en.wikipedia.org/wiki/Sinc_function -->
&= \lim_{\varepsilon\rightarrow 0} \frac{1}{\varepsilon}{\rm sinc}\left(\frac{x}{\varepsilon}\right)
= \lim_{\varepsilon\rightarrow \infty} \frac{1}{\varepsilon} ~ {\rm rect}\left(\frac{x}{\varepsilon}\right)
= \lim_{\varepsilon\rightarrow 0} \varepsilon ~ {\rm rect}\left(\varepsilon x\right)
\end{aligned} δ ( x ) = ε → ∞ lim ∣ ε ∣ π 1 e − ε 2 x 2 = ε → ∞ lim 2 ε 1 e − ε ∣ x ∣ = ε → 0 lim π x sin ( ε x ) = ε → 0 lim π x sin ( ε π x ) = ε → 0 lim ε 1 s i n c ( ε x ) = ε → ∞ lim ε 1 r e c t ( ε x ) = ε → 0 lim ε r e c t ( ε x )
∫ − T / 2 T / 2 e − i 2 π f t d t = ∫ r e c t ( t T ) e − i 2 π f t d t = T s i n c ( T f ) δ ( x ) = lim T → ∞ ∫ − T / 2 T / 2 e − i 2 π f t d t = lim T → ∞ T s i n c ( T f ) \boxed{\begin{gathered}
\int_{-T/2}^{T/2} e^{-i 2\pi f t} dt = \int {\rm rect}\left({\small \frac{t}{\footnotesize T}}\right) e^{-i 2\pi f t} dt = T {\rm sinc}(Tf)\\
\delta(x) = \lim_{T\rightarrow\infty} \int_{-T/2}^{T/2} e^{-i 2\pi f t} dt = \lim_{T\rightarrow\infty} T {\rm sinc}(Tf)
\end{gathered}} ∫ − T / 2 T / 2 e − i 2 π f t d t = ∫ r e c t ( T t ) e − i 2 π f t d t = T s i n c ( T f ) δ ( x ) = T → ∞ lim ∫ − T / 2 T / 2 e − i 2 π f t d t = T → ∞ lim T s i n c ( T f )
δ T ( f ) ≡ T s i n c ( T f ) = sin ( π T f ) π f δ T ( 0 ) = T , lim T → ∞ δ T ( f ) = δ ( f ) \begin{gathered}
\delta_T(f) \equiv T {\rm sinc}(Tf) = \frac{\sin(\pi Tf)}{\pi f}\\
\delta_T(0) = T, ~~~ \lim_{T\rightarrow\infty}\delta_T(f) = \delta(f)
\end{gathered} δ T ( f ) ≡ T s i n c ( T f ) = π f sin ( π T f ) δ T ( 0 ) = T , T → ∞ lim δ T ( f ) = δ ( f )
σ S = ∫ − T / 2 T / 2 d t ∫ − T / 2 T / 2 d t ′ ∫ d f ∫ d f ′ e i 2 π f ( t − t ′ ) e − i 2 π f ′ ( t − t ′ ) N I ( f ) N J ( f ′ ) = ∫ d f ∫ d f ′ N I ( f ) N J ( f ′ ) ∫ − T / 2 T / 2 d t ∫ − T / 2 T / 2 d t ′ e i 2 π f ( t − t ′ ) e i 2 π f ′ ( t − t ′ ) = ∫ d f ∫ d f ′ N I ( f ) N J ( f ′ ) δ T 2 ( f − f ′ ) ≈ ∫ d f ∫ d f ′ N I ( f ) N J ( f ′ ) δ ( f − f ′ ) δ T ( f − f ′ ) = ∫ d f N I ( f ) N J ( f ) δ T ( 0 ) = T ∫ d f N I ( f ) N J ( f ) \begin{aligned}
\sigma_{\mathbb{S}} &= \int_{-T/2}^{T/2} dt \int_{-T/2}^{T/2} dt' \int df \int df' e^{i 2\pi f (t-t')} e^{-i 2\pi f' (t-t')} N_I(f) N_J(f')\\
&= \int df \int df' N_I(f) N_J(f') \int_{-T/2}^{T/2} dt \int_{-T/2}^{T/2} dt' e^{i2\pi f(t-t')} e^{i2\pi f'(t-t')} \\
&= \int df \int df' ~ N_I(f) N_J(f') ~ \delta^2_T(f-f')\\
&\approx \int df \int df' ~ N_I(f) N_J(f') ~ \delta(f-f') ~ \delta_T(f-f')\\
&= \int df ~ N_I(f) N_J(f) ~ \delta_T(0)\\
&= T \int df ~ N_I(f) N_J(f)\\
\end{aligned} σ S = ∫ − T / 2 T / 2 d t ∫ − T / 2 T / 2 d t ′ ∫ d f ∫ d f ′ e i 2 π f ( t − t ′ ) e − i 2 π f ′ ( t − t ′ ) N I ( f ) N J ( f ′ ) = ∫ d f ∫ d f ′ N I ( f ) N J ( f ′ ) ∫ − T / 2 T / 2 d t ∫ − T / 2 T / 2 d t ′ e i 2 π f ( t − t ′ ) e i 2 π f ′ ( t − t ′ ) = ∫ d f ∫ d f ′ N I ( f ) N J ( f ′ ) δ T 2 ( f − f ′ ) ≈ ∫ d f ∫ d f ′ N I ( f ) N J ( f ′ ) δ ( f − f ′ ) δ T ( f − f ′ ) = ∫ d f N I ( f ) N J ( f ) δ T ( 0 ) = T ∫ d f N I ( f ) N J ( f )
噪声功率谱作为频率的函数,在Hz的尺度上变动,而信号的观测时间T T T 在月的量级,对应δ T ( f ) = T s i n c ( T f ) \delta_T(f) = T {\rm sinc}(Tf) δ T ( f ) = T s i n c ( T f ) 中心主峰宽度∼ 1 T \sim \frac{1}{T} ∼ T 1 ,远小于噪声功率谱变化尺度。在上面的计算中其中的一个δ T ( f ) \delta_T(f) δ T ( f ) 被近似替换为δ ( f ) \delta(f) δ ( f ) 。
∫ − T / 2 T / 2 d t ∫ − T / 2 T / 2 d t ′ e i 2 π f ( t − t ′ ) e − i 2 π f ′ ( t − t ′ ) ↔ δ T 2 ( f − f ′ ) ? ? ? \int_{-T/2}^{T/2} dt \int_{-T/2}^{T/2} dt' e^{i 2\pi f (t-t')} e^{-i 2\pi f' (t-t')} \leftrightarrow \delta^2_T(f-f')???
∫ − T / 2 T / 2 d t ∫ − T / 2 T / 2 d t ′ e i 2 π f ( t − t ′ ) e − i 2 π f ′ ( t − t ′ ) ↔ δ T 2 ( f − f ′ ) ? ? ?
准周期连续源
波源简介
Monochromatic
旋转中子星、双白矮星、EMRI系统(FastEMRIWaveform)
方法概述
对于单频信号,理想情况下,直接傅里叶变换,频谱峰值即对应信号频率,实际情况则复杂很多。假设信号很弱,积累足够信噪比需要1年时间,为保证DFT后信号落入单个频率区间,观测时间内信号频率演化需小于DFT的频率分辨率,即f ˙ T o b s ≲ 1 / T o b s ∼ 30 n H z \dot{f}T_{\rm obs}\lesssim 1/T_{\rm obs} \sim 30\rm nHz f ˙ T o b s ≲ 1 / T o b s ∼ 3 0 n H z ,f ˙ ≲ 1 / T o b s 2 ∼ 1 0 − 15 H z / s \dot{f}\lesssim 1/T^2_{\rm obs} \sim 10^{-15}\rm Hz/s f ˙ ≲ 1 / T o b s 2 ∼ 1 0 − 1 5 H z / s 。即便不考虑探测器运动引入的多普勒调制,波源本身由于引力波辐射导致的频率演化通常也不满足该要求,而且信号越强,演化越明显。显然,只有对多普勒调制及波源演化都进行建模和修正之后,简单的傅里叶分析才可行,而这需要对波源位置及参数有足够精确先验,如电磁观测精确建模的波源。
对周期性信号的搜索,根据观测目标可分为靶向搜索、定向搜索以及全天搜索,分别对应波源已知(方位及频率)、方位已知以及波源未知三种情况。根据计算方法又可分为全相干法和半相干法:全相干法一次性分析所有观测数据,考虑信号频率和相位跨越全部观测时间的演化以及探测器噪声的变化;半相干法则是将数据分段,每段内进行相干分析(同时考虑频率和相位演化),各数据段之间则忽略相位的演化,非相干叠加。
显然,全相干法具有更好的灵敏度,但通常并不现实,下面从波源的模板和定位两方面估算。对于波源本身,将其相位演化一般性地展开:
ϕ ( t ) ≈ ϕ 0 + 2 π [ f s ( t − t 0 ) + 1 2 f ˙ s ( t − t 0 ) 2 + 1 6 f ¨ s ( t − t 0 ) 3 ] \phi(t) \approx \phi_0 + 2\pi \left[f_s(t-t_0) + \frac{1}{2}\dot{f}_s(t-t_0)^2 + \frac{1}{6}\ddot{f}_s(t-t_0)^3\right]
ϕ ( t ) ≈ ϕ 0 + 2 π [ f s ( t − t 0 ) + 2 1 f ˙ s ( t − t 0 ) 2 + 6 1 f ¨ s ( t − t 0 ) 3 ]
假设要求在相干时间T ∼ t − t 0 T \sim t-t_0 T ∼ t − t 0 内,模型对相位的预测偏差不超过Δ ϕ \Delta\phi Δ ϕ ,则粗略地,对频率及其导数误差要求为:
Δ f s ∼ 1 2 π Δ ϕ T , Δ f ˙ s ∼ 2 2 π Δ ϕ T 2 , Δ f ¨ s ∼ 6 2 π Δ ϕ T 3 \Delta f_s \sim \frac{1}{2\pi}\frac{\Delta\phi}{T}, ~~~~ \Delta \dot{f}_s \sim \frac{2}{2\pi}\frac{\Delta\phi}{T^2}, ~~~~ \Delta \ddot{f}_s \sim \frac{6}{2\pi}\frac{\Delta\phi}{T^3}
Δ f s ∼ 2 π 1 T Δ ϕ , Δ f ˙ s ∼ 2 π 2 T 2 Δ ϕ , Δ f ¨ s ∼ 2 π 6 T 3 Δ ϕ
这大致对应于模板精度的需求,从而近似到二阶导,所需模板数量将∼ T 6 \sim T^6 ∼ T 6 ,考虑到模板匹配的计算量同样正比于相干时间T T T ,总计量算∼ T 7 \sim T^7 ∼ T 7 。这只是粗略估算,实际中并不会如此极端,低阶效应主导时无需考虑更高阶效应,比如对于二阶导数f ¨ s \ddot{f}_s f ¨ s ,只有当相干时间足够长时影响才显著。除了波源本身演化,探测器运动引入的多普勒调制是另一个重要因素,且依赖于具体波源方位,不同方位需要使用不同修正,大致地:
Δ θ ∼ 1 2 Δ f s f s v ⊥ / c ∼ 1 2 v ⊥ / c 1 f s T , Δ Ω ∼ ( Δ θ ) 2 \Delta\theta \sim \frac{\frac{1}{2}\Delta f_s}{f_s v_{\perp}/c} \sim \frac{1}{2 v_{\perp}/c} \frac{1}{f_s T}, ~~~~ \Delta\Omega \sim (\Delta\theta)^2
Δ θ ∼ f s v ⊥ / c 2 1 Δ f s ∼ 2 v ⊥ / c 1 f s T 1 , Δ Ω ∼ ( Δ θ ) 2
其中v ⊥ v_{\perp} v ⊥ 是探测器速度在垂直波源方向上的投影。由此多普勒修正的模板数∼ f s 2 T 2 \sim f_s^2 T^2 ∼ f s 2 T 2 ,结合前面波源本身模板的需求,随着相干时间增加,计算量将迅速上升,因此全相干分析只有在电磁观测或半相干法提供了足够精确的参数先验情况下才有可能实现。
实际中通常是基于半相干法的层级搜索(Hierarchical Search),即先在完整参数空间中使用相对粗略的模板快速搜索,之后在潜在目标区域精细化搜索,以平衡搜索精度和计算量需求,具体地可大致分为搜索、否决和证认三个阶段:
搜索:基于半相干法全局参数搜索(Search Pipeline)
显然半相干分析会降低探测灵敏度,且不同的搜索策略会有所区别,但大致地,对给定的观测时间,假设数据分为N N N 段,对信号应力强度的灵敏度损失∼ N 1 4 \sim N^\frac{1}{4} ∼ N 4 1 。
F \mathcal{F} F -统计量:GCT、Wave、
频谱分析:弱相干法PowerFlux/Falcon、互相关法Cross-Correlation
时频谱分析:霍夫变换SkyHough/FrequencyHough
频谱分析基于频谱构建统计量,时频谱法则更关注频率演化,通过对二值化时频谱进行霍夫变换,定位频率演化曲线。
维特比算法:Viterbi search
机器学习
否决:通过后处理(Post-Processing)分析,消除误报
后处理的核心是参数聚类(Clustering)以及一致性分析(Coincidence),其中一致性分析可在同一探测器不同时间段或不同探测器之间进行。
一致性否决
χ 2 \chi^2 χ 2 卡方否决
窄带噪声否决
零假设否决
证认:增加相干时间,对候选波源进行后随(Follow-up)证认
后随证认本质是用更高精度对候选信号进行分析,基本的策略是增加相干时间,提升参数网格精度,或基于候选区间对目标参数进行采样(非网格化方法)。具体实现可分为单阶段或多阶段(精度依次提升),数据可复用旧数据或使用新观测数据。
旧数据 v.s. 新数据:证认阶段可复用旧数据或使用新的观测数据。
注意,后随认证阶段增加了分析结果的信噪比和可靠性,但无法提升探测灵敏度,灵敏度完全取决于半相干分析阶段的搜索策略。增加相干时间在增加相位一致性要求的同时也提升了频谱的频率分辨率?[两者是一个东西吗?]。
搜索算法
相干分析:外差法、重采样、狄利克雷滤波、时域分析、五矢量法、F \mathcal{F} F -统计量
半相干法:stack-slide、PowerFlux、HoughTransform、stacked F \mathcal{F} F -statistic
F \mathcal{F} F -统计量
服从卡方分布,而非F分布,与通常的F检验不同?
滑动堆栈 Stack-slide
Stack:对功率平方求和
霍夫变换 Hough Transform:搜索频率演化曲线
时频谱:短时傅里叶变换
“白化”:噪声功率谱估计并非通常的Welch法,而是对周期图进行移动平均,近似于Daniell法。
二值化:阈值、峰值
变换:拉东变换(Radon Transform)、霍夫变换(Hough Transform)
随机背景信号
波源简介
能谱/分数能量密度谱
假设背景是稳态 、各向同性 、无偏振 ;高斯
稳态:随机信号的统计性质不随时间演化,具体的,互相关信号只与时间差有关而于具体时间选择无关。
各向同性:信号源自分布相对均匀的遥远天体或早期宇宙过程,与本地物质不存在关联。
无偏振:各方向偏振强度相等,没有明确的“主”偏振方向。
圆偏振涉及的是两种偏振模式的相位差,这里所谓的“无偏振”是指偏振角随机分布,或者说只是无线偏振,两者并不矛盾?
高斯性:多个探测器信号s ( t ) s(t) s ( t ) 的联合概率密度函数为多元正态分布。对于高斯稳态随机过程,假设均值为零,统计性质完全由其功率谱(自相关函数/二阶距)描述。
对于稳定高斯随机过程,其性质完全由均值(一阶矩)和互相关/自相关(二阶矩)决定。也意味着N点关联可表示为两点关联乘积之和(均值通常设为零)。
其合理性源自大数定律,
根据大数定律,任意足够多的独立随机变量之和都渐近服从正态分布,而与随机变量本身分布无关。因此对于大量独立天体源叠加而成的随机背景,高斯假设是合理的,对于相互关联的早期宇宙源?或少数天体源的叠加则需要谨慎。
在长波近似下,地基观测
在短波近似下,就是Hellings-Downs(HD)曲线。
PTA中观测量通常为多普勒红移(对应频率变化),而非应变响应(对应相位变化)。
此时$z(t) = $
对于单个探测器而言,随机背景与探测器噪声原则是没法区分的,因此对于随机背景的探测通常需要多个探测器互相关,或者对于仪器噪声有准确的测量或模型描述。
随机背景分析
多探测器互相关:多个探测器进行联合观测
探测器噪声先验:通过零通道获取噪声信息
h a b ( t , r ⃗ ) = ∫ d f ∫ h ~ a b ( f , k ^ ) e − i 2 π f ( t − k ^ ⋅ r ⃗ c ) h_{ab}(t, \vec{r}) = \int df \int \tilde{h}_{ab}(f, \hat{k})e^{-i 2\pi f (t-\hat{k}\cdot\frac{\vec{r}}{c})}
h a b ( t , r ) = ∫ d f ∫ h ~ a b ( f , k ^ ) e − i 2 π f ( t − k ^ ⋅ c r )
h ~ a b ( f , k ^ ) = ∑ P h ~ P ( f , k ^ ) e a b P ( k ^ ) \tilde{h}_{ab}(f, \hat{k}) = \sum_P \tilde{h}_P(f, \hat{k})e_{ab}^P(\hat{k}) h ~ a b ( f , k ^ ) = ∑ P h ~ P ( f , k ^ ) e a b P ( k ^ )
∑ P e a b P e a b P = 4 \sum_P e_{ab}^P e_{ab}^P = 4 ∑ P e a b P e a b P = 4
⟨ h ~ P ( f , k ^ ) h ~ P ′ ∗ ( f ′ , k ^ ′ ) ⟩ ‾ = 1 4 π δ ( f − f ′ ) δ 2 ( k ^ , k ^ ′ ) δ P P ′ S h ( f ) \overline{\left\langle \tilde{h}_P(f,\hat{k})\tilde{h}^*_{P'}(f',\hat{k}') \right\rangle} = \frac{1}{4\pi} \delta(f-f')~\delta^2(\hat{k},\hat{k}')~\delta_{PP'}~S_h(f)
⟨ h ~ P ( f , k ^ ) h ~ P ′ ∗ ( f ′ , k ^ ′ ) ⟩ = 4 π 1 δ ( f − f ′ ) δ 2 ( k ^ , k ^ ′ ) δ P P ′ S h ( f )
δ 2 ( k ^ , k ^ ′ ) = δ ( cos θ − cos θ ′ ) δ ( φ − φ ′ ) \delta^2(\hat{k},\hat{k}') = \delta(\cos\theta-\cos\theta')\delta(\varphi-\varphi') δ 2 ( k ^ , k ^ ′ ) = δ ( cos θ − cos θ ′ ) δ ( φ − φ ′ )
∫ d 2 k ^ d 2 k ^ ′ ⟨ h ~ P ( f , k ^ ) h ~ P ′ ∗ ( f ′ , k ^ ′ ) ⟩ ‾ = δ ( f − f ′ ) δ P P ′ S h ( f ) \int d^2\hat{k}~d^2\hat{k}' \overline{\left\langle \tilde{h}_P(f,\hat{k})\tilde{h}^*_{P'}(f',\hat{k}') \right\rangle} = \delta(f-f')~\delta_{PP'}~S_h(f)
∫ d 2 k ^ d 2 k ^ ′ ⟨ h ~ P ( f , k ^ ) h ~ P ′ ∗ ( f ′ , k ^ ′ ) ⟩ = δ ( f − f ′ ) δ P P ′ S h ( f )
d 2 k ^ ′ = d cos θ d φ d^2\hat{k}' = d\cos\theta d\varphi d 2 k ^ ′ = d cos θ d φ
1 2 ∫ d 2 k ^ d 2 k ^ ′ [ ⟨ h ~ + ( f , k ^ ) h ~ + ∗ ( f ′ , k ^ ′ ) ⟩ ‾ + ⟨ h ~ × ( f , k ^ ) h ~ × ∗ ( f ′ , k ^ ′ ) ⟩ ‾ ] = δ ( f − f ′ ) S h ( f ) \frac{1}{2}\int d^2\hat{k}~d^2\hat{k}'\left[\overline{\left\langle \tilde{h}_+(f,\hat{k})\tilde{h}^*_+(f',\hat{k}') \right\rangle} + \overline{\left\langle \tilde{h}_\times(f,\hat{k})\tilde{h}^*_\times(f',\hat{k}') \right\rangle}\right] = \delta(f-f')~ S_h(f)
2 1 ∫ d 2 k ^ d 2 k ^ ′ [ ⟨ h ~ + ( f , k ^ ) h ~ + ∗ ( f ′ , k ^ ′ ) ⟩ + ⟨ h ~ × ( f , k ^ ) h ~ × ∗ ( f ′ , k ^ ′ ) ⟩ ] = δ ( f − f ′ ) S h ( f )
⟨ h ~ P ( f ) h ~ P ′ ∗ ( f ′ ) ⟩ = ∫ d 2 k ^ d 2 k ^ ′ ⟨ h ~ P ( f , k ^ ) h ~ P ′ ∗ ( f ′ , k ^ ′ ) ⟩ ‾ = δ ( f − f ′ ) δ P P ′ S h ( f ) \left\langle \tilde{h}_P(f)\tilde{h}^*_{P'}(f') \right\rangle = \int d^2\hat{k}~d^2\hat{k}'\overline{\left\langle \tilde{h}_P(f,\hat{k})\tilde{h}^*_{P'}(f',\hat{k}') \right\rangle} = \delta(f-f')~\delta_{PP'} S_h(f)
⟨ h ~ P ( f ) h ~ P ′ ∗ ( f ′ ) ⟩ = ∫ d 2 k ^ d 2 k ^ ′ ⟨ h ~ P ( f , k ^ ) h ~ P ′ ∗ ( f ′ , k ^ ′ ) ⟩ = δ ( f − f ′ ) δ P P ′ S h ( f )
⟨ h ~ + ( f ) h ~ + ∗ ( f ′ ) ⟩ = ⟨ h ~ × ( f ) h ~ × ∗ ( f ′ ) ⟩ = δ ( f − f ′ ) S h ( f ) \left\langle \tilde{h}_+(f)\tilde{h}^*_+(f') \right\rangle = \left\langle \tilde{h}_\times(f)\tilde{h}^*_\times(f') \right\rangle = \delta(f-f')~ S_h(f)
⟨ h ~ + ( f ) h ~ + ∗ ( f ′ ) ⟩ = ⟨ h ~ × ( f ) h ~ × ∗ ( f ′ ) ⟩ = δ ( f − f ′ ) S h ( f )
1 2 ( ⟨ h ~ + ( f ) h ~ + ∗ ( f ′ ) ⟩ + ⟨ h ~ × ( f ) h ~ × ∗ ( f ′ ) ⟩ ) = δ ( f − f ′ ) S h ( f ) \frac{1}{2}\left(\left\langle \tilde{h}_+(f)\tilde{h}^*_+(f') \right\rangle + \left\langle \tilde{h}_\times(f)\tilde{h}^*_\times(f') \right\rangle\right) = \delta(f-f')~ S_h(f)
2 1 ( ⟨ h ~ + ( f ) h ~ + ∗ ( f ′ ) ⟩ + ⟨ h ~ × ( f ) h ~ × ∗ ( f ′ ) ⟩ ) = δ ( f − f ′ ) S h ( f )
S h ( f ) S_h(f) S h ( f ) 是偏振平均的功率谱,如果便是为单边谱,则1 / 2 1/2 1 / 2 抵消了,成了求和?如果考虑更多偏振模式?Ω G W \Omega_GW Ω G W 表示随机引力波能量占比,应该是对偏振求和?
⟨ h a b ( t ) h a b ( t ) ⟩ = 4 ∫ d f S h ( f ) ? ? ? \boxed{ \left\langle h_{ab}(t)h^{ab}(t)\right\rangle = 4 \int df S_h(f) }???
⟨ h a b ( t ) h a b ( t ) ⟩ = 4 ∫ d f S h ( f ) ? ? ?
⟨ h + 2 ( t ) + h × 2 ( t ) ⟩ = 2 ∫ d f S h ( f ) h a b h a b = 2 ( h + 2 + h × 2 ) \left\langle h_+^2(t) + h_\times^2(t)\right\rangle = 2 \int df S_h(f) ~~~~~~~ h_{ab}h^{ab}=2(h_+^2 + h_\times^2)
⟨ h + 2 ( t ) + h × 2 ( t ) ⟩ = 2 ∫ d f S h ( f ) h a b h a b = 2 ( h + 2 + h × 2 )
S h ( f ) = f 2 h c 2 ( f ) , ⟨ h + 2 ( t ) + h × 2 ( t ) ⟩ 2 = ∫ d f f h c ( f ) S_h(f) = f^2 h^2_c(f), \frac{\left\langle h_+^2(t) + h_\times^2(t)\right\rangle}{2} = \int df f h_c(f)
S h ( f ) = f 2 h c 2 ( f ) , 2 ⟨ h + 2 ( t ) + h × 2 ( t ) ⟩ = ∫ d f f h c ( f )
S h ( f ) = 3 H 0 2 4 π 2 f 3 Ω G W ( f ) S_h(f) = \frac{3H_0^2}{4\pi^2 f^3} \Omega_{\rm \tiny GW}(f)
S h ( f ) = 4 π 2 f 3 3 H 0 2 Ω G W ( f )
S h ( f ) S_h(f) S h ( f ) 似乎有两种定义方式,一种是各种偏振模式的平均,一种是各种偏振模式的求和。对+ , × +, \times + , × 模式就相差两倍,因此会看到有的文章使用S h ( f ) = 3 H 0 2 2 π 2 f 3 Ω G W ( f ) S_h(f) = \frac{3H_0^2}{2\pi^2 f^3} \Omega_{\rm \tiny GW}(f) S h ( f ) = 2 π 2 f 3 3 H 0 2 Ω G W ( f ) 。跟单边功率谱或双边功率谱似乎是没有关系的,因为如果这里S h S_h S h 是单边的,双边的应该是1/2而非2倍。如果考虑更多偏振模式,这里应该就不再是2倍的关系?
S h ( f ) = ⟨ h ~ P ( f , k ^ ) h ~ P ∗ ( f , k ^ ) ⟩ , S h ( f ) = 4 π ⟨ h ~ P ( f ) h ~ P ∗ ( f ) ⟩ ‾ S_h(f) = \left\langle \tilde{h}_P(f, \hat{k})\tilde{h}^*_P(f, \hat{k}) \right\rangle, ~~~ S_h(f) = 4\pi \overline{\left\langle \tilde{h}_P(f)\tilde{h}^*_P(f) \right\rangle}
S h ( f ) = ⟨ h ~ P ( f , k ^ ) h ~ P ∗ ( f , k ^ ) ⟩ , S h ( f ) = 4 π ⟨ h ~ P ( f ) h ~ P ∗ ( f ) ⟩
⟨ n ~ ( f ) n ~ ∗ ( f ′ ) ⟩ T = N ( f ) δ ( f − f ′ ) \left\langle \tilde{n}(f)\tilde{n}^*(f')\right\rangle_T = N(f)\delta(f-f')
⟨ n ~ ( f ) n ~ ∗ ( f ′ ) ⟩ T = N ( f ) δ ( f − f ′ )
注意⟨ n ~ ( f ) n ~ ∗ ( f ) ⟩ = δ ( 0 ) N ( f ) → ∞ \left\langle \tilde{n}(f) \tilde{n}^*(f) \right\rangle = \delta(0) N(f) \rightarrow \infty ⟨ n ~ ( f ) n ~ ∗ ( f ) ⟩ = δ ( 0 ) N ( f ) → ∞ 趋向无穷大,而非等于N ( f ) N(f) N ( f ) !
⟨ n ( t ) n ( t ′ ) ⟩ ↔ ⟨ n ~ ( f ) n ~ ∗ ( f ′ ) ⟩ \boxed{\left\langle n(t) n(t') \right\rangle \leftrightarrow \left\langle \tilde{n}(f) \tilde{n}^*(f') \right\rangle}
⟨ n ( t ) n ( t ′ ) ⟩ ↔ ⟨ n ~ ( f ) n ~ ∗ ( f ′ ) ⟩
⟨ n ( t ) n ( t ′ ) ⟩ = ⟨ ∫ n ( f ) e i 2 π f t d f ∫ n ∗ ( f ′ ) e − i 2 π f ′ t ′ d f ′ ⟩ = ∫ d f ∫ d f ′ ⟨ n ( f ) n ∗ ( f ′ ) ⟩ e i 2 π f t e − i 2 π f ′ t ′ = ∫ d f ∫ d f ′ N ( f ) δ ( f − f ′ ) e i 2 π f t e − i 2 π f ′ t ′ = ∫ d f N ( f ) e i 2 π f ( t − t ′ ) \begin{aligned}
\left\langle n(t) n(t') \right\rangle &= \left\langle \int n(f) e^{i2\pi f t} df \int n^*(f') e^{-i2\pi f' t'} df' \right\rangle\\
&= \int df \int df'\left\langle n(f) n^*(f') \right\rangle e^{i2\pi f t} e^{-i2\pi f' t'}\\
&= \int df \int df'N(f)\delta(f-f') e^{i2\pi f t} e^{-i2\pi f' t'}\\
&= \int df N(f) e^{i2\pi f (t-t')}\\
\end{aligned} ⟨ n ( t ) n ( t ′ ) ⟩ = ⟨ ∫ n ( f ) e i 2 π f t d f ∫ n ∗ ( f ′ ) e − i 2 π f ′ t ′ d f ′ ⟩ = ∫ d f ∫ d f ′ ⟨ n ( f ) n ∗ ( f ′ ) ⟩ e i 2 π f t e − i 2 π f ′ t ′ = ∫ d f ∫ d f ′ N ( f ) δ ( f − f ′ ) e i 2 π f t e − i 2 π f ′ t ′ = ∫ d f N ( f ) e i 2 π f ( t − t ′ )
⟨ n ( t ) n ( t ′ ) ⟩ = r n ( t − t ′ ) = ∫ N ( f ) e i 2 π f ( t − t ′ ) d f \left\langle n(t) n(t') \right\rangle = r^n(t-t') = \int N(f) e^{i2\pi f(t-t')} df
⟨ n ( t ) n ( t ′ ) ⟩ = r n ( t − t ′ ) = ∫ N ( f ) e i 2 π f ( t − t ′ ) d f
⟨ n 2 ( t ) ⟩ = ⟨ n ( t ) n ( t ) ⟩ = r n ( 0 ) = ∫ N ( f ) d f \left\langle n^2(t) \right\rangle = \left\langle n(t) n(t) \right\rangle = r^n(0) = \int N(f) df
⟨ n 2 ( t ) ⟩ = ⟨ n ( t ) n ( t ) ⟩ = r n ( 0 ) = ∫ N ( f ) d f
平稳随机过程的系综平均等于时间平均
分析方法
单个探测器观测
如果对探测器噪声有明确的先验信息,则可通过功率超出得到随机背景的能谱S ( f ) = S d ( f ) − N ( f ) , S h ( f ) = S ( f ) R ( f ) S(f) = S_d(f) - N(f), S_h(f) = \frac{S(f)}{\mathcal{R}(f)} S ( f ) = S d ( f ) − N ( f ) , S h ( f ) = R ( f ) S ( f ) ,最初宇宙微波背景的发现就是因为观测到了不能被任何仪器噪声所解释的信号。但现实中,仪器噪声通常很难准确描述,因此这个方法并不可靠,尤其是在弱信号强噪声的情况,引力波背景的地面探测就是这种情况。不过对于空间探测,根据理论推算系内白矮星的随机背景信号在相应频段会显著超出仪器噪声。
另一种情况则是,对仪器噪声能有精确测量,引力波信号虽然难以屏蔽,但对于三角构型可以构建,压制引力波信号,被称为零通道(null channel),对称Sagnac组合就是这种情况。此时可以将零通道作为仪器噪声基准,通过不同通道噪声转换关系(转移函数)得到数据通道的噪声。S ( f ) = S d ( f ) − H ( f ) N ( f ) S(f) = S_d(f) - H(f)N(f) S ( f ) = S d ( f ) − H ( f ) N ( f )
对于随机背景而言,单个探测情况下,时间积分对于信噪比提升是没有帮助的。探测器运行后,如果能测到马上就会被测量,反之如果测不到,持续再长时间也测不到。但增加观测时长的好处在于可以提升频率的分辨率,从而对于频谱形状有更好的限制。
d ( t ) = s ( t ) + n ( t ) , ⟨ d 2 ( t ) ⟩ = ⟨ s 2 ( t ) ⟩ + ⟨ n 2 ( t ) ⟩ d(t) = s(t) + n(t), ~~~ \langle d^2(t) \rangle = \langle s^2(t) \rangle + \langle n^2(t)\rangle
d ( t ) = s ( t ) + n ( t ) , ⟨ d 2 ( t ) ⟩ = ⟨ s 2 ( t ) ⟩ + ⟨ n 2 ( t ) ⟩
⟨ s 2 ( t ) ⟩ = ∫ d f S ( f ) = ⟨ R ⟩ ∫ d f S h ( f ) ; ⟨ n 2 ( t ) ⟩ = ∫ d f N ( f ) \langle s^2(t) \rangle = \int df S(f) = \langle\mathcal{R}\rangle \int df S_h(f) ; ~~~ \langle n^2(t)\rangle = \int df N(f)
⟨ s 2 ( t ) ⟩ = ∫ d f S ( f ) = ⟨ R ⟩ ∫ d f S h ( f ) ; ⟨ n 2 ( t ) ⟩ = ∫ d f N ( f )
ρ 2 ≡ ( S N ) 2 = S ( f ) N ( f ) = ⟨ R ⟩ S h ( f ) N ( f ) \rho^2 \equiv \left(\frac{S}{N}\right)^2 = \frac{S(f)}{N(f)} = \langle \mathcal{R}\rangle \frac{S_h(f)}{N(f)}
ρ 2 ≡ ( N S ) 2 = N ( f ) S ( f ) = ⟨ R ⟩ N ( f ) S h ( f )
对于背景信号,无法使用匹配滤波,因为随机背景并没有确定的波形h ( t ) h(t) h ( t ) 。而如果有多个探测器,可以在不同探测器的信号之间进行匹配。
s I ( t ) = ∫ d f ∫ d k ^ D I a b h ~ a b ( f , k ^ ) e i 2 π f ( t − k ^ ⋅ r ⃗ I c ) s_I(t) = \int df \int d\hat{k} ~ D_I^{ab} \tilde{h}_{ab}(f, \hat{k}) e^{i 2\pi f (t-\hat{k}\cdot\frac{\vec{r}_I}{c})}
s I ( t ) = ∫ d f ∫ d k ^ D I a b h ~ a b ( f , k ^ ) e i 2 π f ( t − k ^ ⋅ c r I )
s ~ I ( f ) = ∫ d k ^ D I a b h ~ a b ( f , k ^ ) e − i 2 π f k ^ ⋅ r ⃗ I c \tilde{s}_I(f) = \int d\hat{k} ~ D_I^{ab} \tilde{h}_{ab}(f, \hat{k}) e^{-i 2\pi f \hat{k}\cdot\frac{\vec{r}_I}{c}}
s ~ I ( f ) = ∫ d k ^ D I a b h ~ a b ( f , k ^ ) e − i 2 π f k ^ ⋅ c r I
因为观测时间有限,傅里叶变换后出现sinc函数,但在关注的频段上可将其替换为δ \delta δ 函数????
Even on a relatively short stretch of data with, say, T = 103 s, at f = 10 Hz we have f T = 104. Over the whole useful bandwidth of ground-based detectors we can therefore replace sinc by a Dirac delta,
2个探测器联合观测
定义随机变量
S ≡ ∫ − T / 2 T / 2 d I ( t ) d J ( t ) d t \mathbb{S} \equiv \int_{-T/2}^{T/2} d_I(t)d_J(t) dt
S ≡ ∫ − T / 2 T / 2 d I ( t ) d J ( t ) d t
这里定义的S \mathbb{S} S ,相比互相关函数的零点值(未时间平移)r I J ( 0 ) r_{IJ}(0) r I J ( 0 ) ,少了取极限以及时间平均操作,是有限观测时长,未做时间平均的“互相关”,仍是随机变量。
ρ 2 = ( μ S σ S ) 2 = ⟨ S ⟩ 2 ⟨ S 2 ⟩ − ⟨ S ⟩ 2 \rho^2 = \left(\frac{\mu_{\mathbb{S}}}{\sigma_{\mathbb{S}}}\right)^2 = \frac{\langle\mathbb{S}\rangle^2}{\langle\mathbb{S}^2\rangle - \langle\mathbb{S}\rangle^2}
ρ 2 = ( σ S μ S ) 2 = ⟨ S 2 ⟩ − ⟨ S ⟩ 2 ⟨ S ⟩ 2
⟨ S ⟩ = ⟨ ∫ − T / 2 T / 2 d I ( t ) d J ( t ) d t ⟩ = ∫ − T / 2 T / 2 ⟨ s I ( t ) s J ( t ) ⟩ d t = T ⟨ s I ( t ) s J ( t ) ⟩ \langle\mathbb{S}\rangle = \left\langle \int_{-T/2}^{T/2} d_I(t)d_J(t) dt \right\rangle = \int_{-T/2}^{T/2} \big\langle s_I(t)s_J(t)\big\rangle dt = T \big\langle s_I(t)s_J(t)\big\rangle
⟨ S ⟩ = ⟨ ∫ − T / 2 T / 2 d I ( t ) d J ( t ) d t ⟩ = ∫ − T / 2 T / 2 ⟨ s I ( t ) s J ( t ) ⟩ d t = T ⟨ s I ( t ) s J ( t ) ⟩
⟨ s 1 ( t ) s 2 ( t ) ⟩ = r I J s ( 0 ) = ∫ d f S I J ( f ) \left\langle s_1(t)s_2(t)\right\rangle = r^s_{IJ}(0) = \int df S_{IJ}(f)
⟨ s 1 ( t ) s 2 ( t ) ⟩ = r I J s ( 0 ) = ∫ d f S I J ( f )
假设IJ方位指向完全一致(但噪声独立) S I J ( f ) = S ( f ) = ⟨ R ⟩ S h ( f ) = 2 5 S h ( f ) S_{IJ}(f) = S(f) = \langle \mathcal{R} \rangle S_h(f) = \frac{2}{5}S_h(f) S I J ( f ) = S ( f ) = ⟨ R ⟩ S h ( f ) = 5 2 S h ( f )
⟨ R ⟩ = ∑ P ⟨ F P F P ∗ ⟩ ‾ = ∑ P ∫ d ψ 2 π ∫ d 2 k ^ 4 π F P ( k ^ ; ψ ) F P ∗ ( k ^ ; ψ ) = 2 5 \langle \mathcal{R} \rangle = \sum_P \overline{\left\langle F^P {F^P}^*\right\rangle} = \sum_P \int \frac{d\psi}{2\pi} \int \frac{d^2\hat{k}}{4\pi}~ F^P(\hat{k}; \psi) {F^P}^*(\hat{k}; \psi) = \frac{2}{5}
⟨ R ⟩ = P ∑ ⟨ F P F P ∗ ⟩ = P ∑ ∫ 2 π d ψ ∫ 4 π d 2 k ^ F P ( k ^ ; ψ ) F P ∗ ( k ^ ; ψ ) = 5 2
∫ d ψ 2 π F + ( k ^ ; ψ ) = ∫ d ψ 2 π F × ( k ^ ; ψ ) ; ⟨ F + 2 ⟩ ‾ = ⟨ F + 2 ⟩ ‾ = 1 5 \int \frac{d\psi}{2\pi} F^+(\hat{k}; \psi) = \int \frac{d\psi}{2\pi} {F^\times}(\hat{k}; \psi); ~~~ \overline{\left\langle {F^+}^2\right\rangle} = \overline{\left\langle {F^+}^2\right\rangle} = \frac{1}{5}
∫ 2 π d ψ F + ( k ^ ; ψ ) = ∫ 2 π d ψ F × ( k ^ ; ψ ) ; ⟨ F + 2 ⟩ = ⟨ F + 2 ⟩ = 5 1
⟨ S 2 ⟩ = ⟨ ∫ − T / 2 T / 2 d t ∫ − T / 2 T / 2 d t ′ d I ( t ) d J ( t ) d I ( t ′ ) d J ( t ′ ) ⟩ \begin{aligned}
\langle\mathbb{S}^2\rangle &= \left\langle \int_{-T/2}^{T/2} dt \int_{-T/2}^{T/2} dt' ~ d_I(t)d_J(t)d_I(t')d_J(t') \right\rangle
\end{aligned} ⟨ S 2 ⟩ = ⟨ ∫ − T / 2 T / 2 d t ∫ − T / 2 T / 2 d t ′ d I ( t ) d J ( t ) d I ( t ′ ) d J ( t ′ ) ⟩
在弱信号假设下,可进行简化近似:
σ S 2 = ⟨ S 2 ⟩ − ⟨ S ⟩ 2 = ∫ − T / 2 T / 2 d t ∫ − T / 2 T / 2 d t ′ ⟨ d I ( t ) d J ( t ) d I ( t ′ ) d J ( t ′ ) ⟩ − ⟨ s I ( t ) s J ( t ) ⟩ ⟨ s I ( t ′ ) s J ( t ′ ) ⟩ ≈ ∫ − T / 2 T / 2 d t ∫ − T / 2 T / 2 d t ′ ⟨ n I ( t ) n I ( t ′ ) n J ( t ) n J ( t ′ ) ⟩ = ∫ − T / 2 T / 2 d t ∫ − T / 2 T / 2 d t ′ ⟨ n I ( t ) n I ( t ′ ) ⟩ ⟨ n J ( t ) n J ( t ′ ) ⟩ \begin{aligned}
\sigma^2_\mathbb{S} &= \langle\mathbb{S}^2\rangle - \langle\mathbb{S}\rangle^2 \\
&= \int_{-T/2}^{T/2} dt \int_{-T/2}^{T/2} dt' ~ \big\langle d_I(t)d_J(t)d_I(t')d_J(t') \big\rangle - \big\langle s_I(t)s_J(t)\big\rangle\big\langle s_I(t')s_J(t')\big\rangle\\
&\approx \int_{-T/2}^{T/2} dt \int_{-T/2}^{T/2} dt' ~\big\langle n_I(t)n_I(t') n_J(t)n_J(t') \big\rangle\\
&= \int_{-T/2}^{T/2} dt \int_{-T/2}^{T/2} dt' ~\big\langle n_I(t)n_I(t')\big\rangle \big\langle n_J(t)n_J(t') \big\rangle\\
\end{aligned} σ S 2 = ⟨ S 2 ⟩ − ⟨ S ⟩ 2 = ∫ − T / 2 T / 2 d t ∫ − T / 2 T / 2 d t ′ ⟨ d I ( t ) d J ( t ) d I ( t ′ ) d J ( t ′ ) ⟩ − ⟨ s I ( t ) s J ( t ) ⟩ ⟨ s I ( t ′ ) s J ( t ′ ) ⟩ ≈ ∫ − T / 2 T / 2 d t ∫ − T / 2 T / 2 d t ′ ⟨ n I ( t ) n I ( t ′ ) n J ( t ) n J ( t ′ ) ⟩ = ∫ − T / 2 T / 2 d t ∫ − T / 2 T / 2 d t ′ ⟨ n I ( t ) n I ( t ′ ) ⟩ ⟨ n J ( t ) n J ( t ′ ) ⟩
⟨ n I ( t ) n I ( t ′ ) ⟩ = r n ( t − t ′ ) = ∫ d f e i 2 π f ( t − t ′ ) N I ( f ) \begin{aligned}
\big\langle n_I(t)n_I(t')\big\rangle = r^n(t-t') = \int df e^{i2\pi f(t-t')} N_I(f)\\
\end{aligned} ⟨ n I ( t ) n I ( t ′ ) ⟩ = r n ( t − t ′ ) = ∫ d f e i 2 π f ( t − t ′ ) N I ( f )
σ S 2 ≈ ∫ − T / 2 T / 2 d t ∫ − T / 2 T / 2 d t ′ ⟨ n I ( t ) n I ( t ′ ) ⟩ ⟨ n J ( t ) n J ( t ′ ) ⟩ = ∫ − T / 2 T / 2 d t ∫ − T / 2 T / 2 d t ′ ⟨ n I ( t ) n I ( t ′ ) ⟩ ⟨ n J ( t ) n J ( t ′ ) ⟩ ∗ = ∫ − T / 2 T / 2 d t ∫ − T / 2 T / 2 d t ′ ∫ d f ∫ d f ′ e i 2 π f ( t − t ′ ) e − i 2 π f ′ ( t − t ′ ) N I ( f ) N J ∗ ( f ′ ) = ∫ − T / 2 T / 2 d t ∫ − T / 2 T / 2 d t ′ ∫ d f ∫ d f ′ e i 2 π f ( t − t ′ ) e − i 2 π f ′ ( t − t ′ ) N I ( f ) N J ( f ′ ) = ∫ − T / 2 T / 2 d t ∫ − T / 2 T / 2 d t ′ ∫ d f ∫ d f ′ N I ( f ) N J ( f ′ ) ? ? ? \begin{aligned}
\sigma^2_\mathbb{S} &\approx \int_{-T/2}^{T/2} dt \int_{-T/2}^{T/2} dt' ~\big\langle n_I(t)n_I(t')\big\rangle \big\langle n_J(t)n_J(t') \big\rangle\\
&= \int_{-T/2}^{T/2} dt \int_{-T/2}^{T/2} dt' ~\big\langle n_I(t)n_I(t')\big\rangle \big\langle n_J(t)n_J(t')\big\rangle^*\\
&= \int_{-T/2}^{T/2} dt \int_{-T/2}^{T/2} dt' \int df \int df' e^{i2\pi f(t-t')} e^{-i2\pi f'(t-t')} N_I(f) N^*_J(f')\\
&= \int_{-T/2}^{T/2} dt \int_{-T/2}^{T/2} dt' \int df \int df' e^{i2\pi f(t-t')} e^{-i2\pi f'(t-t')} N_I(f) N_J(f')\\
&= \int_{-T/2}^{T/2} dt \int_{-T/2}^{T/2} dt' \int df \int df' N_I(f) N_J(f')???\\
\end{aligned} σ S 2 ≈ ∫ − T / 2 T / 2 d t ∫ − T / 2 T / 2 d t ′ ⟨ n I ( t ) n I ( t ′ ) ⟩ ⟨ n J ( t ) n J ( t ′ ) ⟩ = ∫ − T / 2 T / 2 d t ∫ − T / 2 T / 2 d t ′ ⟨ n I ( t ) n I ( t ′ ) ⟩ ⟨ n J ( t ) n J ( t ′ ) ⟩ ∗ = ∫ − T / 2 T / 2 d t ∫ − T / 2 T / 2 d t ′ ∫ d f ∫ d f ′ e i 2 π f ( t − t ′ ) e − i 2 π f ′ ( t − t ′ ) N I ( f ) N J ∗ ( f ′ ) = ∫ − T / 2 T / 2 d t ∫ − T / 2 T / 2 d t ′ ∫ d f ∫ d f ′ e i 2 π f ( t − t ′ ) e − i 2 π f ′ ( t − t ′ ) N I ( f ) N J ( f ′ ) = ∫ − T / 2 T / 2 d t ∫ − T / 2 T / 2 d t ′ ∫ d f ∫ d f ′ N I ( f ) N J ( f ′ ) ? ? ?
这里利用了n J ( t ) n_J(t) n J ( t ) 和N J ( f ) N_J(f) N J ( f ) 均为实函数,消去了指数项。
σ S 2 ≈ ∫ d f ∫ d f ′ δ 2 ( f − f ′ ) N I ( f ) N J ( f ′ ) ? ? ? \begin{aligned}
\sigma^2_\mathbb{S} &\approx \int df \int df' \delta^2(f-f') N_I(f) N_J(f')???\\
\end{aligned} σ S 2 ≈ ∫ d f ∫ d f ′ δ 2 ( f − f ′ ) N I ( f ) N J ( f ′ ) ? ? ?
σ S 2 ≈ T ∫ d f N I ( f ) N J ( f ) \begin{aligned}
\sigma^2_\mathbb{S} &\approx T \int df N_I(f) N_J(f)\\
\end{aligned} σ S 2 ≈ T ∫ d f N I ( f ) N J ( f )
ρ I J = μ S σ S = T ∫ d f S I J ( f ) ∫ d f N I ( f ) N J ( f ) = ⟨ R ⟩ T ∫ d f S h ( f ) ∫ d f N I ( f ) N J ( f ) \rho_{\tiny IJ} =\frac{\mu_{\mathbb{S}}}{\sigma_{\mathbb{S}}} =\sqrt{T} \frac{ \displaystyle \int df S_{IJ}(f) }{\displaystyle \sqrt{\int df N_I(f) N_J(f) } } = \langle\mathcal{R}\rangle\sqrt{T} \frac{ \displaystyle \int df S_h(f) }{\displaystyle \sqrt{\int df N_I(f) N_J(f) } }
ρ I J = σ S μ S = T ∫ d f N I ( f ) N J ( f ) ∫ d f S I J ( f ) = ⟨ R ⟩ T ∫ d f N I ( f ) N J ( f ) ∫ d f S h ( f )
这里假设了探测器位置指向相同,即重叠约化函数γ I J ( f ) = 1 \gamma_{IJ}(f) =1 γ I J ( f ) = 1 。当探测器位置或指向不同时需要考虑重叠约化因子。
而且这个信噪比也并非γ I J ( f ) = 1 \gamma_{IJ}(f) =1 γ I J ( f ) = 1 时的最佳信噪比,最佳滤波 :
S ≡ ∫ − T / 2 T / 2 d t ∫ − T / 2 T / 2 d t ′ d I ( t ) d J ( t ′ ) Q ( t − t ′ ) \mathbb{S} \equiv \int_{-T/2}^{T/2}dt \int_{-T/2}^{T/2} dt' ~ d_I(t)d_J(t')Q(t-t')
S ≡ ∫ − T / 2 T / 2 d t ∫ − T / 2 T / 2 d t ′ d I ( t ) d J ( t ′ ) Q ( t − t ′ )
S = ∫ d f ∫ d f ′ δ T ( f − f ′ ) d ~ I ( f ) d ~ J ∗ ( f ′ ) Q ~ ( f ) \mathbb{S} = \int df \int df' ~ \delta_T(f-f') \tilde{d}_I(f) \tilde{d}^*_J(f')\tilde{Q}(f)
S = ∫ d f ∫ d f ′ δ T ( f − f ′ ) d ~ I ( f ) d ~ J ∗ ( f ′ ) Q ~ ( f )
S = ∫ d f ∫ d f ′ δ T ( f − f ′ ) d ~ I ( f ) d ~ J ∗ ( f ′ ) Q ~ ( f ′ ) \mathbb{S} = \int df \int df' ~ \delta_T(f-f') \tilde{d}_I(f) \tilde{d}^*_J(f')\tilde{Q}(f')
S = ∫ d f ∫ d f ′ δ T ( f − f ′ ) d ~ I ( f ) d ~ J ∗ ( f ′ ) Q ~ ( f ′ )
S = ∫ d f ∫ d f ′ δ T ( f − f ′ ) d ~ I ( f ) d ~ J ∗ ( f ′ ) Q ~ ∗ ( f ′ ) \mathbb{S} = \int df \int df' ~ \delta_T(f-f') \tilde{d}_I(f) \tilde{d}^*_J(f')\tilde{Q}^*(f')
S = ∫ d f ∫ d f ′ δ T ( f − f ′ ) d ~ I ( f ) d ~ J ∗ ( f ′ ) Q ~ ∗ ( f ′ )
Q ( t − t ′ ) Q(t-t') Q ( t − t ′ ) 为对称实函数, Q ~ ( − f ) = Q ~ ∗ ( f ) \tilde{Q}(-f) = \tilde{Q}^*(f) Q ~ ( − f ) = Q ~ ∗ ( f ) 。
μ S = ⟨ S ⟩ = ∫ d f ∫ d f ′ δ T ( f − f ′ ) ⟨ s ~ I ( f ) s ~ J ∗ ( f ′ ) ⟩ Q ~ ( f ) = ⟨ R ⟩ T ∫ d f S h ( f ) Q ~ ( f ) \mu_\mathbb{S} = \langle \mathbb{S} \rangle = \int df \int df' ~ \delta_T(f-f') \big\langle\tilde{s}_I(f) \tilde{s}^*_J(f')\big\rangle \tilde{Q}(f) = \langle\mathcal{R}\rangle T \int df ~ S_h(f) \tilde{Q}(f)
μ S = ⟨ S ⟩ = ∫ d f ∫ d f ′ δ T ( f − f ′ ) ⟨ s ~ I ( f ) s ~ J ∗ ( f ′ ) ⟩ Q ~ ( f ) = ⟨ R ⟩ T ∫ d f S h ( f ) Q ~ ( f )
σ S 2 = T ∫ d f N I ( f ) N J ( f ) Q ~ ( f ) Q ~ ∗ ( f ) \sigma^2_\mathbb{S} = T \int df N_I(f) N_J(f) \tilde{Q}(f)\tilde{Q}^*(f)
σ S 2 = T ∫ d f N I ( f ) N J ( f ) Q ~ ( f ) Q ~ ∗ ( f )
ρ I J 2 = μ S 2 σ S 2 = ⟨ R ⟩ 2 T [ ∫ d f S h ( f ) Q ~ ∗ ( f ) ] 2 ∫ d f N I ( f ) N J ( f ) Q ~ ( f ) Q ~ ∗ ( f ) \rho_{\tiny IJ}^2 = \frac{ \mu^2_\mathbb{S}}{\sigma^2_\mathbb{S}} = \langle\mathcal{R}\rangle^2 T \frac{ \left[\displaystyle \int df ~ S_h(f) \tilde{Q}^*(f) \right]^2 }{ \displaystyle \int df N_I(f) N_J(f) \tilde{Q}(f)\tilde{Q}^*(f) }
ρ I J 2 = σ S 2 μ S 2 = ⟨ R ⟩ 2 T ∫ d f N I ( f ) N J ( f ) Q ~ ( f ) Q ~ ∗ ( f ) [ ∫ d f S h ( f ) Q ~ ∗ ( f ) ] 2
柯西-施瓦兹不等式或者定义内积
⟨ A ∣ B ⟩ = ∫ d f A ∗ ( f ) B ( f ) N I ( f ) N J ( f ) \langle A|B\rangle = \int df A^*(f) B(f) N_I(f) N_J(f)
⟨ A ∣ B ⟩ = ∫ d f A ∗ ( f ) B ( f ) N I ( f ) N J ( f )
ρ I J 2 = ⟨ R ⟩ 2 T ⟨ Q ~ ∣ S h ( f ) N I ( f ) N J ( f ) ⟩ 2 ⟨ Q ~ ∣ Q ~ ⟩ \rho_{\tiny IJ}^2 = \langle\mathcal{R}\rangle^2 T \frac{ \left\langle \tilde{Q}\Big|\frac{S_h(f)}{ N_I(f) N_J(f)} \right\rangle^2 }{ \left\langle \tilde{Q}|\tilde{Q}\right\rangle }
ρ I J 2 = ⟨ R ⟩ 2 T ⟨ Q ~ ∣ Q ~ ⟩ ⟨ Q ~ ∣ ∣ ∣ ∣ N I ( f ) N J ( f ) S h ( f ) ⟩ 2
分子是Q ~ \tilde{Q} Q ~ 矢量模方,显然当Q ~ \tilde{Q} Q ~ 与S h ( f ) N I ( f ) N J ( f ) \frac{S_h(f)}{N_I(f)N_J(f)} N I ( f ) N J ( f ) S h ( f ) 共线时信噪比取最大:
Q ~ ( f ) = q S h ( f ) N I ( f ) N J ( f ) \tilde{Q}(f) = q \frac{S_h(f)}{N_I(f)N_J(f)}
Q ~ ( f ) = q N I ( f ) N J ( f ) S h ( f )
ρ I J 2 = ⟨ R ⟩ 2 T ⟨ S h ( f ) N I ( f ) N J ( f ) ∣ S h ( f ) N I ( f ) N J ( f ) ⟩ \rho_{\tiny IJ}^2 = \langle\mathcal{R}\rangle^2 T \left\langle \frac{S_h(f)}{ N_I(f) N_J(f)} \Big|\frac{S_h(f)}{ N_I(f) N_J(f)} \right\rangle
ρ I J 2 = ⟨ R ⟩ 2 T ⟨ N I ( f ) N J ( f ) S h ( f ) ∣ ∣ ∣ ∣ N I ( f ) N J ( f ) S h ( f ) ⟩
ρ I J = ⟨ R ⟩ T ∫ d f S h 2 ( f ) N I ( f ) N J ( f ) \rho_{\tiny IJ} = \langle \mathcal{R} \rangle \sqrt{T} \sqrt{\int df \frac{ S^2_h(f)}{N_I(f) N_J(f)} }
ρ I J = ⟨ R ⟩ T ∫ d f N I ( f ) N J ( f ) S h 2 ( f )
需要注意的是,这里得到的信噪比只是理论最优值,其中最佳滤波Q ~ ( f ) \tilde{Q}(f) Q ~ ( f ) 依赖于仪器噪声谱N ( f ) N(f) N ( f ) 及背景信号谱S h ( f ) S_h(f) S h ( f ) ,而S h ( f ) S_h(f) S h ( f ) 或者说Ω G W \Omega_{\rm \tiny GW} Ω G W 是预先未知的,因此需要额外先验信息。天体物理波源背景通常预期为幂率谱,比如致密双星系统叠加的背景谱Ω G W ∝ f 2 / 3 \Omega_{\rm \tiny GW}\propto f^{2/3} Ω G W ∝ f 2 / 3 ,旋转中子星Ω G W ∝ f 2 \Omega_{\rm \tiny GW}\propto f^{2} Ω G W ∝ f 2 。实际处理数据时可以尝试不同幂指数,搜索最高信噪比。对于宇宙学背景,理论预言上除了源自暴涨的平谱,还有早期宇宙一阶相变的分段幂率谱,以及其他物理机制产生的具有特征频率的单峰背景信号,各种信号叠加在一起,问题会变得复杂。
最初的结果相当于取滤波函数Q ( t − t ′ ) = δ ( t − t ′ ) ; Q ~ ( f ) = 1 Q(t-t')=\delta(t-t'); \tilde{Q}(f) = 1 Q ( t − t ′ ) = δ ( t − t ′ ) ; Q ~ ( f ) = 1 。如果假设背景噪声分数密度谱为常数Ω G W ( f ) = Ω 0 \Omega_{\rm \tiny GW}(f) = \Omega_0 Ω G W ( f ) = Ω 0 ,Q ~ ( f ) ∝ Ω 0 f − 3 N I ( f ) N J ( f ) \tilde{Q}(f) \propto \frac{\Omega_0 f^{-3}}{N_I(f)N_J(f)} Q ~ ( f ) ∝ N I ( f ) N J ( f ) Ω 0 f − 3
频域函数形状(绘图)
时域函数形状(绘图)
与单探测器相比:假设N I ( f ) = N J ( f ) N_I(f) = N_J(f) N I ( f ) = N J ( f ) ,且噪声及信号功率谱在目标频段范围内与频率无关,从而
ρ I J = ⟨ R ⟩ T ∫ d f S h 2 ( f ) N I ( f ) N J ( f ) ∼ ⟨ R ⟩ T S h ( f ) N ( f ) ∫ b a n d d f = ⟨ R ⟩ T 2 Δ f S h ( f ) N ( f ) \rho_{\tiny IJ} = \langle \mathcal{R} \rangle \sqrt{T} \sqrt{\int df \frac{ S^2_h(f)}{N_I(f) N_J(f)} } \sim \langle \mathcal{R} \rangle \sqrt{T} \frac{ S_h(f)}{N(f)} \sqrt{\int_{\rm band} df} = \langle \mathcal{R} \rangle \sqrt{T 2\Delta f} \frac{S_h(f)}{N(f)}
ρ I J = ⟨ R ⟩ T ∫ d f N I ( f ) N J ( f ) S h 2 ( f ) ∼ ⟨ R ⟩ T N ( f ) S h ( f ) ∫ b a n d d f = ⟨ R ⟩ T 2 Δ f N ( f ) S h ( f )
对比单探测器情况,相差T 2 Δ f ∼ 8 × 1 0 3 T 1 y r Δ f 1 H z \sqrt{T 2\Delta f} \sim 8\times10^3 \sqrt{\frac{T}{1 \rm yr}} \sqrt{\frac{\Delta f}{1 \rm Hz}} T 2 Δ f ∼ 8 × 1 0 3 1 y r T 1 H z Δ f 倍,其中Δ f \Delta f Δ f 为探测器的(单边)频带宽度。对于空间探测器而言Δ f ∼ 1 H z \Delta f \sim 1 \rm Hz Δ f ∼ 1 H z ,一年积分时间信噪比将提升约四个量级。
这意味着时间小于带宽分之一的情况下,互相关的信噪比反而不如单个探测器,该如何理解???
对于地面探测器,(单边)带宽约150Hz,对应时间1/300秒,对于秒量级的双黑洞信号并和信号,互相关似乎并不降低信噪比,为什么不做相关??如果考虑到重叠约化??
与通常的匹配滤波作对比
最后,考虑探测器位置或指向不同时的重叠约化函数 。
S = ∫ d f ∫ d f ′ δ T ( f − f ′ ) d ~ I ( f ) d ~ J ∗ ( f ′ ) Q ~ ∗ ( f ′ ) \mathbb{S} = \int df \int df' ~ \delta_T(f-f') \tilde{d}_I(f) \tilde{d}^*_J(f')\tilde{Q}^*(f')
S = ∫ d f ∫ d f ′ δ T ( f − f ′ ) d ~ I ( f ) d ~ J ∗ ( f ′ ) Q ~ ∗ ( f ′ )
⟨ S ⟩ = ∫ d f ∫ d f ′ δ T ( f − f ′ ) ⟨ s ~ I ( f ) s ~ J ∗ ( f ′ ) ⟩ Q ~ ∗ ( f ′ ) \langle \mathbb{S} \rangle = \int df \int df' ~ \delta_T(f-f') \big\langle \tilde{s}_I(f) \tilde{s}^*_J(f')\big\rangle \tilde{Q}^*(f')
⟨ S ⟩ = ∫ d f ∫ d f ′ δ T ( f − f ′ ) ⟨ s ~ I ( f ) s ~ J ∗ ( f ′ ) ⟩ Q ~ ∗ ( f ′ )
s ~ I ( f ) = ∫ d 2 k ^ D I a b h ~ a b ( f , k ^ ) e − i 2 π f k ^ ⋅ r ⃗ I c = ∑ P ∫ d 2 k ^ F I P ( k ^ ) h ~ P ( f , k ^ ) e − i 2 π f k ^ ⋅ r ⃗ I c \tilde{s}_I(f) = \int d^2\hat{k} ~ D_I^{ab} \tilde{h}_{ab}(f, \hat{k}) e^{-i2\pi f \hat{k}\cdot \frac{\vec{r}_I}{c}} = \sum_P \int d^2\hat{k} ~ F^P_I(\hat{k}) \tilde{h}_P(f, \hat{k}) e^{-i2\pi f \hat{k}\cdot \frac{\vec{r}_I}{c}}
s ~ I ( f ) = ∫ d 2 k ^ D I a b h ~ a b ( f , k ^ ) e − i 2 π f k ^ ⋅ c r I = P ∑ ∫ d 2 k ^ F I P ( k ^ ) h ~ P ( f , k ^ ) e − i 2 π f k ^ ⋅ c r I
⟨ s ~ I ( f ) s ~ J ∗ ( f ′ ) ⟩ = ∑ P ∑ P ′ ∫ d 2 k ^ ∫ d 2 k ^ ′ F I P ( k ^ ) F J P ′ ( k ^ ′ ) ⟨ h ~ P ( f , k ^ ) h ~ P ′ ∗ ( f ′ , k ^ ′ ) ⟩ e − i 2 π f k ^ ⋅ r ⃗ I − r ⃗ J c = 1 4 π δ ( f − f ′ ) S h ( f ) ∑ P ∫ d 2 k ^ F I P ( k ^ ) F J P ( k ^ ) e − i 2 π f k ^ ⋅ r ⃗ I − r ⃗ J c = δ ( f − f ′ ) S h ( f ) Γ I J ( f ) \begin{aligned}
\big\langle \tilde{s}_I(f) \tilde{s}^*_J(f')\big\rangle &= \small \sum_P \sum_{P'} \int d^2\hat{k}\int d^2\hat{k}' ~ F^P_I(\hat{k})F^{P'}_J(\hat{k}') \big\langle \tilde{h}_P(f, \hat{k})\tilde{h}^*_{P'}(f', \hat{k}') \big\rangle e^{-i2\pi f \hat{k}\cdot \frac{\vec{r}_I-\vec{r}_J}{c}}\\
&= \frac{1}{4\pi}\delta(f-f')S_h(f) \sum_P \int d^2\hat{k} ~ F^P_I(\hat{k})F^P_J(\hat{k}) e^{-i2\pi f \hat{k}\cdot \frac{\vec{r}_I-\vec{r}_J}{c}}\\
&= \delta(f-f')S_h(f) \Gamma_{IJ}(f)\\
\end{aligned} ⟨ s ~ I ( f ) s ~ J ∗ ( f ′ ) ⟩ = P ∑ P ′ ∑ ∫ d 2 k ^ ∫ d 2 k ^ ′ F I P ( k ^ ) F J P ′ ( k ^ ′ ) ⟨ h ~ P ( f , k ^ ) h ~ P ′ ∗ ( f ′ , k ^ ′ ) ⟩ e − i 2 π f k ^ ⋅ c r I − r J = 4 π 1 δ ( f − f ′ ) S h ( f ) P ∑ ∫ d 2 k ^ F I P ( k ^ ) F J P ( k ^ ) e − i 2 π f k ^ ⋅ c r I − r J = δ ( f − f ′ ) S h ( f ) Γ I J ( f )
Γ I J ( f ) = ∑ P ∫ d 2 k ^ 4 π ∫ d ψ 2 π F I P ( k ^ ) F J P ( k ^ ) e − i 2 π f k ^ ⋅ r ⃗ I − r ⃗ J c ; γ I J ( f ) = Γ I J ( f ) Γ I J ( f ) ∣ a l i g n e d \boxed{\Gamma_{IJ}(f) = \sum_P \int \frac{d^2\hat{k}}{4\pi} \int \frac{d\psi}{2\pi} ~ F^P_I(\hat{k})F^P_J(\hat{k}) e^{-i2\pi f \hat{k}\cdot \frac{\vec{r}_I-\vec{r}_J}{c}}; ~~~ \gamma_{IJ}(f) = \frac{\Gamma_{IJ}(f)}{\Gamma_{IJ}(f)|_{\rm aligned}}}
Γ I J ( f ) = P ∑ ∫ 4 π d 2 k ^ ∫ 2 π d ψ F I P ( k ^ ) F J P ( k ^ ) e − i 2 π f k ^ ⋅ c r I − r J ; γ I J ( f ) = Γ I J ( f ) ∣ a l i g n e d Γ I J ( f )
Γ I J ( f ) ∣ a l i g n e d = ∑ P ∫ d 2 k ^ 4 π ∫ d ψ 2 π F I P ( k ^ ) F J P ( k ^ ) ∣ a l i g n e d = ⟨ R ⟩ = 2 5 \Gamma_{IJ}(f)\big|_{\rm aligned} = \sum_P \int \frac{d^2\hat{k}}{4\pi} \int \frac{d\psi}{2\pi} ~ F^P_I(\hat{k})F^P_J(\hat{k})\big|_{\rm aligned} = \langle \mathcal{R}\rangle = \frac{2}{5}
Γ I J ( f ) ∣ ∣ ∣ a l i g n e d = P ∑ ∫ 4 π d 2 k ^ ∫ 2 π d ψ F I P ( k ^ ) F J P ( k ^ ) ∣ ∣ ∣ a l i g n e d = ⟨ R ⟩ = 5 2
注意这里⟨ R ⟩ \langle \mathcal{R}\rangle ⟨ R ⟩ 为直角迈克逊干涉对全天方位的平均响应,对于非直角情况为2 5 sin 2 Υ \frac{2}{5}\sin^2\Upsilon 5 2 sin 2 Υ ,Υ \Upsilon Υ 为干涉臂张角。此外,这里Γ I J ( f ) \Gamma_{IJ}(f) Γ I J ( f ) 定义中包含了对方位k ^ \hat{k} k ^ 的全天平均1 4 π \frac{1}{4\pi} 4 π 1 ,因此归一化因此 Γ I J ( f ) ∣ a l i g n e d = 2 5 \Gamma_{IJ}(f)|_{\rm aligned} = \frac{2}{5} Γ I J ( f ) ∣ a l i g n e d = 5 2 。文献中另一种常见形式是对全天积分而未进行平均,此时归一化因子 Γ I J ( f ) ∣ a l i g n e d = 8 π 5 \Gamma_{IJ}(f)|_{\rm aligned} = \frac{8\pi}{5} Γ I J ( f ) ∣ a l i g n e d = 5 8 π
⟨ S ⟩ = ∫ d f ∫ d f ′ δ T ( f − f ′ ) ⟨ s ~ I ( f ) s ~ J ∗ ( f ′ ) ⟩ Q ~ ∗ ( f ′ ) = ∫ d f ∫ d f ′ δ T ( f − f ′ ) δ ( f − f ′ ) S h ( f ) ⟨ R ⟩ γ I J ( f ) = ⟨ R ⟩ T ∫ d f S h ( f ) γ I J ( f ) \begin{aligned}
\langle \mathbb{S} \rangle &= \int df \int df' ~ \delta_T(f-f') \big\langle \tilde{s}_I(f) \tilde{s}^*_J(f')\big\rangle \tilde{Q}^*(f')\\
& = \int df \int df' ~ \delta_T(f-f') \delta(f-f') S_h(f) \langle\mathcal{R}\rangle \gamma_{IJ}(f)\\
&= \langle\mathcal{R}\rangle T \int df ~ S_h(f) \gamma_{IJ}(f)
\end{aligned} ⟨ S ⟩ = ∫ d f ∫ d f ′ δ T ( f − f ′ ) ⟨ s ~ I ( f ) s ~ J ∗ ( f ′ ) ⟩ Q ~ ∗ ( f ′ ) = ∫ d f ∫ d f ′ δ T ( f − f ′ ) δ ( f − f ′ ) S h ( f ) ⟨ R ⟩ γ I J ( f ) = ⟨ R ⟩ T ∫ d f S h ( f ) γ I J ( f )
σ ⟨ S ⟩ 2 = ⟨ S 2 ⟩ − ⟨ S ⟩ 2 ≈ ∫ d f ∫ d f ′ ∫ d f 1 ∫ d f 1 ′ δ T ( f − f ′ ) δ T ( f 1 − f 1 ′ ) × ⟨ n ~ I ( f ) n ~ J ∗ ( f ′ ) n ~ I ( f 1 ) n ~ J ∗ ( f 1 ′ ) ⟩ Q ~ ( f ′ ) Q ~ ( f 1 ′ ) = ∫ d f ∫ d f ′ ∫ d f 1 ∫ d f 1 ′ δ T ( f − f ′ ) δ T ( f 1 − f 1 ′ ) × ⟨ n ~ I ( f ) n ~ I ( f 1 ) ⟩ ⟨ n ~ J ∗ ( f ) n ~ J ∗ ( f 1 ′ ) ⟩ Q ~ ( f ′ ) Q ~ ( f 1 ′ ) = ∫ d f ∫ d f ′ ∫ d f 1 ∫ d f 1 ′ δ T ( f − f ′ ) δ T ( f 1 − f 1 ′ ) × ⟨ n ~ I ( f ) n ~ I ∗ ( − f 1 ) ⟩ ⟨ n ~ J ∗ ( f ′ ) n ~ J ( − f 1 ′ ) ⟩ Q ~ ( f ′ ) Q ~ ( f 1 ′ ) = ∫ d f ∫ d f ′ ∫ d f 1 ∫ d f 1 ′ δ T ( f − f ′ ) δ T ( f 1 − f 1 ′ ) × δ ( f + f 1 ) N I ( f ) δ ( f ′ + f 1 ′ ) N J ( f ′ ) Q ~ ( f ′ ) Q ~ ( f 1 ′ ) = ∫ d f ∫ d f ′ δ T ( f − f ′ ) δ T ( f ′ − f ) N I ( f ) N J ( f ′ ) Q ~ ( f ′ ) Q ~ ( − f ′ ) = ∫ d f ∫ d f ′ δ T ( f − f ′ ) δ ( f − f ′ ) N I ( f ) N J ( f ′ ) Q ~ ( f ′ ) Q ~ ∗ ( f ′ ) = T ∫ d f N I ( f ) N J ( f ) Q ~ ( f ) Q ~ ∗ ( f ) \begin{aligned}
\sigma^2_{\langle \mathbb{S} \rangle} &= \langle \mathbb{S}^2 \rangle - \langle \mathbb{S} \rangle^2\\
&\approx \int df \int df' \int df_1 \int df'_1 ~ \delta_T(f-f') \delta_T(f_1-f'_1)\\
& ~~~~~~~~ \times \big\langle \tilde{n}_I(f) \tilde{n}^*_J(f') \tilde{n}_I(f_1) \tilde{n}^*_J(f'_1)\big\rangle \tilde{Q}(f')\tilde{Q}(f'_1)\\
&= \int df \int df' \int df_1 \int df'_1 ~ \delta_T(f-f') \delta_T(f_1-f'_1)\\
& ~~~~~~~~ \times \big\langle \tilde{n}_I(f) \tilde{n}_I(f_1) \big\rangle \big\langle\tilde{n}^*_J(f) \tilde{n}^*_J(f'_1)\big\rangle \tilde{Q}(f')\tilde{Q}(f'_1)\\
&= \int df \int df' \int df_1 \int df'_1 ~ \delta_T(f-f') \delta_T(f_1-f'_1)\\
& ~~~~~~~~ \times \big\langle \tilde{n}_I(f) \tilde{n}^*_I(-f_1) \big\rangle \big\langle\tilde{n}^*_J(f') \tilde{n}_J(-f'_1)\big\rangle \tilde{Q}(f')\tilde{Q}(f'_1)\\
&= \int df \int df' \int df_1 \int df'_1 ~ \delta_T(f-f') \delta_T(f_1-f'_1)\\
& ~~~~~~~~ \times \delta(f+f_1)N_I(f) \delta(f'+f'_1)N_J(f') \tilde{Q}(f')\tilde{Q}(f'_1)\\
&= \int df \int df' ~ \delta_T(f-f') \delta_T(f'-f)~ N_I(f) N_J(f') \tilde{Q}(f')\tilde{Q}(-f')\\
&= \int df \int df' ~ \delta_T(f-f') \delta(f-f')~ N_I(f) N_J(f') \tilde{Q}(f')\tilde{Q}^*(f')\\
&= T \int df N_I(f) N_J(f) \tilde{Q}(f)\tilde{Q}^*(f)\\
\end{aligned} σ ⟨ S ⟩ 2 = ⟨ S 2 ⟩ − ⟨ S ⟩ 2 ≈ ∫ d f ∫ d f ′ ∫ d f 1 ∫ d f 1 ′ δ T ( f − f ′ ) δ T ( f 1 − f 1 ′ ) × ⟨ n ~ I ( f ) n ~ J ∗ ( f ′ ) n ~ I ( f 1 ) n ~ J ∗ ( f 1 ′ ) ⟩ Q ~ ( f ′ ) Q ~ ( f 1 ′ ) = ∫ d f ∫ d f ′ ∫ d f 1 ∫ d f 1 ′ δ T ( f − f ′ ) δ T ( f 1 − f 1 ′ ) × ⟨ n ~ I ( f ) n ~ I ( f 1 ) ⟩ ⟨ n ~ J ∗ ( f ) n ~ J ∗ ( f 1 ′ ) ⟩ Q ~ ( f ′ ) Q ~ ( f 1 ′ ) = ∫ d f ∫ d f ′ ∫ d f 1 ∫ d f 1 ′ δ T ( f − f ′ ) δ T ( f 1 − f 1 ′ ) × ⟨ n ~ I ( f ) n ~ I ∗ ( − f 1 ) ⟩ ⟨ n ~ J ∗ ( f ′ ) n ~ J ( − f 1 ′ ) ⟩ Q ~ ( f ′ ) Q ~ ( f 1 ′ ) = ∫ d f ∫ d f ′ ∫ d f 1 ∫ d f 1 ′ δ T ( f − f ′ ) δ T ( f 1 − f 1 ′ ) × δ ( f + f 1 ) N I ( f ) δ ( f ′ + f 1 ′ ) N J ( f ′ ) Q ~ ( f ′ ) Q ~ ( f 1 ′ ) = ∫ d f ∫ d f ′ δ T ( f − f ′ ) δ T ( f ′ − f ) N I ( f ) N J ( f ′ ) Q ~ ( f ′ ) Q ~ ( − f ′ ) = ∫ d f ∫ d f ′ δ T ( f − f ′ ) δ ( f − f ′ ) N I ( f ) N J ( f ′ ) Q ~ ( f ′ ) Q ~ ∗ ( f ′ ) = T ∫ d f N I ( f ) N J ( f ) Q ~ ( f ) Q ~ ∗ ( f )
ρ I J 2 = ⟨ R ⟩ 2 T o b s ⟨ Q ~ ∣ γ I J ( f ) S h ( f ) N I ( f ) N J ( f ) ⟩ 2 ⟨ Q ~ ∣ Q ~ ⟩ \rho_{\tiny IJ}^2 = \langle\mathcal{R}\rangle^2 T_{\rm obs} \frac{ \left\langle \tilde{Q}\Big|\frac{\gamma_{IJ}(f) S_h(f)}{ N_I(f) N_J(f)} \right\rangle^2 }{ \left\langle \tilde{Q}|\tilde{Q}\right\rangle }
ρ I J 2 = ⟨ R ⟩ 2 T o b s ⟨ Q ~ ∣ Q ~ ⟩ ⟨ Q ~ ∣ ∣ ∣ ∣ N I ( f ) N J ( f ) γ I J ( f ) S h ( f ) ⟩ 2
Q ~ ( f ) = q γ I J ( f ) S h ( f ) N I ( f ) N J ( f ) \tilde{Q}(f) = q \frac{\gamma_{IJ}(f) S_h(f)}{N_I(f)N_J(f)}
Q ~ ( f ) = q N I ( f ) N J ( f ) γ I J ( f ) S h ( f )
ρ I J 2 = ⟨ R ⟩ 2 T o b s ⟨ γ I J ( f ) S h ( f ) N I ( f ) N J ( f ) ∣ γ I J ( f ) S h ( f ) N I ( f ) N J ( f ) ⟩ \rho_{\tiny IJ}^2 = \langle\mathcal{R}\rangle^2 T_{\rm obs} \left\langle \frac{\gamma_{IJ}(f) S_h(f)}{ N_I(f) N_J(f)} \Big|\frac{\gamma_{IJ}(f) S_h(f)}{ N_I(f) N_J(f)} \right\rangle
ρ I J 2 = ⟨ R ⟩ 2 T o b s ⟨ N I ( f ) N J ( f ) γ I J ( f ) S h ( f ) ∣ ∣ ∣ ∣ N I ( f ) N J ( f ) γ I J ( f ) S h ( f ) ⟩
ρ I J = ⟨ R ⟩ T o b s ∫ d f γ I J 2 ( f ) S h 2 ( f ) N I ( f ) N J ( f ) \boxed{\rho_{\tiny IJ} = \langle \mathcal{R} \rangle \sqrt{T_{\rm obs}} \sqrt{\int df \frac{ \gamma_{IJ}^2(f) S^2_h(f)}{N_I(f) N_J(f)} } }
ρ I J = ⟨ R ⟩ T o b s ∫ d f N I ( f ) N J ( f ) γ I J 2 ( f ) S h 2 ( f )
ρ I J = ⟨ R ⟩ T o b s ∫ d f ( γ I J ( f ) Ω G W ( f ) Ω n I J ( f ) ) 2 \rho_{\tiny IJ} = \langle R\rangle \sqrt{T_{\rm obs}} \sqrt{\int df \left( \frac{\gamma_{IJ}(f) \Omega_{\rm \tiny GW}(f)}{ \Omega_{n_{IJ}}(f) }\right)^2 }
ρ I J = ⟨ R ⟩ T o b s ∫ d f ( Ω n I J ( f ) γ I J ( f ) Ω G W ( f ) ) 2
其中Ω n I J ( f ) ≡ 3 H 0 2 4 π 2 f − 3 N I ( f ) N J ( f ) \Omega_{n_{IJ}}(f) \equiv \frac{3H_0^2}{4\pi^2} f^{-3} N_I(f)N_J(f) Ω n I J ( f ) ≡ 4 π 2 3 H 0 2 f − 3 N I ( f ) N J ( f )
T o b s T_{\rm obs} T o b s
S = ∫ d f ∫ d f ′ δ T ( f − f ′ ) d ~ i ( f ) d ~ j ∗ ( f ′ ) Q ~ ( f ′ ) \mathbb{S} = \int df \int df' ~ \delta_T(f-f') \tilde{d}_i(f) \tilde{d}^*_j(f') \tilde{Q}(f')
S = ∫ d f ∫ d f ′ δ T ( f − f ′ ) d ~ i ( f ) d ~ j ∗ ( f ′ ) Q ~ ( f ′ )
⟨ S 2 ⟩ − ⟨ S ⟩ 2 = ∫ d f ∫ d f ′ ∫ d f 1 ∫ d f 1 ′ δ T ( f − f ′ ) δ T ( f 1 − f 1 ′ ) Q ~ ( f ′ ) Q ~ ( f 1 ′ ) [ ⟨ d ~ i ( f ) d ~ j ∗ ( f ′ ) d ~ i ( f 1 ) d ~ j ∗ ( f 1 ′ ) ⟩ − ⟨ d ~ i ( f ) d ~ j ∗ ( f ′ ) ⟩ ⟨ d ~ i ( f 1 ) d ~ j ∗ ( f 1 ′ ) ⟩ ] \begin{aligned}
\langle \mathbb{S}^2\rangle - \langle \mathbb{S}\rangle^2 &= \int df \int df'\int df_1 \int df'_1 ~ \delta_T(f-f')\delta_T(f_1-f'_1)\tilde{Q}(f')\tilde{Q}(f'_1)\\
&~~~~~\left[\langle\tilde{d}_i(f) \tilde{d}^*_j(f')\tilde{d}_i(f_1) \tilde{d}^*_j(f'_1)\rangle - \langle\tilde{d}_i(f) \tilde{d}^*_j(f')\rangle \langle\tilde{d}_i(f_1) \tilde{d}^*_j(f'_1)\rangle\right]
\end{aligned} ⟨ S 2 ⟩ − ⟨ S ⟩ 2 = ∫ d f ∫ d f ′ ∫ d f 1 ∫ d f 1 ′ δ T ( f − f ′ ) δ T ( f 1 − f 1 ′ ) Q ~ ( f ′ ) Q ~ ( f 1 ′ ) [ ⟨ d ~ i ( f ) d ~ j ∗ ( f ′ ) d ~ i ( f 1 ) d ~ j ∗ ( f 1 ′ ) ⟩ − ⟨ d ~ i ( f ) d ~ j ∗ ( f ′ ) ⟩ ⟨ d ~ i ( f 1 ) d ~ j ∗ ( f 1 ′ ) ⟩ ]
⟨ S i j S k l ⟩ − ⟨ S i j ⟩ ⟨ S k l ⟩ = ∫ d f ∫ d f ′ ∫ d f 1 ∫ d f 1 ′ δ T ( f − f ′ ) δ T ( f 1 − f 1 ′ ) Q ~ ( f ′ ) Q ~ ( f 1 ′ ) [ ⟨ d ~ i ( f ) d ~ j ∗ ( f ′ ) d ~ k ( f 1 ) d ~ l ∗ ( f 1 ′ ) ⟩ − ⟨ d ~ i ( f ) d ~ j ∗ ( f ′ ) ⟩ ⟨ d ~ k ( f 1 ) d ~ l ∗ ( f 1 ′ ) ⟩ ] \begin{aligned}
\langle \mathbb{S}_{ij}\mathbb{S}_{kl}\rangle - \langle \mathbb{S}_{ij}\rangle \langle \mathbb{S}_{kl}\rangle &= \int df \int df'\int df_1 \int df'_1 ~ \delta_T(f-f')\delta_T(f_1-f'_1)\tilde{Q}(f')\tilde{Q}(f'_1)\\
&~~~~~\left[\langle\tilde{d}_i(f) \tilde{d}^*_j(f')\tilde{d}_k(f_1) \tilde{d}^*_l(f'_1)\rangle - \langle\tilde{d}_i(f) \tilde{d}^*_j(f')\rangle \langle\tilde{d}_k(f_1) \tilde{d}^*_l(f'_1)\rangle\right]
\end{aligned} ⟨ S i j S k l ⟩ − ⟨ S i j ⟩ ⟨ S k l ⟩ = ∫ d f ∫ d f ′ ∫ d f 1 ∫ d f 1 ′ δ T ( f − f ′ ) δ T ( f 1 − f 1 ′ ) Q ~ ( f ′ ) Q ~ ( f 1 ′ ) [ ⟨ d ~ i ( f ) d ~ j ∗ ( f ′ ) d ~ k ( f 1 ) d ~ l ∗ ( f 1 ′ ) ⟩ − ⟨ d ~ i ( f ) d ~ j ∗ ( f ′ ) ⟩ ⟨ d ~ k ( f 1 ) d ~ l ∗ ( f 1 ′ ) ⟩ ]
(Orlando et al. 2021) The four-point correlators is expressed as a sum of products of two-point correlators using the Wick theorem.
假设i , k i, k i , k 和j , l j, l j , l 非别为LISA、Taiji指标,从而必然有i ≠ j , k ≠ l i \neq j, k \neq l i = j , k = l 以及i ≠ l , k ≠ j i \neq l, k \neq j i = l , k = j 。
⟨ C i j ( f , f ′ ) C k l ∗ ( f 1 , f 1 ′ ) ⟩ = ⟨ d ~ i ( f ) d ~ j ∗ ( f ′ ) d ~ k ∗ ( f 1 ) d ~ l ( f 1 ′ ) ⟩ = ⟨ d ~ i d ~ j ∗ ⟩ ⟨ d ~ k ∗ d ~ l ⟩ + ⟨ d ~ i d ~ l ⟩ ⟨ d ~ k ∗ d ~ j ∗ ⟩ + ⟨ d ~ i d ~ k ∗ ⟩ ⟨ d ~ j ∗ d ~ l ⟩ \begin{aligned}
\langle \mathcal{C}_{ij}(f, f') \mathcal{C}^*_{kl}(f_1, f'_1)\rangle &= \langle \tilde{d}_i(f) \tilde{d}^*_j(f') \tilde{d}^*_k(f_1) \tilde{d}_l(f'_1) \rangle \\
&= \langle \tilde{d}_i \tilde{d}^*_j \rangle \langle \tilde{d}^*_k \tilde{d}_l \rangle + \langle \tilde{d}_i \tilde{d}_l \rangle \langle \tilde{d}^*_k \tilde{d}^*_j \rangle + \langle \tilde{d}_i \tilde{d}^*_k \rangle \langle \tilde{d}^*_j \tilde{d}_l \rangle
\end{aligned} ⟨ C i j ( f , f ′ ) C k l ∗ ( f 1 , f 1 ′ ) ⟩ = ⟨ d ~ i ( f ) d ~ j ∗ ( f ′ ) d ~ k ∗ ( f 1 ) d ~ l ( f 1 ′ ) ⟩ = ⟨ d ~ i d ~ j ∗ ⟩ ⟨ d ~ k ∗ d ~ l ⟩ + ⟨ d ~ i d ~ l ⟩ ⟨ d ~ k ∗ d ~ j ∗ ⟩ + ⟨ d ~ i d ~ k ∗ ⟩ ⟨ d ~ j ∗ d ~ l ⟩
⟨ C i j C k l ∗ ⟩ − ⟨ C i j ⟩ ⟨ C k l ∗ ⟩ = ⟨ d ~ i d ~ l ⟩ ⟨ d ~ k ∗ d ~ j ∗ ⟩ + ⟨ d ~ i d ~ k ∗ ⟩ ⟨ d ~ j ∗ d ~ l ⟩ = ⟨ s ~ i s ~ l ⟩ ⟨ s ~ k ∗ s ~ j ∗ ⟩ + ⟨ s ~ i s ~ k ∗ + n ~ i n ~ k ∗ ⟩ ⟨ s ~ j ∗ s ~ l + n ~ j ∗ n ~ l ⟩ = Γ i l 1 2 S h ( f ) δ ( f − f 1 ′ ) Γ k j 1 2 S h ( f 1 ) δ ( f 1 − f ′ ) + 1 2 [ Γ i k S h ( f ) + δ i k N i ( f ) ] δ ( f − f 1 ) × 1 2 [ Γ j l S h ( f ′ ) + δ j l N j ( f ′ ) ] δ ( f ′ − f 1 ′ ) \begin{aligned}
\langle \mathcal{C}_{ij}\mathcal{C}^*_{kl}\rangle - \langle\mathcal{C}_{ij} \rangle \langle \mathcal{C}^*_{kl} \rangle &= \langle \tilde{d}_i \tilde{d}_l \rangle \langle \tilde{d}^*_k \tilde{d}^*_j \rangle + \langle \tilde{d}_i \tilde{d}^*_k \rangle \langle \tilde{d}^*_j \tilde{d}_l \rangle\\
&= \langle \tilde{s}_i \tilde{s}_l \rangle \langle \tilde{s}^*_k \tilde{s}^*_j \rangle + \langle \tilde{s}_i \tilde{s}^*_k + \tilde{n}_i \tilde{n}^*_k \rangle \langle \tilde{s}^*_j \tilde{s}_l + \tilde{n}^*_j \tilde{n}_l\rangle\\
&= \Gamma_{il} \frac{1}{2}S_h(f) \delta(f-f'_1) ~ \Gamma_{kj} \frac{1}{2}S_h(f_1)\delta(f_1-f') \\
&~~~ + \frac{1}{2}[\Gamma_{ik} S_h(f) + \delta_{ik}N_i(f)]\delta(f-f_1)~\times\\
&~~~~~~~~\frac{1}{2}[\Gamma_{jl} S_h(f') + \delta_{jl} N_j(f')]\delta(f'-f'_1)
\end{aligned} ⟨ C i j C k l ∗ ⟩ − ⟨ C i j ⟩ ⟨ C k l ∗ ⟩ = ⟨ d ~ i d ~ l ⟩ ⟨ d ~ k ∗ d ~ j ∗ ⟩ + ⟨ d ~ i d ~ k ∗ ⟩ ⟨ d ~ j ∗ d ~ l ⟩ = ⟨ s ~ i s ~ l ⟩ ⟨ s ~ k ∗ s ~ j ∗ ⟩ + ⟨ s ~ i s ~ k ∗ + n ~ i n ~ k ∗ ⟩ ⟨ s ~ j ∗ s ~ l + n ~ j ∗ n ~ l ⟩ = Γ i l 2 1 S h ( f ) δ ( f − f 1 ′ ) Γ k j 2 1 S h ( f 1 ) δ ( f 1 − f ′ ) + 2 1 [ Γ i k S h ( f ) + δ i k N i ( f ) ] δ ( f − f 1 ) × 2 1 [ Γ j l S h ( f ′ ) + δ j l N j ( f ′ ) ] δ ( f ′ − f 1 ′ )
仅考虑对角元有:
1 4 δ i j , k l { [ Γ i j S h ( f ) ] 2 δ ( f − f 1 ′ ) δ ( f 1 − f ′ ) + 1 2 [ R i S h ( f ) + N i ( f ) ] δ ( f − f 1 ) [ R j S h ( f ) + N j ( f ) ] δ ( f ′ − f 1 ′ ) } \begin{aligned}
\frac{1}{4}\delta_{ij, kl} \Big\{& \left[\Gamma_{ij} S_h(f)\right]^2 \delta(f-f'_1)\delta(f_1-f') \\
&+ \frac{1}{2}[\mathcal{R}_i S_h(f) + N_i(f)]\delta(f-f_1)~[\mathcal{R}_j S_h(f) + N_j(f)]\delta(f'-f'_1)\Big\}\\
\end{aligned} 4 1 δ i j , k l { [ Γ i j S h ( f ) ] 2 δ ( f − f 1 ′ ) δ ( f 1 − f ′ ) + 2 1 [ R i S h ( f ) + N i ( f ) ] δ ( f − f 1 ) [ R j S h ( f ) + N j ( f ) ] δ ( f ′ − f 1 ′ ) }
弱信号近似:
⟨ C i j C k l ∗ ⟩ − ⟨ C i j ⟩ ⟨ C k l ∗ ⟩ = ⟨ n ~ i n ~ k ∗ ⟩ ⟨ n ~ j ∗ n ~ l ⟩ = 1 4 δ i k δ j l N i ( f ) δ ( f − f 1 ) N j ( f ) δ ( f ′ − f 1 ′ ) \begin{aligned}
\langle \mathcal{C}_{ij} \mathcal{C}^*_{kl}\rangle - \langle\mathcal{C}_{ij} \rangle \langle \mathcal{C}^*_{kl} \rangle &= \langle \tilde{n}_i \tilde{n}^*_k \rangle \langle \tilde{n}^*_j \tilde{n}_l\rangle = \frac{1}{4} \delta_{ik}\delta_{jl} N_i(f)\delta(f-f_1)~ N_j(f)\delta(f'-f'_1)
\end{aligned} ⟨ C i j C k l ∗ ⟩ − ⟨ C i j ⟩ ⟨ C k l ∗ ⟩ = ⟨ n ~ i n ~ k ∗ ⟩ ⟨ n ~ j ∗ n ~ l ⟩ = 4 1 δ i k δ j l N i ( f ) δ ( f − f 1 ) N j ( f ) δ ( f ′ − f 1 ′ )
配合S \mathbb{S} S 协方差中积分及 δ T ( f − f ′ ) , δ T ( f 1 − f 1 ′ ) \delta_T(f-f'), \delta_T(f_1-f'_1) δ T ( f − f ′ ) , δ T ( f 1 − f 1 ′ ) 可以将这里的δ \delta δ 约化。
⟨ n ~ i ( f ) n ~ k ∗ ( f 1 ) ⟩ ⟨ n ~ j ∗ ( f ′ ) n ~ l ( f 1 ′ ) ⟩ = ∬ ⟨ x i ( t ) x k ∗ ( t 1 ) ⟩ e i 2 π f t e − i 2 π f 1 t 1 d t d t 1 ∬ ⟨ x j ( t ′ ) x l ∗ ( t 1 ′ ) ⟩ e − i 2 π f ′ t ′ e i 2 π f 1 ′ t 1 ′ d t ′ d t 1 ′ = ∫ e − i 2 π ( f 1 − f ) t d t ∫ r i ( τ ) δ i k e i 2 π f 1 τ d τ ∫ e − i 2 π ( f ′ − f 1 ′ ) t ′ d t ′ ∫ r j ( τ ′ ) δ j l e i 2 π f 1 ′ τ ′ d τ ′ = δ ( f 1 − f ) δ i k 1 2 N i ( f 1 ) δ ( f ′ − f 1 ′ ) δ j l 1 2 N ( f 1 ′ ) = 1 4 δ i j , k l δ ( f − f 1 ) N i ( f ) δ ( f ′ − f 1 ′ ) N ( f ′ ) \begin{aligned}
&\langle \tilde{n}_i(f) \tilde{n}^*_k(f_1) \rangle \langle \tilde{n}^*_j(f') \tilde{n}_l(f'_1) \rangle \\
&= \iint \langle x_i(t) x^*_k(t_1)\rangle e^{i 2\pi f t} e^{-i 2\pi f_1 t_1} dt dt_1 \iint \langle x_j(t') x^*_l(t'_1) \rangle e^{-i 2\pi f' t'} e^{i 2\pi f'_1 t'_1} dt'dt'_1 \\
&= \int e^{-i 2\pi (f_1-f) t} dt \int r_i (\tau) \delta_{ik} e^{i 2\pi f_1 \tau}d\tau \int e^{-i 2\pi (f'-f'_1) t'}dt' \int r_j(\tau') \delta_{jl} e^{i 2\pi f'_1 \tau'}d\tau'\\
&= \delta(f_1-f) \delta_{ik} \frac{1}{2}N_i(f_1) \delta(f'-f'_1)\delta_{jl} \frac{1}{2}N(f'_1)\\
&= \frac{1}{4} \delta_{ij,kl} \delta(f-f_1) N_i(f) \delta(f'-f'_1) N(f')\\
\end{aligned} ⟨ n ~ i ( f ) n ~ k ∗ ( f 1 ) ⟩ ⟨ n ~ j ∗ ( f ′ ) n ~ l ( f 1 ′ ) ⟩ = ∬ ⟨ x i ( t ) x k ∗ ( t 1 ) ⟩ e i 2 π f t e − i 2 π f 1 t 1 d t d t 1 ∬ ⟨ x j ( t ′ ) x l ∗ ( t 1 ′ ) ⟩ e − i 2 π f ′ t ′ e i 2 π f 1 ′ t 1 ′ d t ′ d t 1 ′ = ∫ e − i 2 π ( f 1 − f ) t d t ∫ r i ( τ ) δ i k e i 2 π f 1 τ d τ ∫ e − i 2 π ( f ′ − f 1 ′ ) t ′ d t ′ ∫ r j ( τ ′ ) δ j l e i 2 π f 1 ′ τ ′ d τ ′ = δ ( f 1 − f ) δ i k 2 1 N i ( f 1 ) δ ( f ′ − f 1 ′ ) δ j l 2 1 N ( f 1 ′ ) = 4 1 δ i j , k l δ ( f − f 1 ) N i ( f ) δ ( f ′ − f 1 ′ ) N ( f ′ )
S \mathbb{S} S 均值为两重积分,有1个δ T \delta_T δ T ,对应谱也有1个δ \delta δ ;方差为四重积分,有2个δ T \delta_T δ T ,对应这里还有2个δ \delta δ 。
⟨ C i j C k l ∗ ⟩ − ⟨ C i j ⟩ ⟨ C k l ∗ ⟩ = 1 4 T o b s N i ( f ) N j ( f ) \langle \mathcal{C}_{ij} \mathcal{C}^*_{kl}\rangle - \langle\mathcal{C}_{ij} \rangle \langle \mathcal{C}^*_{kl} \rangle = \frac{1}{4}T_{\rm obs} N_i(f)N_j(f)
⟨ C i j C k l ∗ ⟩ − ⟨ C i j ⟩ ⟨ C k l ∗ ⟩ = 4 1 T o b s N i ( f ) N j ( f )
s ~ ( f ) = ∫ d k ^ D a b h ~ a b ( f , k ^ ) e − i 2 π f k ^ ⋅ r ⃗ c \tilde{s}(f) = \int d\hat{k} D^{ab}\tilde{h}_{ab}(f, \hat{k}) e^{-i2\pi f \hat{k}\cdot\frac{\vec{r}}{c}}
s ~ ( f ) = ∫ d k ^ D a b h ~ a b ( f , k ^ ) e − i 2 π f k ^ ⋅ c r
s ~ ( f ) = ∑ P = R , L ∫ d k ^ D a b h ~ P ( f , k ^ ) e a b P e − i 2 π f k ^ ⋅ r ⃗ c \tilde{s}(f) = \sum_{P=R,L} \int d\hat{k} D^{ab}\tilde{h}_P(f, \hat{k}) e^P_{ab} e^{-i2\pi f \hat{k}\cdot\frac{\vec{r}}{c}}
s ~ ( f ) = P = R , L ∑ ∫ d k ^ D a b h ~ P ( f , k ^ ) e a b P e − i 2 π f k ^ ⋅ c r
s ~ ( f ) = ∑ P = R , L ∫ d k ^ F P h ~ P ( f , k ^ ) e − i 2 π f k ^ ⋅ r ⃗ c \tilde{s}(f) = \sum_{P=R,L} \int d\hat{k} F^P \tilde{h}_P(f, \hat{k}) e^{-i2\pi f \hat{k}\cdot\frac{\vec{r}}{c}}
s ~ ( f ) = P = R , L ∑ ∫ d k ^ F P h ~ P ( f , k ^ ) e − i 2 π f k ^ ⋅ c r
S = ∫ d f ∫ d f ′ δ T ( f − f ′ ) d ~ I ( f ) d ~ J ∗ ( f ′ ) Q ~ ∗ ( f ′ ) \mathbb{S} = \int df \int df' ~ \delta_T(f-f') \tilde{d}_I(f) \tilde{d}^*_J(f')\tilde{Q}^*(f')
S = ∫ d f ∫ d f ′ δ T ( f − f ′ ) d ~ I ( f ) d ~ J ∗ ( f ′ ) Q ~ ∗ ( f ′ )
⟨ S ⟩ = ∫ d f ∫ d f ′ δ T ( f − f ′ ) ⟨ s ~ I ( f ) s ~ J ∗ ( f ′ ) ⟩ Q ~ ∗ ( f ′ ) \langle \mathbb{S} \rangle = \int df \int df' ~ \delta_T(f-f') \big\langle \tilde{s}_I(f) \tilde{s}^*_J(f')\big\rangle \tilde{Q}^*(f')
⟨ S ⟩ = ∫ d f ∫ d f ′ δ T ( f − f ′ ) ⟨ s ~ I ( f ) s ~ J ∗ ( f ′ ) ⟩ Q ~ ∗ ( f ′ )
s ~ I ( f ) = ∫ d 2 k ^ D I a b h ~ a b ( f , k ^ ) e − i 2 π f k ^ ⋅ r ⃗ I c = ∑ P ∫ d 2 k ^ F I P ( k ^ ) h ~ P ( f , k ^ ) e − i 2 π f k ^ ⋅ r ⃗ I c \tilde{s}_I(f) = \int d^2\hat{k} ~ D_I^{ab} \tilde{h}_{ab}(f, \hat{k}) e^{-i2\pi f \hat{k}\cdot \frac{\vec{r}_I}{c}} = \sum_P \int d^2\hat{k} ~ F^P_I(\hat{k}) \tilde{h}_P(f, \hat{k}) e^{-i2\pi f \hat{k}\cdot \frac{\vec{r}_I}{c}}
s ~ I ( f ) = ∫ d 2 k ^ D I a b h ~ a b ( f , k ^ ) e − i 2 π f k ^ ⋅ c r I = P ∑ ∫ d 2 k ^ F I P ( k ^ ) h ~ P ( f , k ^ ) e − i 2 π f k ^ ⋅ c r I
⟨ s ~ I ( f ) s ~ J ∗ ( f ′ ) ⟩ = ∑ P ∑ P ′ ∫ d 2 k ^ ∫ d 2 k ^ ′ F I P ( k ^ ) F J P ′ ( k ^ ′ ) ⟨ h ~ P ( f , k ^ ) h ~ P ′ ∗ ( f ′ , k ^ ′ ) ⟩ e − i 2 π f k ^ ⋅ r ⃗ I − r ⃗ J c = 1 4 π δ ( f − f ′ ) S h ( f ) ∑ P ∫ d 2 k ^ F I P ( k ^ ) F J P ( k ^ ) e − i 2 π f k ^ ⋅ r ⃗ I − r ⃗ J c = δ ( f − f ′ ) S h ( f ) Γ I J ( f ) \begin{aligned}
\big\langle \tilde{s}_I(f) \tilde{s}^*_J(f')\big\rangle &= \small \sum_P \sum_{P'} \int d^2\hat{k}\int d^2\hat{k}' ~ F^P_I(\hat{k})F^{P'}_J(\hat{k}') \big\langle \tilde{h}_P(f, \hat{k})\tilde{h}^*_{P'}(f', \hat{k}') \big\rangle e^{-i2\pi f \hat{k}\cdot \frac{\vec{r}_I-\vec{r}_J}{c}}\\
&= \frac{1}{4\pi}\delta(f-f')S_h(f) \sum_P \int d^2\hat{k} ~ F^P_I(\hat{k})F^P_J(\hat{k}) e^{-i2\pi f \hat{k}\cdot \frac{\vec{r}_I-\vec{r}_J}{c}}\\
&= \delta(f-f')S_h(f) \Gamma_{IJ}(f)\\
\end{aligned} ⟨ s ~ I ( f ) s ~ J ∗ ( f ′ ) ⟩ = P ∑ P ′ ∑ ∫ d 2 k ^ ∫ d 2 k ^ ′ F I P ( k ^ ) F J P ′ ( k ^ ′ ) ⟨ h ~ P ( f , k ^ ) h ~ P ′ ∗ ( f ′ , k ^ ′ ) ⟩ e − i 2 π f k ^ ⋅ c r I − r J = 4 π 1 δ ( f − f ′ ) S h ( f ) P ∑ ∫ d 2 k ^ F I P ( k ^ ) F J P ( k ^ ) e − i 2 π f k ^ ⋅ c r I − r J = δ ( f − f ′ ) S h ( f ) Γ I J ( f )
⟨ h ~ P ( f , k ^ ) h ~ P ′ ∗ ( f ′ , k ^ ′ ) ⟩ ‾ = 1 4 π δ ( f − f ′ ) δ 2 ( k ^ , k ^ ′ ) δ P P ′ S h ( f ) \overline{\left\langle \tilde{h}_P(f,\hat{k})\tilde{h}^*_{P'}(f',\hat{k}') \right\rangle} = \frac{1}{4\pi} \delta(f-f')~\delta^2(\hat{k},\hat{k}')~\delta_{PP'}~S_h(f)
⟨ h ~ P ( f , k ^ ) h ~ P ′ ∗ ( f ′ , k ^ ′ ) ⟩ = 4 π 1 δ ( f − f ′ ) δ 2 ( k ^ , k ^ ′ ) δ P P ′ S h ( f )
对于有圆偏振情况,上述结果不再成立!此时:
( ⟨ h ~ + ( f , k ^ ) h ~ + ∗ ( f ′ , k ^ ′ ) ⟩ ⟨ h ~ + ( f , k ^ ) h ~ × ∗ ( f ′ , k ^ ′ ) ⟩ ⟨ h ~ + ∗ ( f , k ^ ) h ~ × ( f ′ , k ^ ′ ) ⟩ ⟨ h ~ × ( f , k ^ ) h ~ × ∗ ( f ′ , k ^ ′ ) ⟩ ) = δ ( f − f ′ ) δ 2 ( k ^ , k ^ ′ ) 4 π ( I + Q U − i V U + i V I − Q ) \begin{pmatrix}
\langle \tilde{h}_+(f, \hat{k}) \tilde{h}_+^*(f', \hat{k}')\rangle & \langle \tilde{h}_+(f, \hat{k}) \tilde{h}_\times^*(f', \hat{k}')\rangle \\
\langle \tilde{h}_+^*(f, \hat{k}) \tilde{h}_\times(f', \hat{k}')\rangle & \langle \tilde{h}_\times(f, \hat{k}) \tilde{h}_\times^*(f', \hat{k}')\rangle
\end{pmatrix} =
\frac{\delta(f-f')~\delta^2(\hat{k},\hat{k}')}{4\pi}
\begin{pmatrix}
I + Q & U-iV \\
U + iV & I-Q
\end{pmatrix} ( ⟨ h ~ + ( f , k ^ ) h ~ + ∗ ( f ′ , k ^ ′ ) ⟩ ⟨ h ~ + ∗ ( f , k ^ ) h ~ × ( f ′ , k ^ ′ ) ⟩ ⟨ h ~ + ( f , k ^ ) h ~ × ∗ ( f ′ , k ^ ′ ) ⟩ ⟨ h ~ × ( f , k ^ ) h ~ × ∗ ( f ′ , k ^ ′ ) ⟩ ) = 4 π δ ( f − f ′ ) δ 2 ( k ^ , k ^ ′ ) ( I + Q U + i V U − i V I − Q )
I = ⟨ ∣ h ~ + ∣ 2 + ∣ h ~ × ∣ 2 ⟩ / 2 , Q = ⟨ ∣ h ~ + ∣ 2 − ∣ h ~ × ∣ 2 ⟩ / 2 I = \langle |\tilde{h}_+|^2 + |\tilde{h}_\times|^2\rangle /2, ~~~ Q = \langle |\tilde{h}_+|^2 - |\tilde{h}_\times|^2\rangle /2
I = ⟨ ∣ h ~ + ∣ 2 + ∣ h ~ × ∣ 2 ⟩ / 2 , Q = ⟨ ∣ h ~ + ∣ 2 − ∣ h ~ × ∣ 2 ⟩ / 2
U = ⟨ h ~ + h ~ × ∗ + h ~ + ∗ h ~ × ⟩ / 2 , V = i ⟨ h ~ + h ~ × ∗ − h ~ + ∗ h ~ × ⟩ / 2 U = \langle \tilde{h}_+\tilde{h}_\times^* + \tilde{h}_+^*\tilde{h}_\times\rangle /2, ~~~ V= i\langle \tilde{h}_+\tilde{h}_\times^* - \tilde{h}_+^*\tilde{h}_\times\rangle /2
U = ⟨ h ~ + h ~ × ∗ + h ~ + ∗ h ~ × ⟩ / 2 , V = i ⟨ h ~ + h ~ × ∗ − h ~ + ∗ h ~ × ⟩ / 2
其中Q , U Q, U Q , U 对应线偏振的强度差异,分别为当前坐标下x y xy x y 与斜4 5 ∘ 45^\circ 4 5 ∘ 方向线偏振,即h ~ + , h ~ × \tilde{h}_+,\tilde{h}_\times h ~ + , h ~ × 的差异,以及旋转π 4 \frac{\pi}{4} 4 π 后两个相对的张量方向的差异。当基矢旋转时,Q , U Q,U Q , U 具有自旋± 4 \pm 4 ± 4 ,多级矩展开时从4阶矩开始,对于各向同性的背景,两者取值均为0。I I I 为背景总强度,V V V 对应左右旋信号的差异,即总体圆偏振强度,两者在基矢旋转时保持不变。从而对于各项同性背景有
( ⟨ h ~ + ( f , k ^ ) h ~ + ∗ ( f ′ , k ^ ′ ) ⟩ ⟨ h ~ + ( f , k ^ ) h ~ × ∗ ( f ′ , k ^ ′ ) ⟩ ⟨ h ~ + ∗ ( f , k ^ ) h ~ × ( f ′ , k ^ ′ ) ⟩ ⟨ h ~ × ( f , k ^ ) h ~ × ∗ ( f ′ , k ^ ′ ) ⟩ ) = δ ( f − f ′ ) δ 2 ( k ^ , k ^ ′ ) 4 π ( I − i V i V I ) \begin{pmatrix}
\langle \tilde{h}_+(f, \hat{k}) \tilde{h}_+^*(f', \hat{k}')\rangle & \langle \tilde{h}_+(f, \hat{k}) \tilde{h}_\times^*(f', \hat{k}')\rangle \\
\langle \tilde{h}_+^*(f, \hat{k}) \tilde{h}_\times(f', \hat{k}')\rangle & \langle \tilde{h}_\times(f, \hat{k}) \tilde{h}_\times^*(f', \hat{k}')\rangle
\end{pmatrix} =
\frac{\delta(f-f')~\delta^2(\hat{k},\hat{k}')}{4\pi}
\begin{pmatrix}
I & -iV \\
iV & I
\end{pmatrix} ( ⟨ h ~ + ( f , k ^ ) h ~ + ∗ ( f ′ , k ^ ′ ) ⟩ ⟨ h ~ + ∗ ( f , k ^ ) h ~ × ( f ′ , k ^ ′ ) ⟩ ⟨ h ~ + ( f , k ^ ) h ~ × ∗ ( f ′ , k ^ ′ ) ⟩ ⟨ h ~ × ( f , k ^ ) h ~ × ∗ ( f ′ , k ^ ′ ) ⟩ ) = 4 π δ ( f − f ′ ) δ 2 ( k ^ , k ^ ′ ) ( I i V − i V I )
当没有总体圆偏振,才回归到最初的公式。
引入左右旋的张量基,及对应的左右旋应力张量。
e R = e + + i e × 2 , e L = e + − i e × 2 h R = h + − i h × 2 , h L = h + + i h × 2 \begin{gathered}
\bm{e}^R = \frac{\bm{e}^+ + i\bm{e}^\times}{\sqrt{2}}, ~~~~~ \bm{e}^L = \frac{\bm{e}^+ - i\bm{e}^\times}{\sqrt{2}}\\
h_R = \frac{h_+ - i h_\times}{\sqrt{2}}, ~~~~~ h_L = \frac{h_+ + i h_\times}{\sqrt{2}}\\
% h_+ = \frac{h_R + h_L}{\sqrt{2}}, ~~~~~ h_\times = i \frac{h_R - h_L}{\sqrt{2}}
\end{gathered} e R = 2 e + + i e × , e L = 2 e + − i e × h R = 2 h + − i h × , h L = 2 h + + i h ×
h a b = h + e + + h × e × = h R e R + h L e L h_{ab} = h_+ \bm{e}^+ + h_\times \bm{e}^\times = h_R~\bm{e}^R + h_L~\bm{e}^L
h a b = h + e + + h × e × = h R e R + h L e L
⟨ h ~ R h ~ R ∗ ⟩ + ⟨ h ~ L h ~ L ∗ ⟩ = ⟨ h ~ + h ~ + ∗ ⟩ + ⟨ h ~ × h ~ × ∗ ⟩ ⟨ h ~ R h ~ R ∗ ⟩ − ⟨ h ~ L h ~ L ∗ ⟩ = i [ ⟨ h ~ + h ~ × ∗ ⟩ − ⟨ h ~ × h ~ + ∗ ⟩ ] \begin{gathered}
\langle \tilde{h}_R \tilde{h}_R^*\rangle + \langle \tilde{h}_L \tilde{h}_L^*\rangle = \langle \tilde{h}_+ \tilde{h}_+^*\rangle + \langle \tilde{h}_\times \tilde{h}_\times^*\rangle\\
\langle \tilde{h}_R \tilde{h}_R^*\rangle - \langle \tilde{h}_L \tilde{h}_L^*\rangle = i\left[\langle \tilde{h}_+ \tilde{h}_\times^*\rangle - \langle \tilde{h}_\times \tilde{h}_+^*\rangle\right]
\end{gathered} ⟨ h ~ R h ~ R ∗ ⟩ + ⟨ h ~ L h ~ L ∗ ⟩ = ⟨ h ~ + h ~ + ∗ ⟩ + ⟨ h ~ × h ~ × ∗ ⟩ ⟨ h ~ R h ~ R ∗ ⟩ − ⟨ h ~ L h ~ L ∗ ⟩ = i [ ⟨ h ~ + h ~ × ∗ ⟩ − ⟨ h ~ × h ~ + ∗ ⟩ ]
I ( f ) = 1 2 [ S R ( f ) + S L ( f ) ] , V ( f ) = 1 2 [ S R ( f ) − S L ( f ) ] S R ( f ) = I ( f ) + V ( f ) , S L ( f ) = I ( f ) − V ( f ) v . s . I ( f ) = S R ( f ) + S L ( f ) , V ( f ) = S R ( f ) − S L ( f ) S R ( f ) = 1 2 [ I ( f ) + V ( f ) ] , S L ( f ) = 1 2 [ I ( f ) − V ( f ) ] \begin{gathered}
I(f) = \frac{1}{2}[S_R(f) + S_L(f)], ~~~ V(f) = \frac{1}{2}[S_R(f) - S_L(f)]\\ %, ~~ \chi(f) = \frac{V(f)}{I(f)}\\
S_R(f) = I(f) + V(f), ~~~ S_L(f) = I(f) - V(f)\\
%\mathcal{R}^I(f) = R_R(f) + R_L(f), ~~~~~ V(f) = S_R(f) - S_L(f)\\
{\rm \color{red} v.s. }\\
I(f) = S_R(f) + S_L(f), ~~~ V(f) = S_R(f) - S_L(f)\\ %, ~~ \chi(f) = \frac{V(f)}{I(f)}\\
S_R(f) = \frac{1}{2}[I(f) + V(f)], ~~~ S_L(f) = \frac{1}{2}[I(f) - V(f)]\\
%\mathcal{R}^I(f) = R_R(f) + R_L(f), ~~~~~ V(f) = S_R(f) - S_L(f)
\end{gathered} I ( f ) = 2 1 [ S R ( f ) + S L ( f ) ] , V ( f ) = 2 1 [ S R ( f ) − S L ( f ) ] S R ( f ) = I ( f ) + V ( f ) , S L ( f ) = I ( f ) − V ( f ) v . s . I ( f ) = S R ( f ) + S L ( f ) , V ( f ) = S R ( f ) − S L ( f ) S R ( f ) = 2 1 [ I ( f ) + V ( f ) ] , S L ( f ) = 2 1 [ I ( f ) − V ( f ) ]
写作左右旋形式则有:
( ⟨ h ~ R ( f , k ^ ) h ~ R ∗ ( f ′ , k ^ ′ ) ⟩ ⟨ h ~ R ( f , k ^ ) h ~ L ∗ ( f ′ , k ^ ′ ) ⟩ ⟨ h ~ R ∗ ( f , k ^ ) h ~ L ( f ′ , k ^ ′ ) ⟩ ⟨ h ~ L ( f , k ^ ) h ~ L ∗ ( f ′ , k ^ ′ ) ⟩ ) = δ ( f − f ′ ) δ 2 ( k ^ , k ^ ′ ) 4 π ( I + V 0 0 I − V ) \begin{pmatrix}
\langle \tilde{h}_R(f, \hat{k}) \tilde{h}_R^*(f', \hat{k}')\rangle & \langle \tilde{h}_R(f, \hat{k}) \tilde{h}_L^*(f', \hat{k}')\rangle \\
\langle \tilde{h}_R^*(f, \hat{k}) \tilde{h}_L(f', \hat{k}')\rangle & \langle \tilde{h}_L(f, \hat{k}) \tilde{h}_L^*(f', \hat{k}')\rangle
\end{pmatrix} =
\frac{\delta(f-f')~\delta^2(\hat{k},\hat{k}')}{4\pi}
\begin{pmatrix}
I+V & 0 \\
0 & I-V
\end{pmatrix} ( ⟨ h ~ R ( f , k ^ ) h ~ R ∗ ( f ′ , k ^ ′ ) ⟩ ⟨ h ~ R ∗ ( f , k ^ ) h ~ L ( f ′ , k ^ ′ ) ⟩ ⟨ h ~ R ( f , k ^ ) h ~ L ∗ ( f ′ , k ^ ′ ) ⟩ ⟨ h ~ L ( f , k ^ ) h ~ L ∗ ( f ′ , k ^ ′ ) ⟩ ) = 4 π δ ( f − f ′ ) δ 2 ( k ^ , k ^ ′ ) ( I + V 0 0 I − V )
或者简写做
⟨ h ~ P ( f , k ^ ) h ~ P ′ ∗ ( f ′ , k ^ ′ ) ⟩ = 1 4 π δ ( f − f ′ ) δ 2 ( k ^ , k ^ ′ ) δ P P ′ S P ( f ) \left\langle \tilde{h}_P(f,\hat{k})\tilde{h}^*_{P'}(f',\hat{k}') \right\rangle = \frac{1}{4\pi} \delta(f-f')~\delta^2(\hat{k},\hat{k}')~\delta_{PP'}~S_P(f)
⟨ h ~ P ( f , k ^ ) h ~ P ′ ∗ ( f ′ , k ^ ′ ) ⟩ = 4 π 1 δ ( f − f ′ ) δ 2 ( k ^ , k ^ ′ ) δ P P ′ S P ( f )
注意此时左右旋对应各自S P ( f ) S_P(f) S P ( f ) ,而不再是同一个功率谱。
当存在园偏振时,F P F^P F P 不再是实数
⟨ s ~ I ( f ) s ~ J ∗ ( f ′ ) ⟩ = ∑ P ∑ P ′ ∫ d 2 k ^ ∫ d 2 k ^ ′ F I P ( k ^ ) F J P ′ ∗ ( k ^ ′ ) ⟨ h ~ P ( f , k ^ ) h ~ P ′ ∗ ( f ′ , k ^ ′ ) ⟩ e − i 2 π f k ^ ⋅ r ⃗ I − r ⃗ J c = δ f , f ′ ∑ P ∫ d 2 k ^ F I P ( k ^ ) F J P ∗ ( k ^ ) S P ( f ) e − i 2 π f k ^ ⋅ r ⃗ I − r ⃗ J c \begin{aligned}
\big\langle \tilde{s}_I(f) \tilde{s}^*_J(f')\big\rangle &= \small \sum_P \sum_{P'} \int d^2\hat{k}\int d^2\hat{k}' ~ F^P_I(\hat{k})F^{P'*}_J(\hat{k}') \big\langle \tilde{h}_P(f, \hat{k}) \tilde{h}^*_{P'}(f', \hat{k}') \big\rangle e^{-i2\pi f \hat{k}\cdot \frac{\vec{r}_I-\vec{r}_J}{c}}\\
&= \delta_{f, f'}\sum_P \int d^2\hat{k} ~ F^P_I(\hat{k})F^{P*}_J(\hat{k}) S_P(f) e^{-i2\pi f \hat{k}\cdot \frac{\vec{r}_I-\vec{r}_J}{c}}
\end{aligned} ⟨ s ~ I ( f ) s ~ J ∗ ( f ′ ) ⟩ = P ∑ P ′ ∑ ∫ d 2 k ^ ∫ d 2 k ^ ′ F I P ( k ^ ) F J P ′ ∗ ( k ^ ′ ) ⟨ h ~ P ( f , k ^ ) h ~ P ′ ∗ ( f ′ , k ^ ′ ) ⟩ e − i 2 π f k ^ ⋅ c r I − r J = δ f , f ′ P ∑ ∫ d 2 k ^ F I P ( k ^ ) F J P ∗ ( k ^ ) S P ( f ) e − i 2 π f k ^ ⋅ c r I − r J
F I R F J R ∗ = D a b D c d e a b R e c d R ∗ = D I a b D J c d [ e a b + e c d + + e a b × e c d × − i e a b + e c d × + i e a b × e c d + ] F_I^R F_J^{R*} = D_{ab}D_{cd} e_{ab}^{R}e_{cd}^{R*} = D_I^{ab}D_J^{cd} [e_{ab}^{+}e_{cd}^{+} + e_{ab}^{\times}e_{cd}^{\times} - i e_{ab}^+e_{cd}^\times + i e_{ab}^\times e_{cd}^+]
F I R F J R ∗ = D a b D c d e a b R e c d R ∗ = D I a b D J c d [ e a b + e c d + + e a b × e c d × − i e a b + e c d × + i e a b × e c d + ]
F I L F J L ∗ = D a b D c d e a b R e c d R ∗ = D a b D c d [ e a b + e c d + + e a b × e c d × + i e a b + e c d × − i e a b × e c d + ] F_I^L F_J^{L*} = D_{ab}D_{cd} e_{ab}^{R}e_{cd}^{R*} = D_{ab}D_{cd} [e_{ab}^{+}e_{cd}^{+} + e_{ab}^{\times}e_{cd}^{\times} + i e_{ab}^+e_{cd}^\times - i e_{ab}^\times e_{cd}^+]
F I L F J L ∗ = D a b D c d e a b R e c d R ∗ = D a b D c d [ e a b + e c d + + e a b × e c d × + i e a b + e c d × − i e a b × e c d + ]
⟨ s ~ I ( f ) s ~ J ∗ ( f ′ ) ⟩ = 4 π ⟨ R ⟩ [ γ I J I ( f ) I ( f ) + γ I J V ( f ) V ( f ) ] δ f , f ′ \langle \tilde{s}_I(f) \tilde{s}^*_J(f')\rangle = 4\pi \langle\mathcal{R}\rangle \left[\gamma_{IJ}^I(f) I(f) + \gamma_{IJ}^V(f) V(f) \right]\delta_{f, f'}
⟨ s ~ I ( f ) s ~ J ∗ ( f ′ ) ⟩ = 4 π ⟨ R ⟩ [ γ I J I ( f ) I ( f ) + γ I J V ( f ) V ( f ) ] δ f , f ′
γ I J I ( f ) = D I a b D J c d 1 ⟨ R ⟩ ∫ d 2 k ^ 4 π ( e a b + e c d + + e a b × e c d × ) e i k ^ ⋅ z ^ ı \gamma_{IJ}^{I}(f) = D_I^{~~ab} D_J^{~~cd} \frac{1}{\langle\mathcal{R}\rangle} \int \frac{d^2\hat{k}}{4\pi} ~ \left(e^+_{ab} e^+_{cd} + e^\times_{ab} e^\times_{cd}\right) e^{i\hat{k}\cdot \hat{z}~\imath}
γ I J I ( f ) = D I a b D J c d ⟨ R ⟩ 1 ∫ 4 π d 2 k ^ ( e a b + e c d + + e a b × e c d × ) e i k ^ ⋅ z ^
γ I J V ( f ) = D I a b D J c d i ⟨ R ⟩ ∫ d 2 k ^ 4 π ( e a b + e c d × − e a b × e c d + ) e i k ^ ⋅ z ^ ı \gamma_{IJ}^{V}(f) = D_I^{~~ab} D_J^{~~cd} \frac{i}{\langle\mathcal{R}\rangle} \int \frac{d^2\hat{k}}{4\pi} ~ \left(e^+_{ab} e^\times_{cd} - e^\times_{ab} e^+_{cd}\right) e^{i\hat{k}\cdot \hat{z}~\imath}
γ I J V ( f ) = D I a b D J c d ⟨ R ⟩ i ∫ 4 π d 2 k ^ ( e a b + e c d × − e a b × e c d + ) e i k ^ ⋅ z ^
D i j , k l V ( f , f ′ ) = d ~ i ( f ) d ~ j ∗ ( f ′ ) R i j I − d ~ k ( f ) d ~ l ∗ ( f ′ ) R k l I \mathcal{D}^V_{ij,kl}(f,f') = \frac{\tilde{d}_i(f)\tilde{d}^*_j(f')}{\mathcal{R}^I_{ij}} - \frac{\tilde{d}_k(f)\tilde{d}^*_l(f')}{\mathcal{R}^I_{kl}}
D i j , k l V ( f , f ′ ) = R i j I d ~ i ( f ) d ~ j ∗ ( f ′ ) − R k l I d ~ k ( f ) d ~ l ∗ ( f ′ )
⟨ D i j , k l V ( f , f ′ ) ⟩ = ⟨ s ~ i ( f ) s ~ j ∗ ( f ′ ) ⟩ R i j I − ⟨ s ~ k ( f ) s ~ l ∗ ( f ′ ) ⟩ R i j I = ( R i j V R i j I − R k l V R k l I ) V ( f ) δ ( f − f ′ ) \Big\langle \mathcal{D}^V_{ij,kl}(f,f') \Big\rangle = \frac{\langle \tilde{s}_i(f) \tilde{s}^*_j(f')\rangle}{\mathcal{R}^I_{ij}} - \frac{\langle \tilde{s}_k(f) \tilde{s}^*_l(f')\rangle}{\mathcal{R}^I_{ij}} = \left(\frac{\mathcal{R}^V_{ij}}{\mathcal{R}^I_{ij}}-\frac{\mathcal{R}^V_{kl}}{\mathcal{R}^I_{kl}}\right)V(f)\delta(f-f')
⟨ D i j , k l V ( f , f ′ ) ⟩ = R i j I ⟨ s ~ i ( f ) s ~ j ∗ ( f ′ ) ⟩ − R i j I ⟨ s ~ k ( f ) s ~ l ∗ ( f ′ ) ⟩ = ( R i j I R i j V − R k l I R k l V ) V ( f ) δ ( f − f ′ )
S V = ∫ d f ∫ d f ′ δ T ( f − f ′ ) D i j , k l V ( f , f ′ ) Q i j , k l ( f ) \mathbb{S}^V = \int df\int df' ~ \delta_T(f-f') \mathcal{D}^V_{ij,kl}(f,f') Q_{ij,kl}(f)
S V = ∫ d f ∫ d f ′ δ T ( f − f ′ ) D i j , k l V ( f , f ′ ) Q i j , k l ( f )
⟨ S V ⟩ = ∫ d f ∫ d f ′ δ T ( f − f ′ ) ⟨ D i j , k l V ( f , f ′ ) ⟩ Q i j , k l ( f ) = ∫ d f δ T ( 0 ) ( R i j V R i j I − R k l V R k l I ) V ( f ) Q i j , k l ( f ) = T ∫ d f ( R i j V R i j I − R k l V R k l I ) V ( f ) Q i j , k l ( f ) \begin{aligned}
\langle\mathbb{S}^V\rangle &= \int df\int df' ~ \delta_T(f-f')\Big\langle \mathcal{D}^V_{ij,kl}(f,f') \Big\rangle Q_{ij,kl}(f)\\
&= \int df ~ \delta_T(0) \left(\frac{\mathcal{R}^V_{ij}}{\mathcal{R}^I_{ij}}-\frac{\mathcal{R}^V_{kl}}{\mathcal{R}^I_{kl}}\right)V(f) Q_{ij,kl}(f)\\
&= T \int df ~ \left(\frac{\mathcal{R}^V_{ij}}{\mathcal{R}^I_{ij}}-\frac{\mathcal{R}^V_{kl}}{\mathcal{R}^I_{kl}}\right)V(f) Q_{ij,kl}(f)
\end{aligned} ⟨ S V ⟩ = ∫ d f ∫ d f ′ δ T ( f − f ′ ) ⟨ D i j , k l V ( f , f ′ ) ⟩ Q i j , k l ( f ) = ∫ d f δ T ( 0 ) ( R i j I R i j V − R k l I R k l V ) V ( f ) Q i j , k l ( f ) = T ∫ d f ( R i j I R i j V − R k l I R k l V ) V ( f ) Q i j , k l ( f )
S t o t a l V = ∑ i j , k l ∫ d f ∫ d f ′ δ T ( f − f ′ ) D i j , k l V ( f , f ′ ) Q i j , k l ( f ) = ∑ i j , k l ∫ d f ∫ d f ′ δ T ( f − f ′ ) [ d ~ i ( f ) d ~ j ∗ ( f ′ ) R i j I − d ~ k ( f ) d ~ l ∗ ( f ′ ) R k l I ] Q i j , k l ( f ) = ∑ i j ∫ d f ∫ d f ′ δ T ( f − f ′ ) d ~ i ( f ) d ~ j ∗ ( f ′ ) W i j ( f ) \begin{aligned}
\mathbb{S}^V_{\rm total} =& \sum_{ij,kl}\int df\int df' ~ \delta_T(f-f') \mathcal{D}^V_{ij,kl}(f,f') Q_{ij,kl}(f)\\
=& \sum_{ij,kl}\int df\int df' ~ \delta_T(f-f') \left[\frac{\tilde{d}_i(f)\tilde{d}^*_j(f')}{\mathcal{R}^I_{ij}} - \frac{\tilde{d}_k(f)\tilde{d}^*_l(f')}{\mathcal{R}^I_{kl}}\right] Q_{ij,kl}(f)\\
=& \sum_{ij}\int df\int df' ~ \delta_T(f-f') \tilde{d}_i(f)\tilde{d}^*_j(f') W_{ij}(f)
\end{aligned} S t o t a l V = = = i j , k l ∑ ∫ d f ∫ d f ′ δ T ( f − f ′ ) D i j , k l V ( f , f ′ ) Q i j , k l ( f ) i j , k l ∑ ∫ d f ∫ d f ′ δ T ( f − f ′ ) [ R i j I d ~ i ( f ) d ~ j ∗ ( f ′ ) − R k l I d ~ k ( f ) d ~ l ∗ ( f ′ ) ] Q i j , k l ( f ) i j ∑ ∫ d f ∫ d f ′ δ T ( f − f ′ ) d ~ i ( f ) d ~ j ∗ ( f ′ ) W i j ( f )
⟨ S t o t a l V ⟩ = ∑ i j , k l ∫ d f ∫ d f ′ δ T ( f − f ′ ) ⟨ D i j , k l V ( f , f ′ ) ⟩ Q i j , k l ( f ) = ∑ i j , k l ∫ d f ∫ d f ′ δ T ( f − f ′ ) ( R i j V R i j I − R k l V R k l I ) V ( f ) δ ( f − f ′ ) Q i j , k l ( f ) = ∑ i j ∫ d f δ T ( 0 ) R i j V V ( f ) W i j ( f ) \begin{aligned}
\langle \mathbb{S}^V_{\rm total} \rangle=& \sum_{ij,kl}\int df\int df' ~ \delta_T(f-f') \Big\langle \mathcal{D}^V_{ij,kl}(f,f')\Big\rangle Q_{ij,kl}(f)\\
=& \sum_{ij,kl}\int df\int df' ~ \delta_T(f-f') \left(\frac{\mathcal{R}^V_{ij}}{\mathcal{R}^I_{ij}}-\frac{\mathcal{R}^V_{kl}}{\mathcal{R}^I_{kl}}\right)V(f)\delta(f-f') Q_{ij,kl}(f)\\
=& \sum_{ij}\int df ~ \delta_T(0) \mathcal{R}^V_{ij} V(f) W_{ij}(f)
\end{aligned} ⟨ S t o t a l V ⟩ = = = i j , k l ∑ ∫ d f ∫ d f ′ δ T ( f − f ′ ) ⟨ D i j , k l V ( f , f ′ ) ⟩ Q i j , k l ( f ) i j , k l ∑ ∫ d f ∫ d f ′ δ T ( f − f ′ ) ( R i j I R i j V − R k l I R k l V ) V ( f ) δ ( f − f ′ ) Q i j , k l ( f ) i j ∑ ∫ d f δ T ( 0 ) R i j V V ( f ) W i j ( f )
⟨ S t o t a l V ⟩ = ∑ i j ∫ d f ∫ d f ′ δ T ( f − f ′ ) ⟨ d ~ i ( f ) d ~ j ∗ ( f ′ ) ⟩ W i j ( f ) = ∑ i j ∫ d f δ T ( 0 ) 4 π ⟨ R ⟩ [ γ I J I ( f ) I ( f ) + γ I J V ( f ) V ( f ) ] W i j ( f ) \begin{aligned}
\langle \mathbb{S}^V_{\rm total} \rangle=& \sum_{ij}\int df\int df' ~ \delta_T(f-f') \big\langle \tilde{d}_i(f)\tilde{d}^*_j(f')\big\rangle W_{ij}(f)\\
= & \sum_{ij}\int df ~ \delta_T(0) 4\pi \langle\mathcal{R}\rangle \Big[ \gamma_{IJ}^I(f) I(f) + \gamma_{IJ}^V(f) V(f) \Big] W_{ij}(f)
\end{aligned} ⟨ S t o t a l V ⟩ = = i j ∑ ∫ d f ∫ d f ′ δ T ( f − f ′ ) ⟨ d ~ i ( f ) d ~ j ∗ ( f ′ ) ⟩ W i j ( f ) i j ∑ ∫ d f δ T ( 0 ) 4 π ⟨ R ⟩ [ γ I J I ( f ) I ( f ) + γ I J V ( f ) V ( f ) ] W i j ( f )
∑ i j γ i j I ( f ) I ( f ) W i j ( f ) = ∑ i j γ i j I ( f ) I ( f ) γ i j V ( f ) V ( f ) N i ( f ) N j ( f ) ∑ i j γ i j I ( f ) γ i j V ( f ) = 0 \begin{aligned}
& \sum_{ij} \gamma_{ij}^I(f) I(f) W_{ij}(f) = \sum_{ij} \gamma_{ij}^I(f) I(f) \frac{\gamma_{ij}^V(f) V(f)}{N_i(f)N_j(f)}\\
& \boxed{\sum_{ij} \gamma_{ij}^I(f) \gamma_{ij}^V(f) = 0}
\end{aligned} i j ∑ γ i j I ( f ) I ( f ) W i j ( f ) = i j ∑ γ i j I ( f ) I ( f ) N i ( f ) N j ( f ) γ i j V ( f ) V ( f ) i j ∑ γ i j I ( f ) γ i j V ( f ) = 0
⟨ ( S t o t a l V ) 2 ⟩ = ∫ d f ∫ d f ′ ∫ d f 1 ∫ d f 1 ′ δ T ( f − f ′ ) δ T ( f 1 − f 1 ′ ) ∑ i j ∑ k l ⟨ d ~ i ( f ) d ~ j ∗ ( f ′ ) d ~ k ∗ ( f ) d ~ l ( f ′ ) ⟩ W i j ( f ) W k l ∗ ( f 1 ) ≈ ∫ d f ∫ d f ′ ∫ d f 1 ∫ d f 1 ′ δ T ( f − f ′ ) δ T ( f 1 − f 1 ′ ) ∑ i j ∑ k l ⟨ n ~ i ( f ) n ~ j ∗ ( f ′ ) n ~ k ∗ ( f ) n ~ l ( f ′ ) ⟩ W i j ( f ) W k l ∗ ( f 1 ) = ∫ d f δ T ( 0 ) δ T ( 0 ) ∑ i j N i ( f ) N j ( f ) W i j ( f ) W i j ∗ ( f ) \begin{aligned}
\langle(\mathbb{S}^V_{\rm total})^2\rangle = & \int df\int df'\int df_1\int df'_1 ~ \delta_T(f-f') \delta_T(f_1-f'_1) \\
& \sum_{ij} \sum_{kl} \Big\langle\tilde{d}_i(f)\tilde{d}^*_j(f') \tilde{d}^*_k(f)\tilde{d}_l(f')\Big\rangle W_{ij}(f)W^*_{kl}(f_1)\\
\approx & \int df\int df'\int df_1\int df'_1 ~ \delta_T(f-f') \delta_T(f_1-f'_1) \\
& \sum_{ij} \sum_{kl} \Big\langle\tilde{n}_i(f)\tilde{n}^*_j(f') \tilde{n}^*_k(f)\tilde{n}_l(f')\Big\rangle W_{ij}(f)W^*_{kl}(f_1)\\
= & \int df~ \delta_T(0) \delta_T(0) \sum_{ij} N_{i}(f)N_j(f) W_{ij}(f)W^*_{ij}(f)
\end{aligned} ⟨ ( S t o t a l V ) 2 ⟩ = ≈ = ∫ d f ∫ d f ′ ∫ d f 1 ∫ d f 1 ′ δ T ( f − f ′ ) δ T ( f 1 − f 1 ′ ) i j ∑ k l ∑ ⟨ d ~ i ( f ) d ~ j ∗ ( f ′ ) d ~ k ∗ ( f ) d ~ l ( f ′ ) ⟩ W i j ( f ) W k l ∗ ( f 1 ) ∫ d f ∫ d f ′ ∫ d f 1 ∫ d f 1 ′ δ T ( f − f ′ ) δ T ( f 1 − f 1 ′ ) i j ∑ k l ∑ ⟨ n ~ i ( f ) n ~ j ∗ ( f ′ ) n ~ k ∗ ( f ) n ~ l ( f ′ ) ⟩ W i j ( f ) W k l ∗ ( f 1 ) ∫ d f δ T ( 0 ) δ T ( 0 ) i j ∑ N i ( f ) N j ( f ) W i j ( f ) W i j ∗ ( f )
Q ~ ( f ) = q γ I J I ( f ) I ( f ) + γ I J V ( f ) V ( f ) N I ( f ) N J ( f ) ? ? ? \tilde{Q}(f) = q \frac{\gamma_{IJ}^I(f) I(f) + \gamma_{IJ}^V(f) V(f)}{N_I(f)N_J(f)}???
Q ~ ( f ) = q N I ( f ) N J ( f ) γ I J I ( f ) I ( f ) + γ I J V ( f ) V ( f ) ? ? ?
Q ~ I ( f ) = q γ I J I ( f ) I ( f ) N I ( f ) N J ( f ) ? ? ? \tilde{Q}_I(f) = q \frac{\gamma_{IJ}^I(f) I(f)}{N_I(f)N_J(f)}???
Q ~ I ( f ) = q N I ( f ) N J ( f ) γ I J I ( f ) I ( f ) ? ? ?
Q ~ V ( f ) = q γ I J V ( f ) V ( f ) N I ( f ) N J ( f ) ? ? ? \tilde{Q}_V(f) = q \frac{\gamma_{IJ}^V(f) V(f)}{N_I(f)N_J(f)}???
Q ~ V ( f ) = q N I ( f ) N J ( f ) γ I J V ( f ) V ( f ) ? ? ?
ρ I J 2 = ⟨ R ⟩ 2 T o b s ⟨ γ I J I ( f ) I ( f ) + γ I J V ( f ) V ( f ) N I ( f ) N J ( f ) ∣ γ I J I ( f ) I ( f ) + γ I J V ( f ) V ( f ) N I ( f ) N J ( f ) ⟩ \rho_{\tiny IJ}^2 = \langle\mathcal{R}\rangle^2 T_{\rm obs} \left\langle \frac{\gamma_{IJ}^I(f) I(f) + \gamma_{IJ}^V(f) V(f)}{ N_I(f) N_J(f)} \Big|\frac{\gamma_{IJ}^I(f) I(f) + \gamma_{IJ}^V(f) V(f)}{ N_I(f) N_J(f)} \right\rangle
ρ I J 2 = ⟨ R ⟩ 2 T o b s ⟨ N I ( f ) N J ( f ) γ I J I ( f ) I ( f ) + γ I J V ( f ) V ( f ) ∣ ∣ ∣ ∣ N I ( f ) N J ( f ) γ I J I ( f ) I ( f ) + γ I J V ( f ) V ( f ) ⟩
ρ I J = ⟨ R ⟩ T o b s ∫ d f [ γ I J I ( f ) I ( f ) + γ I J V ( f ) V ( f ) ] 2 N I ( f ) N J ( f ) \boxed{\rho_{\tiny IJ} = \langle \mathcal{R} \rangle \sqrt{T_{\rm obs}} \sqrt{\int df \frac{ \Big[\gamma_{IJ}^I(f) I(f) + \gamma_{IJ}^V(f) V(f)\Big]^2}{N_I(f) N_J(f)} } }
ρ I J = ⟨ R ⟩ T o b s ∫ d f N I ( f ) N J ( f ) [ γ I J I ( f ) I ( f ) + γ I J V ( f ) V ( f ) ] 2
C i j ( f ) = d ~ i ( f ) d ~ j ∗ ( f ) \mathcal{C}_{ij}(f) = \tilde{d}_i(f)\tilde{d}^*_j(f)
C i j ( f ) = d ~ i ( f ) d ~ j ∗ ( f )
C i j = d ~ i d ~ j ∗ = s ~ i s ~ j ∗ + s ~ i n ~ j ∗ + n ~ i s ~ j ∗ + n ~ i n ~ j ∗ \mathcal{C}_{ij} = \tilde{d}_i\tilde{d}^*_j = \tilde{s}_i\tilde{s}^*_j + \tilde{s}_i\tilde{n}^*_j + \tilde{n}_i\tilde{s}^*_j + \tilde{n}_i\tilde{n}^*_j
C i j = d ~ i d ~ j ∗ = s ~ i s ~ j ∗ + s ~ i n ~ j ∗ + n ~ i s ~ j ∗ + n ~ i n ~ j ∗
对应前面数据互相关此,有f = f ′ f=f' f = f ′ , 对应δ ( f − f ′ ) = δ ( 0 ) \delta(f-f')=\delta(0) δ ( f − f ′ ) = δ ( 0 ) ,对于有限时间的观测数据 δ ( 0 ) → T o b s \delta(0)\rightarrow T_{\rm obs} δ ( 0 ) → T o b s
⟨ C i j ( f ) ⟩ = 4 π ⟨ R ⟩ T o b s [ γ i j I I ( f ) + γ i j V V ( f ) ] \langle \mathcal{C}_{ij}(f)\rangle = 4\pi\langle\mathcal{R}\rangle T_{\rm obs} [\gamma^I_{ij} I(f) + \gamma^V_{ij} V(f)]
⟨ C i j ( f ) ⟩ = 4 π ⟨ R ⟩ T o b s [ γ i j I I ( f ) + γ i j V V ( f ) ]
⟨ ( C A A C A E C E A C E E ) ⟩ = 4 π ⟨ R ⟩ T o b s ( γ A A I γ A A V γ A E I γ A E V γ E A I γ E A V γ E E I γ E E V ) ( I V ) \left\langle\begin{pmatrix}
\mathcal{C}_{AA} \\
\mathcal{C}_{AE}\\
\mathcal{C}_{EA} \\
\mathcal{C}_{EE}
\end{pmatrix}\right\rangle = 4\pi\langle\mathcal{R}\rangle T_{\rm obs}
\begin{pmatrix}
\gamma^I_{AA} & \gamma^V_{AA} \\
\gamma^I_{AE} & \gamma^V_{AE} \\
\gamma^I_{EA} & \gamma^V_{EA} \\
\gamma^I_{EE} & \gamma^V_{EE}
\end{pmatrix}
\begin{pmatrix}
I\\
\\
V
\end{pmatrix} ⟨ ⎝ ⎜ ⎜ ⎜ ⎛ C A A C A E C E A C E E ⎠ ⎟ ⎟ ⎟ ⎞ ⟩ = 4 π ⟨ R ⟩ T o b s ⎝ ⎜ ⎜ ⎜ ⎛ γ A A I γ A E I γ E A I γ E E I γ A A V γ A E V γ E A V γ E E V ⎠ ⎟ ⎟ ⎟ ⎞ ⎝ ⎛ I V ⎠ ⎞
⟨ C ⟩ = 4 π ⟨ R ⟩ T o b s M S , S = 1 4 π ⟨ R ⟩ T o b s M − 1 ⟨ C ⟩ \langle\mathcal{C}\rangle = 4\pi\langle\mathcal{R}\rangle T_{\rm obs} \mathcal{M} \mathcal{S}, ~~~~~ \mathcal{S} = \frac{1}{4\pi\langle\mathcal{R}\rangle T_{\rm obs}} \mathcal{M}^{-1} \langle\mathcal{C}\rangle
⟨ C ⟩ = 4 π ⟨ R ⟩ T o b s M S , S = 4 π ⟨ R ⟩ T o b s 1 M − 1 ⟨ C ⟩
这里C \mathcal{C} C 为随机变量,均值为上式。为简化协方差矩阵,将每个探测器对i j ij i j 记为单一标号κ \kappa κ ,则有:
N κ κ ′ ( f , f ′ ) = ⟨ C κ ( f ) C κ ′ ∗ ( f ′ ) ⟩ − ⟨ C κ ( f ) ⟩ ⟨ C κ ′ ∗ ( f ′ ) ⟩ \mathcal{N}_{\kappa\kappa'} (f,f') = \langle \mathcal{C}_\kappa(f) \mathcal{C}^*_{\kappa'}(f')\rangle - \langle\mathcal{C}_\kappa(f)\rangle \langle\mathcal{C}^*_{\kappa'}(f')\rangle
N κ κ ′ ( f , f ′ ) = ⟨ C κ ( f ) C κ ′ ∗ ( f ′ ) ⟩ − ⟨ C κ ( f ) ⟩ ⟨ C κ ′ ∗ ( f ′ ) ⟩
在弱信号近似下:
N κ κ ′ ( f , f ′ ) ≈ ⟨ n κ ( f , f ′ ) n κ ′ ∗ ( f , f ′ ) ⟩ = N κ ( f ) δ κ , κ ′ δ ( f − f ′ ) \mathcal{N}_{\kappa\kappa'(f,f')} \approx \langle \bm{n}_\kappa(f,f') \bm{n}^*_{\kappa'}(f, f') \rangle = N_{\kappa}(f) \delta_{\kappa,\kappa'} \delta(f-f')
N κ κ ′ ( f , f ′ ) ≈ ⟨ n κ ( f , f ′ ) n κ ′ ∗ ( f , f ′ ) ⟩ = N κ ( f ) δ κ , κ ′ δ ( f − f ′ )
即噪声的协方差矩阵 N \mathcal{N} N 为对角阵。注意,这里 n κ ( f , f ′ ) \bm{n}_\kappa(f,f') n κ ( f , f ′ ) 为一对探测器的噪声信号相关,即n i ( f ) n j ∗ ( f ′ ) n_i(f) n^*_j(f') n i ( f ) n j ∗ ( f ′ ) ,从而 ⟨ n κ ( f , f ′ ) n κ ′ ∗ ( f , f ′ ) ⟩ = ⟨ n i ( f ) n j ∗ ( f ) n k ∗ ( f ′ ) n l ( f ′ ) ⟩ \langle \bm{n}_\kappa(f,f') \bm{n}^*_{\kappa'}(f, f') \rangle = \langle n_i(f) n^*_j(f) n^*_k(f') n_l(f') \rangle ⟨ n κ ( f , f ′ ) n κ ′ ∗ ( f , f ′ ) ⟩ = ⟨ n i ( f ) n j ∗ ( f ) n k ∗ ( f ′ ) n l ( f ′ ) ⟩ ,只有当i j ij i j 与k l kl k l 相同,即κ \kappa κ 与κ ′ \kappa' κ ′ 相等(且f = f ′ f=f' f = f ′ )时,结果才非零,此时取值为N i ( f ) N j ( f ) N_i(f) N_j(f) N i ( f ) N j ( f ) ,这里记作N κ ( f ) N_\kappa(f) N κ ( f ) 。
这里δ ( 0 ) → T o b s \delta(0)\rightarrow T_{\rm obs} δ ( 0 ) → T o b s ,同样会出现时间。
最终对于信号S \mathcal{S} S ,引力波探测互相关数据C \mathcal{C} C 的似然函数:
p ( C ∣ S ) ∝ exp [ − 1 2 ( C − 4 π ⟨ R ⟩ T o b s M S ) T ( T o b s N ) − 1 ( C − 4 π ⟨ R ⟩ T o b s M S ) ] p(\mathcal{C}|\mathcal{S}) \propto \exp\left[-\frac{1}{2} \Big(\mathcal{C}- 4\pi\langle\mathcal{R}\rangle T_{\rm obs} \mathcal{M} \mathcal{S}\Big)^T \left(T_{\rm obs}\mathcal{N}\right)^{-1} \Big(\mathcal{C}- 4\pi\langle\mathcal{R}\rangle T_{\rm obs} \mathcal{M} \mathcal{S}\Big)\right]
p ( C ∣ S ) ∝ exp [ − 2 1 ( C − 4 π ⟨ R ⟩ T o b s M S ) T ( T o b s N ) − 1 ( C − 4 π ⟨ R ⟩ T o b s M S ) ]
对S \mathcal{S} S 求导,可得到S \mathcal{S} S 的最大化似然估计:
4 π ⟨ R ⟩ T o b s M T ( T o b s N ) − 1 ( C − 4 π ⟨ R ⟩ T o b s M S ^ ) + ( C − 4 π ⟨ R ⟩ T o b s M S ^ ) T ( T o b s N ) − 1 4 π ⟨ R ⟩ T o b s M = 0 4\pi\langle\mathcal{R}\rangle T_{\rm obs} \mathcal{M}^T \left(T_{\rm obs}\mathcal{N}\right)^{-1} \Big(\mathcal{C}- 4\pi\langle\mathcal{R}\rangle T_{\rm obs} \mathcal{M} \hat{\mathcal{S}}\Big) + \Big(\mathcal{C}- 4\pi\langle\mathcal{R}\rangle T_{\rm obs} \mathcal{M} \hat{\mathcal{S}}\Big)^T \left(T_{\rm obs}\mathcal{N}\right)^{-1} 4\pi\langle\mathcal{R}\rangle T_{\rm obs} \mathcal{M} = 0
4 π ⟨ R ⟩ T o b s M T ( T o b s N ) − 1 ( C − 4 π ⟨ R ⟩ T o b s M S ^ ) + ( C − 4 π ⟨ R ⟩ T o b s M S ^ ) T ( T o b s N ) − 1 4 π ⟨ R ⟩ T o b s M = 0
S ^ = F − 1 X , F = 4 π ⟨ R ⟩ T o b s M T ( T o b s N ) − 1 4 π ⟨ R ⟩ T o b s M , X = 4 π ⟨ R ⟩ T o b s M T ( T o b s N ) − 1 C \hat{\mathcal{S}} = \mathcal{F}^{-1} X, ~~~~ \mathcal{F} = 4\pi\langle\mathcal{R}\rangle T_{\rm obs} \mathcal{M}^T \left(T_{\rm obs}\mathcal{N}\right)^{-1} 4\pi\langle\mathcal{R}\rangle T_{\rm obs}\mathcal{M}, X = 4\pi\langle\mathcal{R}\rangle T_{\rm obs} \mathcal{M}^T \left(T_{\rm obs}\mathcal{N}\right)^{-1} \mathcal{C}
S ^ = F − 1 X , F = 4 π ⟨ R ⟩ T o b s M T ( T o b s N ) − 1 4 π ⟨ R ⟩ T o b s M , X = 4 π ⟨ R ⟩ T o b s M T ( T o b s N ) − 1 C
这里F \mathcal{F} F 是Fisher信息矩阵,具体参考下面正态分布Fisher信息矩阵的结论(目标参数为S \mathcal{S} S ,物理模型对参数的求导为4 π ⟨ R ⟩ T o b s M 4\pi\langle\mathcal{R}\rangle T_{\rm obs} \mathcal{M} 4 π ⟨ R ⟩ T o b s M ),由此信号协方差矩阵:
F = ( 4 π ⟨ R ⟩ ) 2 T o b s M T N − 1 M , X = 4 π ⟨ R ⟩ M T N − 1 C \mathcal{F} = (4\pi\langle\mathcal{R}\rangle)^2 T_{\rm obs} \mathcal{M}^T \mathcal{N}^{-1} \mathcal{M}, ~~~~ X = 4\pi\langle\mathcal{R}\rangle \mathcal{M}^T \mathcal{N}^{-1} \mathcal{C}
F = ( 4 π ⟨ R ⟩ ) 2 T o b s M T N − 1 M , X = 4 π ⟨ R ⟩ M T N − 1 C
F = ( 4 π ⟨ R ⟩ ) 2 T o b s ( ∑ ( γ κ I ) 2 N κ ∑ γ κ I γ κ V N κ ∑ γ κ V γ κ I N κ ∑ ( γ κ V ) 2 N κ ) \mathcal{F} = (4\pi\langle\mathcal{R}\rangle)^2 T_{\rm obs} \begin{pmatrix}
\sum \frac{(\gamma^I_\kappa)^2}{N_\kappa} & \sum \frac{\gamma^I_\kappa\gamma^V_\kappa}{N_\kappa}\\
\sum \frac{\gamma^V_\kappa\gamma^I_\kappa}{N_\kappa} & \sum \frac{(\gamma^V_\kappa)^2}{N_\kappa}
\end{pmatrix} F = ( 4 π ⟨ R ⟩ ) 2 T o b s ( ∑ N κ ( γ κ I ) 2 ∑ N κ γ κ V γ κ I ∑ N κ γ κ I γ κ V ∑ N κ ( γ κ V ) 2 )
Σ ≈ F − 1 \Sigma \approx \mathcal{F}^{-1}
Σ ≈ F − 1
F I I − 1 = F V V ∣ F ∣ , F V V − 1 = F I I ∣ F ∣ \mathcal{F}^{-1}_{II} = \frac{\mathcal{F}_{VV}}{|\mathcal{F}|}, ~~~~ \mathcal{F}^{-1}_{VV} = \frac{\mathcal{F}_{II}}{|\mathcal{F}|}
F I I − 1 = ∣ F ∣ F V V , F V V − 1 = ∣ F ∣ F I I
S N R I 2 = ∫ d f I ^ 2 ( f ) σ I ^ 2 ( f ) = ( 4 π ⟨ R ⟩ ) 2 T o b s ∫ d f ∣ F ∣ F V V I 2 SNR_I^2 = \int df \frac{\hat{I}^2(f)}{\sigma^2_{\hat{I}}(f)} = (4\pi\langle\mathcal{R}\rangle)^2 T_{\rm obs} \int df \frac{|\mathcal{F}|}{\mathcal{F}_{VV}}I^2
S N R I 2 = ∫ d f σ I ^ 2 ( f ) I ^ 2 ( f ) = ( 4 π ⟨ R ⟩ ) 2 T o b s ∫ d f F V V ∣ F ∣ I 2
S N R V 2 = ∫ d f V ^ 2 ( f ) σ V ^ 2 ( f ) = ( 4 π ⟨ R ⟩ ) 2 T o b s ∫ d f ∣ F ∣ F I I V 2 SNR_V^2 = \int df \frac{\hat{V}^2(f)}{\sigma^2_{\hat{V}}(f)} = (4\pi\langle\mathcal{R}\rangle)^2 T_{\rm obs} \int df \frac{|\mathcal{F}|}{\mathcal{F}_{II}}V^2
S N R V 2 = ∫ d f σ V ^ 2 ( f ) V ^ 2 ( f ) = ( 4 π ⟨ R ⟩ ) 2 T o b s ∫ d f F I I ∣ F ∣ V 2
S N R I 2 = ( 4 π ⟨ R ⟩ ) 2 T o b s ∫ d f [ ∑ κ ( γ κ I ) 2 / N κ ] [ ∑ κ ( γ κ V ) 2 / N κ ] − ( ∑ κ γ κ I γ κ V / N κ ) 2 ∑ κ ( γ κ V ) 2 / N κ I 2 SNR_I^2 =(4\pi\langle\mathcal{R}\rangle)^2 T_{\rm obs} \int df \frac{\big[\sum_\kappa (\gamma^I_\kappa)^2/N_\kappa\big] \big[\sum_\kappa (\gamma^V_\kappa)^2/N_\kappa\big] - \big(\sum_\kappa \gamma^I_\kappa\gamma^V_\kappa/N_\kappa\big)^2}{\sum_\kappa (\gamma^V_\kappa)^2/N_\kappa} I^2
S N R I 2 = ( 4 π ⟨ R ⟩ ) 2 T o b s ∫ d f ∑ κ ( γ κ V ) 2 / N κ [ ∑ κ ( γ κ I ) 2 / N κ ] [ ∑ κ ( γ κ V ) 2 / N κ ] − ( ∑ κ γ κ I γ κ V / N κ ) 2 I 2
S N R V 2 = ( 4 π ⟨ R ⟩ ) 2 T o b s ∫ d f [ ∑ κ ( γ κ I ) 2 / N κ ] [ ∑ κ ( γ κ V ) 2 / N κ ] − ( ∑ κ γ κ I γ κ V / N κ ) 2 ∑ κ ( γ κ I ) 2 / N κ V 2 SNR_V^2 =(4\pi\langle\mathcal{R}\rangle)^2 T_{\rm obs} \int df \frac{\big[\sum_\kappa (\gamma^I_\kappa)^2/N_\kappa\big] \big[\sum_\kappa (\gamma^V_\kappa)^2/N_\kappa\big] - \big(\sum_\kappa \gamma^I_\kappa\gamma^V_\kappa/N_\kappa\big)^2}{\sum_\kappa (\gamma^I_\kappa)^2/N_\kappa} V^2
S N R V 2 = ( 4 π ⟨ R ⟩ ) 2 T o b s ∫ d f ∑ κ ( γ κ I ) 2 / N κ [ ∑ κ ( γ κ I ) 2 / N κ ] [ ∑ κ ( γ κ V ) 2 / N κ ] − ( ∑ κ γ κ I γ κ V / N κ ) 2 V 2
对于LISA-Taiji联网,四个有效通道对的噪声功率完全相同,从而:
S N R I 2 = ( 4 π ⟨ R ⟩ ) 2 T o b s ∫ d f [ ∑ κ ( γ κ I ) 2 ] [ ∑ κ ( γ κ V ) 2 ] − ( ∑ κ γ κ I γ κ V ) 2 ∑ κ ( γ κ V ) 2 I 2 / N SNR_I^2 =(4\pi\langle\mathcal{R}\rangle)^2 T_{\rm obs} \int df \frac{\big[\sum_\kappa (\gamma^I_\kappa)^2\big] \big[\sum_\kappa (\gamma^V_\kappa)^2\big] - \big(\sum_\kappa \gamma^I_\kappa\gamma^V_\kappa\big)^2}{\sum_\kappa (\gamma^V_\kappa)^2} I^2/N
S N R I 2 = ( 4 π ⟨ R ⟩ ) 2 T o b s ∫ d f ∑ κ ( γ κ V ) 2 [ ∑ κ ( γ κ I ) 2 ] [ ∑ κ ( γ κ V ) 2 ] − ( ∑ κ γ κ I γ κ V ) 2 I 2 / N
S N R V 2 = ( 4 π ⟨ R ⟩ ) 2 T o b s ∫ d f [ ∑ κ ( γ κ I ) 2 ] [ ∑ κ ( γ κ V ) 2 ] − ( ∑ κ γ κ I γ κ V ) 2 ∑ κ ( γ κ I ) 2 V 2 / N SNR_V^2 =(4\pi\langle\mathcal{R}\rangle)^2 T_{\rm obs} \int df \frac{\big[\sum_\kappa (\gamma^I_\kappa)^2\big] \big[\sum_\kappa (\gamma^V_\kappa)^2\big] - \big(\sum_\kappa \gamma^I_\kappa\gamma^V_\kappa\big)^2}{\sum_\kappa (\gamma^I_\kappa)^2} V^2/N
S N R V 2 = ( 4 π ⟨ R ⟩ ) 2 T o b s ∫ d f ∑ κ ( γ κ I ) 2 [ ∑ κ ( γ κ I ) 2 ] [ ∑ κ ( γ κ V ) 2 ] − ( ∑ κ γ κ I γ κ V ) 2 V 2 / N
此时可引入有效重叠约化函数,更直观地描述探测器网络对于I、V分量的灵敏度
γ I , e f f = [ ∑ κ ( γ κ I ) 2 ] [ ∑ κ ( γ κ V ) 2 ] − ( ∑ κ γ κ I γ κ V ) 2 ∑ κ ( γ κ V ) 2 = ∑ κ ( γ κ I ) 2 − ( ∑ κ γ κ I γ κ V ) 2 ∑ κ ( γ κ V ) 2 \gamma_{I, \rm eff} = \sqrt{\frac{\big[\sum_\kappa (\gamma^I_\kappa)^2\big] \big[\sum_\kappa (\gamma^V_\kappa)^2\big] - \big(\sum_\kappa \gamma^I_\kappa\gamma^V_\kappa\big)^2}{\sum_\kappa (\gamma^V_\kappa)^2}} = \sqrt{\sum_\kappa (\gamma^I_\kappa)^2 - \frac{\big(\sum_\kappa \gamma^I_\kappa\gamma^V_\kappa\big)^2}{\sum_\kappa (\gamma^V_\kappa)^2}}
γ I , e f f = ∑ κ ( γ κ V ) 2 [ ∑ κ ( γ κ I ) 2 ] [ ∑ κ ( γ κ V ) 2 ] − ( ∑ κ γ κ I γ κ V ) 2 = κ ∑ ( γ κ I ) 2 − ∑ κ ( γ κ V ) 2 ( ∑ κ γ κ I γ κ V ) 2
γ V , e f f = [ ∑ κ ( γ κ I ) 2 ] [ ∑ κ ( γ κ V ) 2 ] − ( ∑ κ γ κ I γ κ V ) 2 ∑ κ ( γ κ I ) 2 = ∑ κ ( γ κ V ) 2 − ( ∑ κ γ κ I γ κ V ) 2 ∑ κ ( γ κ I ) 2 \gamma_{V, \rm eff} = \sqrt{\frac{\big[\sum_\kappa (\gamma^I_\kappa)^2\big] \big[\sum_\kappa (\gamma^V_\kappa)^2\big] - \big(\sum_\kappa \gamma^I_\kappa\gamma^V_\kappa\big)^2}{\sum_\kappa (\gamma^I_\kappa)^2}} = \sqrt{\sum_\kappa (\gamma^V_\kappa)^2 - \frac{\big(\sum_\kappa \gamma^I_\kappa\gamma^V_\kappa\big)^2}{\sum_\kappa (\gamma^I_\kappa)^2}}
γ V , e f f = ∑ κ ( γ κ I ) 2 [ ∑ κ ( γ κ I ) 2 ] [ ∑ κ ( γ κ V ) 2 ] − ( ∑ κ γ κ I γ κ V ) 2 = κ ∑ ( γ κ V ) 2 − ∑ κ ( γ κ I ) 2 ( ∑ κ γ κ I γ κ V ) 2
S N R I 2 = ( 4 π ⟨ R ⟩ ) 2 T o b s ∫ d f γ I , e f f 2 I 2 N SNR_I^2 =(4\pi\langle\mathcal{R}\rangle)^2 T_{\rm obs} \int df \frac{\gamma^2_{I, \rm eff} I^2}{N}
S N R I 2 = ( 4 π ⟨ R ⟩ ) 2 T o b s ∫ d f N γ I , e f f 2 I 2
S N R V 2 = ( 4 π ⟨ R ⟩ ) 2 T o b s ∫ d f γ V , e f f 2 V 2 N SNR_V^2 =(4\pi\langle\mathcal{R}\rangle)^2 T_{\rm obs} \int df \frac{\gamma^2_{V, \rm eff} V^2}{N}
S N R V 2 = ( 4 π ⟨ R ⟩ ) 2 T o b s ∫ d f N γ V , e f f 2 V 2
在背景各项异性的天图分析中,这里X X X 通常被称为脏图(dirty map),S ^ \hat{S} S ^ 则称为净图(clean map)。X = M T N − 1 C ∼ M T N − 1 M S ∼ F S X = \mathcal{M}^T\mathcal{N}^{-1}\mathcal{C} \sim \mathcal{M}^T\mathcal{N}^{-1}\mathcal{M}\mathcal{S} \sim \mathcal{F}\mathcal{S} X = M T N − 1 C ∼ M T N − 1 M S ∼ F S ,包含了信号与探测器关联响应,即重叠约化函数的卷积(频域相乘等于时域卷积),以及噪声,而S ^ \hat{S} S ^ 则时经过去卷积出的信号。
如果背景是完全圆偏振的,I ( f ) = ± V ( f ) I(f) = \pm V(f) I ( f ) = ± V ( f ) ,无需要再拆分两者,从而两个探测器的重叠约化可取为γ I J I ± γ I J V \gamma^I_{IJ} \pm \gamma^V_{IJ} γ I J I ± γ I J V ,总的有效重叠约化为
∑ κ ( γ κ I ± γ κ V ) 2 = ∑ κ ( γ κ I ) 2 + ∑ κ ( γ κ I ) 2 ± 2 ∑ κ γ κ I γ κ V \sqrt{\sum_\kappa \left( \gamma^I_\kappa \pm \gamma^V_\kappa \right)^2} = \sqrt{\sum_\kappa ( \gamma^I_\kappa)^2 + \sum_\kappa ( \gamma^I_\kappa)^2 \pm 2\sum_\kappa \gamma^I_\kappa\gamma^V_\kappa}
κ ∑ ( γ κ I ± γ κ V ) 2 = κ ∑ ( γ κ I ) 2 + κ ∑ ( γ κ I ) 2 ± 2 κ ∑ γ κ I γ κ V
critical frequency f = c 2 D ∼ 1.5 m H z f = \frac{c}{2 D}\sim 1.5\rm mHz f = 2 D c ∼ 1 . 5 m H z Wang 2021??
Fisher信息
得分函数 对数似然的梯度
s ( θ ; x ) ≡ ∇ θ log p ( x ; θ ) s(\theta;x) \equiv \nabla_\theta \log p(x; \theta)
s ( θ ; x ) ≡ ∇ θ log p ( x ; θ )
这里下脚标θ \theta θ 代表在参数空间(对参数θ \theta θ 各分量)求梯度,而非方向导数。得分函数反映了似然函数/模型在特定θ \theta θ 处对参数变化的敏感程度。但其具体取值依赖随机变量x x x ,对x x x 的分布求期望,在真实参数处平均取值为0。此时,得分函数的方差/协方差矩阵更好的反映了平均的敏感程度,被称为Fisher信息(矩阵) 。
E ( s ∣ θ ) = ∫ ∇ θ log p ( x ; θ ) p ( x ; θ ) d x = ∫ ∇ θ p ( x ; θ ) d x = ∇ θ ∫ p ( x ; θ ) d x = 0 E ( s s T ∣ θ ) = ∫ [ ∇ θ log p ( x ; θ ) ∇ θ T log p ( x ; θ ) ] p ( x ; θ ) d x = ∫ ∇ θ p ( x ; θ ) ∇ θ T log p ( x ; θ ) d x = ∫ { ∇ θ [ p ( x ; θ ) ∇ θ T log p ( x ; θ ) ] − p ( x ; θ ) ∇ θ ∇ θ T log p ( x ; θ ) } d x = ∫ ∇ θ ∇ θ T p ( x ; θ ) d x − ∫ p ( x ; θ ) ∇ θ ∇ θ T log p ( x ; θ ) d x = ∇ θ ∇ θ T ∫ p ( x ; θ ) d x − ∫ p ( x ; θ ) ∇ θ ∇ θ T log p ( x ; θ ) d x = − E ( ∇ θ ∇ θ T log p ( x ; θ ) ∣ θ ) \begin{aligned}
E(s|\theta) &= \int \nabla_\theta \log p(x; \theta) ~ p(x;\theta) dx\\
&= \int \nabla_\theta p(x; \theta) dx
= \nabla_\theta \int p(x; \theta) dx\\
& = 0\\
E(s s^T|\theta) &= \int \Big[\nabla_\theta \log p(x; \theta) ~ \nabla_\theta^T \log p(x; \theta)\Big] ~ p(x; \theta) dx\\
&= \int \nabla_\theta~ p(x; \theta) ~ \nabla_\theta^T \log p(x; \theta) dx\\
&= \int \left\{\nabla_\theta \Big[p(x;\theta) ~ \nabla_\theta^T \log p(x; \theta)\Big] - p(x;\theta)~ \nabla_\theta \nabla_\theta^T \log p(x; \theta)\right\}dx\\
&= \int \nabla_\theta \nabla_\theta^T p(x; \theta) dx - \int p(x;\theta)~ \nabla_\theta \nabla_\theta^T \log p(x; \theta)dx\\
&= \nabla_\theta \nabla_\theta^T \int p(x; \theta) dx - \int p(x;\theta)~ \nabla_\theta \nabla_\theta^T \log p(x; \theta)dx\\
&= - E\left(\nabla_\theta \nabla_\theta^T \log p(x; \theta) \Big| \theta\right)
\end{aligned} E ( s ∣ θ ) E ( s s T ∣ θ ) = ∫ ∇ θ log p ( x ; θ ) p ( x ; θ ) d x = ∫ ∇ θ p ( x ; θ ) d x = ∇ θ ∫ p ( x ; θ ) d x = 0 = ∫ [ ∇ θ log p ( x ; θ ) ∇ θ T log p ( x ; θ ) ] p ( x ; θ ) d x = ∫ ∇ θ p ( x ; θ ) ∇ θ T log p ( x ; θ ) d x = ∫ { ∇ θ [ p ( x ; θ ) ∇ θ T log p ( x ; θ ) ] − p ( x ; θ ) ∇ θ ∇ θ T log p ( x ; θ ) } d x = ∫ ∇ θ ∇ θ T p ( x ; θ ) d x − ∫ p ( x ; θ ) ∇ θ ∇ θ T log p ( x ; θ ) d x = ∇ θ ∇ θ T ∫ p ( x ; θ ) d x − ∫ p ( x ; θ ) ∇ θ ∇ θ T log p ( x ; θ ) d x = − E ( ∇ θ ∇ θ T log p ( x ; θ ) ∣ ∣ ∣ ∣ θ )
这里∇ ∇ T \nabla \nabla^T ∇ ∇ T 对应黑赛矩阵,有时也被简记为∇ 2 \nabla^2 ∇ 2 wiki ,注意与通常的拉普拉斯算符区分。
根据上述推导:
Γ ≡ C o v ( s ) = E ( ∇ θ log p ( x ; θ ) ∇ θ T log p ( x ; θ ) ∣ θ ) = − E ( ∇ θ ∇ θ T log p ( x ; θ ) ∣ θ ) \small \boxed{\Gamma \equiv {\rm Cov}(s) = E\left( \nabla_\theta \log p(x; \theta) ~ \nabla_\theta^T \log p(x; \theta) \Big| \theta\right) = - E\left( \nabla_\theta \nabla_\theta^T \log p(x; \theta) \Big| \theta\right)}
Γ ≡ C o v ( s ) = E ( ∇ θ log p ( x ; θ ) ∇ θ T log p ( x ; θ ) ∣ ∣ ∣ ∣ θ ) = − E ( ∇ θ ∇ θ T log p ( x ; θ ) ∣ ∣ ∣ ∣ θ )
上述两种表达都可用于计算Fisher信息矩阵,根据具体情况选择,无法解析计算也可通过蒙特卡罗多次采样求均值近似。在参数空间中对数似然在峰值,即真实参数附近梯度为0,此时黑赛矩阵反映了曲率,或者说度规张量。因此,Fisher信息矩阵在几何上,对应于参数空间中对数似然超曲面在真实参数附近的曲率,也即前面讲的参数敏感程度。注意,得分函数作为对数似然的导数,与具体的采样或观测数据有关,但其期望及方差仅由分布决定,与具体数据无关。
V a r ( X + Y ) = E { [ X + Y − E ( X ) − E ( Y ) ] 2 } = V a r ( X ) + V a r ( Y ) + 2 C o v ( X , Y ) \footnotesize {\rm Var}(X+Y) = E\left\{[X+Y - E(X)-E(Y)]^2\right\} = {\rm Var}(X) + {\rm Var}(Y) + 2{\rm Cov}(X,Y) V a r ( X + Y ) = E { [ X + Y − E ( X ) − E ( Y ) ] 2 } = V a r ( X ) + V a r ( Y ) + 2 C o v ( X , Y ) 。对于相互独立的随机变量,协方差为0,从而独立随机变量和的方差为各自方差之和。对于多个独立事件,似然函数为所有事件似然概率的乘积,从而得分函数为各自得分函数之和。进而得分函数的方差,即总的Fisher信息矩阵,为各自信息矩阵之和。概括起来,对于独立事件,Fisher信息矩阵具有可加性。尤其是对于独立同分布的观测数据,N次观测的Fisher信息矩阵为单次观测的N倍。
这种可加性对于先验信息同样成立,参考下面的克拉美-罗下界,对于无偏估计,Fisher信息对应参数分布协方差矩阵的逆。由此,对参数先验分布的协方差矩阵求逆可获得先验Fisher信息,将其与由似然函数获得的信息相加,即为总的后验Fisher信息,再次取逆可得到参数的后验分布的协方差。实际中通常只对个别参数有先验分布,此时可近似将参数方差的倒数与Fisher信息矩阵对应的对角元相加。
进一步的,信息可加性还可用于不同实验结果的整合。这些实验可涉及不同的模型以及观测量,尤其是可能存在不同的冗余参数,为避免合并矩阵维度过高,可边缘化(marginalize)冗余参数,但需要注意不同实验的冗余参数间是否存在关联性:对于无关的冗余参数,可先对相应参数进行边缘化操作,之后对边缘化之后的Fisher信息矩阵求和;对相同或相关的冗余参数,则必须先对Fisher信息矩阵求和,再边缘化冗余参数。边缘化参数需先求逆计算协方差矩阵,从中移除冗余参数,再进行逆运算获得边缘化后的Fisher信息矩阵。这其中的问题在于反复求逆,数值上不太稳定,另一种选择是Γ ‾ = Γ p p − Γ p r Γ r r − 1 Γ r p \overline{\Gamma} = \Gamma_{pp} - \Gamma_{pr}\Gamma_{rr}^{-1}\Gamma_{rp} Γ = Γ p p − Γ p r Γ r r − 1 Γ r p ,这里下标指示了Fisher矩阵的不同分块,p p p 代表目标参数,r r r 代表冗余参数。
注意:参数边缘化需要从协方差矩阵 中移除参数,与之相对的,直接从Fisher信息矩阵 中移除参数所对应的是完全固定相关参数。
克拉美-罗下界 (Cramér–Rao Bound, CRB)
多维参数情况,参照柯西-施瓦茨不等式的矩阵形式⟨ u , u ⟩ ⪰ ⟨ u , v ⟩ ⟨ v , v ⟩ − 1 ⟨ v , u ⟩ \langle u,u \rangle \succeq \langle u,v \rangle \langle v,v \rangle^{-1}\langle v,u\rangle ⟨ u , u ⟩ ⪰ ⟨ u , v ⟩ ⟨ v , v ⟩ − 1 ⟨ v , u ⟩ :
C o v ( θ ^ ∣ θ ) ≥ C o v ( θ ^ , s ∣ θ ) C o v ( s ) − 1 C o v ( s , θ ^ ∣ θ ) {\rm Cov}(\hat{\theta}|\theta) \ge {\rm Cov}(\hat{\theta}, s|\theta) {\rm Cov}(s)^{-1} {\rm Cov}(s, \hat{\theta}|\theta)
C o v ( θ ^ ∣ θ ) ≥ C o v ( θ ^ , s ∣ θ ) C o v ( s ) − 1 C o v ( s , θ ^ ∣ θ )
C o v ( s , θ ^ ∣ θ ) = E { [ s − E ( s ) ] [ θ ^ − E ( θ ^ ) ] T ∣ θ } = E { s [ θ ^ − E ( θ ^ ) ] T ∣ θ } = E ( s θ ^ T ∣ θ ) − E ( s ∣ θ ) E ( θ ^ ∣ θ ) T = E ( s θ ^ T ∣ θ ) = ∫ [ ∇ θ log p ( x ; θ ) ] θ ^ T p ( x ; θ ) d x = ∫ [ ∇ θ p ( x ; θ ) ] θ ^ T d x = ∇ θ ∫ p ( x ; θ ) θ ^ T d x = ∇ E ( θ ^ ∣ θ ) C o v ( θ ^ , s ∣ θ ) = E ( θ ^ s T ∣ θ ) = E ( s θ ^ T ∣ θ ) T = [ ∇ E ( θ ^ ∣ θ ) ] T \begin{aligned}
{\rm Cov}(s, \hat{\theta} | \theta) &= E\left\{ \big[s -E(s)\big] \big[\hat{\theta}-E(\hat{\theta})\big]^T \bigg|\theta \right\}\\
&= E\left\{ s \big[\hat{\theta}-E(\hat{\theta})\big]^T \bigg|\theta \right\} = E(s~ \hat{\theta}^T|\theta) - E(s|\theta) E(\hat{\theta}|\theta)^T\\
&= E(s~ \hat{\theta}^T|\theta) = \int \Big[\nabla_\theta \log p(x; \theta)\Big] ~ \hat{\theta}^T p(x; \theta) dx\\
&= \int \Big[\nabla_\theta ~ p(x; \theta)\Big] ~ \hat{\theta}^T dx
= \nabla_\theta \int p(x; \theta) \hat{\theta}^T dx\\
&= \nabla E(\hat{\theta}|\theta)\\
{\rm Cov}(\hat{\theta}, s | \theta) &= E(\hat{\theta} s^T|\theta) = E(s \hat{\theta}^T|\theta)^T = \left[\nabla E(\hat{\theta}|\theta)\right]^T
\end{aligned} C o v ( s , θ ^ ∣ θ ) C o v ( θ ^ , s ∣ θ ) = E { [ s − E ( s ) ] [ θ ^ − E ( θ ^ ) ] T ∣ ∣ ∣ ∣ ∣ θ } = E { s [ θ ^ − E ( θ ^ ) ] T ∣ ∣ ∣ ∣ ∣ θ } = E ( s θ ^ T ∣ θ ) − E ( s ∣ θ ) E ( θ ^ ∣ θ ) T = E ( s θ ^ T ∣ θ ) = ∫ [ ∇ θ log p ( x ; θ ) ] θ ^ T p ( x ; θ ) d x = ∫ [ ∇ θ p ( x ; θ ) ] θ ^ T d x = ∇ θ ∫ p ( x ; θ ) θ ^ T d x = ∇ E ( θ ^ ∣ θ ) = E ( θ ^ s T ∣ θ ) = E ( s θ ^ T ∣ θ ) T = [ ∇ E ( θ ^ ∣ θ ) ] T
从而,参数估计协方差的下界为:
Σ θ ^ ≡ C o v ( θ ^ ∣ θ ) ≥ [ ∇ E ( θ ^ ∣ θ ) ] T Γ − 1 ∇ E ( θ ^ ∣ θ ) \Sigma_{\hat{\theta}} \equiv {\rm Cov}(\hat{\theta}|\theta) \ge \left[\nabla E(\hat{\theta}|\theta)\right] ^T ~ \Gamma^{-1} ~ \nabla E(\hat{\theta}|\theta)
Σ θ ^ ≡ C o v ( θ ^ ∣ θ ) ≥ [ ∇ E ( θ ^ ∣ θ ) ] T Γ − 1 ∇ E ( θ ^ ∣ θ )
在给定θ \theta θ 情况下,假设θ ^ \hat{\theta} θ ^ 为无偏 估计,有E ( θ ^ ∣ θ ) = θ E(\hat{\theta}|\theta)=\theta E ( θ ^ ∣ θ ) = θ ,从而C o v ( s , θ ^ ∣ θ ) = I {\rm Cov}(s, \hat{\theta}|\theta) = I C o v ( s , θ ^ ∣ θ ) = I ,为单位矩阵。此时,协方差下界简化为Fisher信息矩阵的逆:
Σ θ ^ ≥ Γ − 1 \Sigma_{\hat{\theta}} \ge \Gamma^{-1}
Σ θ ^ ≥ Γ − 1
实际应用时,若Fisher信息矩阵的逆难以计算,也可直接计算Fisher信息矩阵对角元的倒数,获得关于参数方差的宽松下界wiki : σ 2 θ ^ = ( Σ θ ^ ) i i ≥ ( Γ − 1 ) i i ≥ ( Γ i i ) − 1 {\sigma^2}_{\hat{\theta}} = (\Sigma_{\hat{\theta}})_{ii} \ge (\Gamma^{-1})_{ii} \ge (\Gamma_{ii})^{-1} σ 2 θ ^ = ( Σ θ ^ ) i i ≥ ( Γ − 1 ) i i ≥ ( Γ i i ) − 1 。此外需注意,Fisher信息矩阵并非所实际能获得精度的下界,该结论仅针对无偏估计,利用一些有偏估计量反而可能获得更好的参数限制。
对于正态分布 p ( x ; θ ) = 1 ( 2 π ) k ∣ Σ ( θ ) ∣ exp { − 1 2 [ x − μ ( θ ) ] T Σ ( θ ) − 1 [ x − μ ( θ ) ] } p(x;\theta) = \frac{1}{ \sqrt{(2\pi)^k|\Sigma(\theta)|} } \exp \left\{-\frac{1}{2}[x-\mu(\theta)]^T\Sigma(\theta)^{-1} [x-\mu(\theta)]\right\} p ( x ; θ ) = ( 2 π ) k ∣ Σ ( θ ) ∣ 1 exp { − 2 1 [ x − μ ( θ ) ] T Σ ( θ ) − 1 [ x − μ ( θ ) ] } 。对应的Fisher信息矩阵可参考Wikipedia :
Γ i j = [ ( ∇ μ ) T Σ − 1 ∇ μ ] i j + 1 2 T r [ Σ θ − 1 ( ∂ θ i Σ ) Σ − 1 ( ∂ θ j Σ ) ] \Gamma_{ij} = \Big[(\nabla \mu)^T \Sigma^{-1} \nabla \mu\Big]_{ij} + \frac{1}{2}{\rm Tr}\Big[\Sigma_\theta^{-1}~(\partial_{\theta_i}\Sigma) ~\Sigma^{-1} ~ (\partial_{\theta_j}\Sigma)\Big]
Γ i j = [ ( ∇ μ ) T Σ − 1 ∇ μ ] i j + 2 1 T r [ Σ θ − 1 ( ∂ θ i Σ ) Σ − 1 ( ∂ θ j Σ ) ]
其中T r \rm Tr T r 为求迹。注意,这里Σ \Sigma Σ 为残差/噪声的协方差,维度对应于数据维度,不要与前面所讨论的参数估计协方差混淆。后者对于无偏估计,以信息矩阵的逆Γ − 1 \Gamma^{-1} Γ − 1 为下界,维度对应参数维度。暂时忽略后一项,这里前一项反映了参数变换时Fisher信息矩阵的变换 :变量x x x 变换至θ \theta θ ,对应于Fisher信息矩阵Γ θ = ( ∇ θ x ) T Γ x ∇ θ x \Gamma_\theta = (\nabla_\theta x)^T \Gamma_x \nabla_\theta x Γ θ = ( ∇ θ x ) T Γ x ∇ θ x 。注意在x , θ x, \theta x , θ 均为多维变量时,∇ θ x \nabla_\theta x ∇ θ x 即变量代换的雅可比矩阵。考虑到x x x 的协方差为Σ \Sigma Σ ,对应Fisher信息Σ − 1 \Sigma^{-1} Σ − 1 ,而变量代换通过模型μ ( θ ) \mu(\theta) μ ( θ ) 体现,最终θ \theta θ 信息矩阵为( ∇ μ ) T Σ − 1 ∇ μ (\nabla \mu)^T \Sigma^{-1} \nabla \mu ( ∇ μ ) T Σ − 1 ∇ μ ,其逆对应新参数的协方差。
数据分析时,x x x 为观测数据,μ ( θ ) \mu(\theta) μ ( θ ) 对应信号的物理模型,∇ μ \nabla \mu ∇ μ 为模型对参数的梯度。若各次观测的噪声无关,θ \theta θ 仅为物理参数,而残差的协方差Σ \Sigma Σ 与物理参数无关,因此信息矩阵仅有前一项。反之,如果多次观测噪声相关,θ \theta θ 包含物理参数α \alpha α 及噪声参数β \beta β ,同时噪声与信号不相关(两者不依赖共同参数),则Fisher信息矩阵为分块对角:其中物理参数部分,与残差噪声无关,因此后一项为0;噪声参数部分,与物理信号无关,因此前一项为0。
引力波分析中,假设噪声为稳态高斯过程,一次具体观测(relization)的似然概率:
p ( d ∣ θ ) ∝ exp [ − 1 2 ∫ d f [ d ~ ( f ) − s ~ ( f ; θ ) ] ∗ [ d ~ ( f ) − s ~ ( f ; θ ) ] 1 / 2 S n ( f ) ] p(d|\theta) \propto \exp\left[ -\frac{1}{2} \int df \frac{[\tilde{d}(f)-\tilde{s}(f;\theta)]^*[\tilde{d}(f)-\tilde{s}(f;\theta)]}{1/2S_n(f)} \right]
p ( d ∣ θ ) ∝ exp [ − 2 1 ∫ d f 1 / 2 S n ( f ) [ d ~ ( f ) − s ~ ( f ; θ ) ] ∗ [ d ~ ( f ) − s ~ ( f ; θ ) ] ]
这里1 / 2 S n ( f ) 1/2S_n(f) 1 / 2 S n ( f ) 为不同频率处探测器噪声功率,即相应频率的噪声方差。指数上对f f f 的积分,相当于不同频率处概率相乘,对应不同时间全部观测的总体概率。
引入内积定义
⟨ a ∣ b ⟩ ≡ ∫ a ~ ∗ ( f ) b ~ ( f ) 1 / 2 S n ( f ) d f = 4 R ∫ 0 ∞ a ~ ∗ ( f ) b ~ ( f ) S n ( f ) d f \langle a|b\rangle \equiv \int \frac{\tilde{a}^*(f)\tilde{b}(f)}{1/2 S_n(f)}df = 4\mathfrak{R} \int_0^\infty \frac{\tilde{a}^*(f)\tilde{b}(f)}{S_n(f)}df
⟨ a ∣ b ⟩ ≡ ∫ 1 / 2 S n ( f ) a ~ ∗ ( f ) b ~ ( f ) d f = 4 R ∫ 0 ∞ S n ( f ) a ~ ∗ ( f ) b ~ ( f ) d f
上述似然概率可简化为:
p ( d ∣ θ ) ∝ exp [ − 1 2 ⟨ d − s ( θ ) ∣ d − s ( θ ) ⟩ ] p(d|\theta) \propto \exp\left[ -\frac{1}{2} \langle d-s(\theta)|d-s(\theta)\rangle \right]
p ( d ∣ θ ) ∝ exp [ − 2 1 ⟨ d − s ( θ ) ∣ d − s ( θ ) ⟩ ]
对于多探测联合观测,可类似定义
⟨ A ∣ B ⟩ ≡ 4 R ∫ 0 ∞ A ~ † ( f ) N − 1 B ~ ( f ) d f \langle A|B\rangle \equiv 4\mathfrak{R} \int_0^\infty \tilde{A}^\dagger(f) \mathcal{N}^{-1} \tilde{B}(f) df
⟨ A ∣ B ⟩ ≡ 4 R ∫ 0 ∞ A ~ † ( f ) N − 1 B ~ ( f ) d f
这里N \mathcal{N} N 探测器噪声协方差矩阵,⟨ n ~ a ( f ) n ~ b ∗ ( f ′ ) ⟩ = 1 2 N a b δ ( f − f ′ ) \langle \tilde{n}_a(f) \tilde{n}^*_b(f')\rangle = \frac{1}{2}\mathcal{N}_{ab}\delta(f-f') ⟨ n ~ a ( f ) n ~ b ∗ ( f ′ ) ⟩ = 2 1 N a b δ ( f − f ′ ) 。假设不同探测器间噪声不相关,则N \mathcal{N} N 为对角阵,⟨ n ~ a ( f ) n ~ b ∗ ( f ′ ) ⟩ = 1 2 N a b δ a , b δ ( f − f ′ ) \langle \tilde{n}_a(f) \tilde{n}^*_b(f')\rangle = \frac{1}{2}\mathcal{N}_{ab}\delta_{a,b}\delta(f-f') ⟨ n ~ a ( f ) n ~ b ∗ ( f ′ ) ⟩ = 2 1 N a b δ a , b δ ( f − f ′ ) 。
根据正态分布的结论,如果不考虑对f f f 的积分,对单个频率Γ = ( ∇ s ) † ( 1 2 N ) − 1 ∇ s \Gamma = (\nabla s)^\dagger (\frac{1}{2}\mathcal{N})^{-1} \nabla s Γ = ( ∇ s ) † ( 2 1 N ) − 1 ∇ s 。而考虑到Fisher信息的可加性:
Γ = ∫ d f ( ∇ s ) † ( 1 / 2 N ) − 1 ∇ s = ⟨ ∇ s ∣ ∇ s ⟩ \Gamma = \int df (\nabla s)^\dagger~ (1/2~\mathcal{N})^{-1} ~\nabla s = \langle \nabla s|\nabla s\rangle
Γ = ∫ d f ( ∇ s ) † ( 1 / 2 N ) − 1 ∇ s = ⟨ ∇ s ∣ ∇ s ⟩
也可直接由Fisher信息的定义式出发,以单探测器情况为例:
Γ = E [ ∇ θ log p ( d ; θ ) ∇ θ T log p ( d ; θ ) ∣ θ ] = E [ ⟨ ∇ s ∣ d − s ⟩ ⟨ d − s ∣ ∇ s ⟩ ] = E [ ⟨ ∇ s ∣ n ⟩ ⟨ n ∣ ∇ s ⟩ ] = E [ ∫ d f ∇ s ~ ( f ; θ ) † n ~ ( f ) 1 / 2 S n ( f ) ∫ d f ′ n ~ ∗ ( f ′ ) ∇ s ~ ( f ′ ; θ ) 1 / 2 S n ( f ′ ) ] = ∫ d f d f ′ ∇ s ~ ( f ; θ ) † ∇ s ~ ( f ′ ; θ ) 1 / 2 S n ( f ) E [ n ~ ( f ) n ~ ∗ ( f ′ ) ] 1 / 2 S n ( f ′ ) = ∫ d f d f ′ ∇ s ~ ( f ; θ ) † ∇ s ~ ( f ′ ; θ ) 1 / 2 S n ( f ) δ ( f − f ′ ) = ∫ d f ∇ s ~ ( f ; θ ) † ∇ s ~ ( f ; θ ) 1 / 2 S n ( f ) = ⟨ ∇ s ∣ ∇ s ⟩ \begin{aligned}
\Gamma &= E\left[ \nabla_\theta \log p(d; \theta) ~ \nabla_\theta^T \log p(d; \theta) \Big| \theta\right]\\
&= E\Big[ \left\langle \nabla s \big|d - s\right\rangle ~ \left\langle d - s\big|\nabla s\right\rangle\Big] = E\Big[ \left\langle \nabla s \big|n\right\rangle ~ \left\langle n\big|\nabla s\right\rangle\Big]\\
&= E\left[\int df \frac{\nabla \tilde{s}(f;\theta)^\dagger \tilde{n}(f)}{1/2S_n(f)} \int df' \frac{\tilde{n}^*(f') \nabla \tilde{s}(f';\theta)}{1/2S_n(f')}\right]\\
&= \int df df' \frac{\nabla \tilde{s}(f;\theta)^\dagger \nabla \tilde{s}(f';\theta)}{1/2S_n(f)} \frac{E\big[\tilde{n}(f) \tilde{n}^*(f')\big]}{1/2S_n(f')}\\
&= \int df df' \frac{\nabla \tilde{s}(f;\theta)^\dagger \nabla \tilde{s}(f';\theta)}{1/2S_n(f)} \delta(f-f')\\
&= \int df \frac{\nabla \tilde{s}(f;\theta)^\dagger \nabla \tilde{s}(f;\theta)}{1/2S_n(f)} = \left\langle \nabla s \big| \nabla s\right\rangle
\end{aligned} Γ = E [ ∇ θ log p ( d ; θ ) ∇ θ T log p ( d ; θ ) ∣ ∣ ∣ ∣ θ ] = E [ ⟨ ∇ s ∣ ∣ ∣ d − s ⟩ ⟨ d − s ∣ ∣ ∣ ∇ s ⟩ ] = E [ ⟨ ∇ s ∣ ∣ ∣ n ⟩ ⟨ n ∣ ∣ ∣ ∇ s ⟩ ] = E [ ∫ d f 1 / 2 S n ( f ) ∇ s ~ ( f ; θ ) † n ~ ( f ) ∫ d f ′ 1 / 2 S n ( f ′ ) n ~ ∗ ( f ′ ) ∇ s ~ ( f ′ ; θ ) ] = ∫ d f d f ′ 1 / 2 S n ( f ) ∇ s ~ ( f ; θ ) † ∇ s ~ ( f ′ ; θ ) 1 / 2 S n ( f ′ ) E [ n ~ ( f ) n ~ ∗ ( f ′ ) ] = ∫ d f d f ′ 1 / 2 S n ( f ) ∇ s ~ ( f ; θ ) † ∇ s ~ ( f ′ ; θ ) δ ( f − f ′ ) = ∫ d f 1 / 2 S n ( f ) ∇ s ~ ( f ; θ ) † ∇ s ~ ( f ; θ ) = ⟨ ∇ s ∣ ∣ ∣ ∇ s ⟩
这里利用了噪声性质⟨ n ~ ( f ) n ~ ∗ ( f ′ ) ⟩ = 1 2 S n ( f ) δ ( f − f ′ ) \left\langle\tilde{n}(f)\tilde{n}^*(f')\right\rangle = \frac{1}{2}S_n(f)\delta(f-f') ⟨ n ~ ( f ) n ~ ∗ ( f ′ ) ⟩ = 2 1 S n ( f ) δ ( f − f ′ ) 。除此之外,还可从小量展开角度理解。假设参数先验为均匀分布,则后验分布p ( θ ∣ d ) ∝ p ( d ∣ θ ) p(\theta|d) \propto p(d|\theta) p ( θ ∣ d ) ∝ p ( d ∣ θ ) 。考虑参数最大似然估计θ ^ \hat{\theta} θ ^ 接近真实参数θ \theta θ ,在θ ^ \hat{\theta} θ ^ 附近对参数分布/统计模型展开到两阶:
p ( θ ∣ d ) ∝ ∼ exp { − 1 2 [ ⟨ d − s ( θ ^ ) ∣ d − s ( θ ^ ) ⟩ − 2 ⟨ d − s ( θ ^ ) ∣ ∇ s ( θ ) ⟩ ( θ − θ ^ ) − ( θ − θ ^ ) T ⟨ d − s ( θ ^ ) ∣ ∇ ∇ T s ( θ ) ⟩ ( θ − θ ^ ) + ( θ − θ ^ ) T ⟨ ∇ s ( θ ) ∣ ∇ s ( θ ) ⟩ ( θ − θ ^ ) ] } \begin{aligned}
p(\theta|d) \underset{\sim}{\propto} \exp\bigg\{ -\frac{1}{2} \Big[ &\left\langle d-s(\hat{\theta})\big|d-s(\hat{\theta})\right\rangle - 2\left\langle d-s(\hat{\theta})\big|\nabla s(\theta)\right\rangle (\theta-\hat{\theta}) \\
&- (\theta-\hat{\theta})^T \left\langle d-s(\hat{\theta})\big|\nabla \nabla^T s(\theta)\right\rangle (\theta-\hat{\theta}) \\
&+ (\theta-\hat{\theta})^T \left\langle \nabla s(\theta)\big|\nabla s(\theta)\right\rangle (\theta-\hat{\theta})\Big]\bigg\}
\end{aligned} p ( θ ∣ d ) ∼ ∝ exp { − 2 1 [ ⟨ d − s ( θ ^ ) ∣ ∣ ∣ d − s ( θ ^ ) ⟩ − 2 ⟨ d − s ( θ ^ ) ∣ ∣ ∣ ∇ s ( θ ) ⟩ ( θ − θ ^ ) − ( θ − θ ^ ) T ⟨ d − s ( θ ^ ) ∣ ∣ ∣ ∇ ∇ T s ( θ ) ⟩ ( θ − θ ^ ) + ( θ − θ ^ ) T ⟨ ∇ s ( θ ) ∣ ∣ ∣ ∇ s ( θ ) ⟩ ( θ − θ ^ ) ] }
忽略零阶的常数项;考虑到最大似然处导数(梯度)为0,即⟨ d − s ( θ ^ ) ∣ ∇ s ⟩ = 0 \langle d-s(\hat{\theta})|\nabla s\rangle = 0 ⟨ d − s ( θ ^ ) ∣ ∇ s ⟩ = 0 ,第二项为0;同时,在高信噪比 情况,∣ d − s ( θ ^ ∣ ≪ s |d-s(\hat{\theta}|\ll s ∣ d − s ( θ ^ ∣ ≪ s ,忽略第三项,最终:
p ( θ ∣ d ) ∝ ∼ exp [ − 1 2 ( θ − θ ^ ) T ⟨ ∇ s ∣ ∇ s ⟩ ( θ − θ ^ ) ] p(\theta|d) \underset{\sim}{\propto} \exp\left[-\frac{1}{2}(\theta-\hat{\theta})^T \left\langle \nabla s \big| \nabla s\right\rangle (\theta-\hat{\theta}) \right]
p ( θ ∣ d ) ∼ ∝ exp [ − 2 1 ( θ − θ ^ ) T ⟨ ∇ s ∣ ∣ ∣ ∇ s ⟩ ( θ − θ ^ ) ]
对比正态分布公式p ( θ ) ∝ exp [ − 1 2 ( θ − θ ^ ) T Σ − 1 ( θ − θ ^ ) ] p(\theta) \propto \exp\left[-\frac{1}{2} (\theta-\hat{\theta})^T \Sigma^{-1} (\theta-\hat{\theta})\right] p ( θ ) ∝ exp [ − 2 1 ( θ − θ ^ ) T Σ − 1 ( θ − θ ^ ) ] ,可得:
Σ = ⟨ ∇ s ∣ ∇ s ⟩ − 1 = Γ − 1 \Sigma = \left\langle \nabla s \big| \nabla s\right\rangle^{-1} = \Gamma^{-1}
Σ = ⟨ ∇ s ∣ ∣ ∣ ∇ s ⟩ − 1 = Γ − 1
注意,展开是在θ ^ \hat{\theta} θ ^ 附近进行的,因此计算梯度/Fisher信息矩阵时需要代入θ ^ \hat{\theta} θ ^ 。这里我们展开到两阶并忽略了高阶项(第三项),本质上是做了信号的线性近似 s ( θ ) ≈ s ( θ ^ ) + ∇ s ∣ θ ^ ( θ − θ ^ ) s(\theta) \approx s(\hat{\theta}) + \nabla s|_{\hat{\theta}} ~~ (\theta-\hat{\theta}) s ( θ ) ≈ s ( θ ^ ) + ∇ s ∣ θ ^ ( θ − θ ^ ) 。高信噪比的要求也是信号线性近似成立的条件。
在文献中,引力波观测响应信号通常用h h h 表示,因此更常见的为如下分量表达
Γ a b = ⟨ ∂ h ∂ θ a ∣ ∂ h ∂ θ b ⟩ = ⟨ ∂ θ a h ∣ ∂ θ b h ⟩ = ⟨ ∂ a h ∣ ∂ b h ⟩ \boxed{\Gamma_{ab}= \left\langle \frac{\partial h}{\partial \theta^a}\bigg|\frac{\partial h}{\partial \theta^b}\right\rangle = \left\langle \partial_{\theta_a} h\Big|\partial_{\theta_b} h\right\rangle = \left\langle \partial_a h\big|\partial_b h\right\rangle}
Γ a b = ⟨ ∂ θ a ∂ h ∣ ∣ ∣ ∣ ∣ ∂ θ b ∂ h ⟩ = ⟨ ∂ θ a h ∣ ∣ ∣ ∣ ∂ θ b h ⟩ = ⟨ ∂ a h ∣ ∣ ∣ ∂ b h ⟩
实际应用中,常见问题是Fisher信息矩阵存在奇异(行列式为0)或病态(行列式接近0),无法直接取逆或者逆矩阵在数值上不稳定。其中数值不稳定性可通过在信息矩阵中引入随机微扰,检查逆矩阵(协方差)的取值变化。存在问题时,通常需要先检查Fisher信息矩阵是否适用,参数间是否存在简并组合。应对的经验措施包括:移除简并参数、引入先验信息、修改量纲或取对数使不同参数在数量级上接近、提高算法的数值精度,以及引入对信息矩阵的高阶修正等。
关于先验,当信息矩阵对参数限制较弱,所给方差接近其先验∼ ( θ max − θ min ) 2 \sim (\theta_{\max}-\theta_{\min})^2 ∼ ( θ m a x − θ m i n ) 2 ,以及参数存在明显简并时,引入先验信息的效果会很显著。对于正态分布,先验的引入相对简单;对于常用的均匀分布,则可能需要借助蒙特卡罗模拟获得更准确的结果,具体可参考Vallisneri (2008) 。同时文中提供了似然函数的失配因子,用于作为信号线形近似(高信噪比)假设自洽性的判据,还进一步分析了更高阶展开。
最后,除了前面所关注的预测实验对于模型参数的限制能力(方差/置信区间),Fisher信息矩阵还常用于确定样本量(实验设计),以及基于预期置信区间构建假设检验。利用Fisher信息矩阵还可构建贝叶斯分析中的Jeffrey无信息先验,此外,某些模型选择方法同样依赖Fisher信息矩阵。
重叠约化函数 (Overlap reduction function)
以太阳质心系为参考,地基观测、空间观测探测器都在做轨道运动,区别只是地基观测探测器都位于地球上,距离小(相应的频段也高),两者重叠约化函数对比起来差别有多大??
PTA则是探测器的一端(地球)在做轨道运动,又是如何考虑轨道调制?
⟨ s ~ I ( f ) s ~ J ∗ ( f ′ ) ⟩ = δ ( f − f ′ ) S I J ( f ) ⟨ h ~ ( f ) h ~ ∗ ( f ′ ) ⟩ = δ ( f − f ′ ) S h ( f ) ⟨ h ~ P ( f , k ^ ) h ~ P ′ ∗ ( f ′ , k ^ ′ ) ⟩ ‾ = 1 4 π δ ( f − f ′ ) δ ( k ^ , k ^ ′ ) δ P P ′ S h ( f ) ∫ d 2 k ^ d 2 k ^ ′ ⟨ h ~ P ( f , k ^ ) h ~ P ′ ∗ ( f ′ , k ^ ′ ) ⟩ ‾ = 1 4 π δ ( f − f ′ ) δ ( k ^ , k ^ ′ ) δ P P ′ S h ( f ) S I J ( f ) = Γ I J ( f ) S h ( f ) = ⟨ R ⟩ γ I J ( f ) S h ( f ) \begin{aligned}
\langle \tilde{s}_I(f) \tilde{s}_J^*(f')\rangle &= \delta(f-f') S_{IJ}(f)\\
\langle \tilde{h}(f) \tilde{h}^*(f')\rangle &= \delta(f-f') S_{h}(f)\\
\overline{\langle \tilde{h}_P(f, \hat{k}) \tilde{h}_{P'}^*(f', \hat{k}')\rangle} &= \frac{1}{4\pi}\delta(f-f')\delta(\hat{k},\hat{k}')\delta_{PP'} S_{h}(f)\\
\int d^2\hat{k}d^2\hat{k}'\overline{\langle \tilde{h}_P(f, \hat{k}) \tilde{h}_{P'}^*(f', \hat{k}')\rangle} &= \frac{1}{4\pi}\delta(f-f')\delta(\hat{k},\hat{k}')\delta_{PP'} S_{h}(f)\\
S_{IJ}(f) = \Gamma_{IJ}(f)S_h(f) &= \langle \mathcal{R} \rangle \gamma_{IJ}(f)S_h(f)
\end{aligned} ⟨ s ~ I ( f ) s ~ J ∗ ( f ′ ) ⟩ ⟨ h ~ ( f ) h ~ ∗ ( f ′ ) ⟩ ⟨ h ~ P ( f , k ^ ) h ~ P ′ ∗ ( f ′ , k ^ ′ ) ⟩ ∫ d 2 k ^ d 2 k ^ ′ ⟨ h ~ P ( f , k ^ ) h ~ P ′ ∗ ( f ′ , k ^ ′ ) ⟩ S I J ( f ) = Γ I J ( f ) S h ( f ) = δ ( f − f ′ ) S I J ( f ) = δ ( f − f ′ ) S h ( f ) = 4 π 1 δ ( f − f ′ ) δ ( k ^ , k ^ ′ ) δ P P ′ S h ( f ) = 4 π 1 δ ( f − f ′ ) δ ( k ^ , k ^ ′ ) δ P P ′ S h ( f ) = ⟨ R ⟩ γ I J ( f ) S h ( f )
Γ I J ( f ) = S I J ( f ) S h ( f ) = s ~ I ( f ) s ~ J ∗ ( f ) ∑ P ∣ h ~ P ( f ) ∣ 2 \boxed{\Gamma_{IJ}(f) = \frac{S_{IJ}(f)}{S_h(f)} = \frac{\tilde{s}_I(f)\tilde{s}_J^*(f)}{\sum_P |\tilde{h}_P(f)|^2} }
Γ I J ( f ) = S h ( f ) S I J ( f ) = ∑ P ∣ h ~ P ( f ) ∣ 2 s ~ I ( f ) s ~ J ∗ ( f )
注意,互相关不满足交换律,在频域S I J ( f ) = S J I ∗ ( f ) S_{IJ}(f) = S^*_{JI}(f) S I J ( f ) = S J I ∗ ( f ) ,对应时域r I J ( t ) = r J I ∗ ( − t ) r_{IJ}(t) = r^*_{JI}(-t) r I J ( t ) = r J I ∗ ( − t ) ,对于实信号r I J ( t ) = r J I ( − t ) r_{IJ}(t) = r_{JI}(-t) r I J ( t ) = r J I ( − t ) 。对应于重叠约化函数则是Γ I J ( f ) = Γ J I ∗ ( f ) \Gamma_{IJ}(f) = \Gamma_{JI}^*(f) Γ I J ( f ) = Γ J I ∗ ( f ) 。
定性分析,对于相同位置、相同指向的探测器,探测到的信号完全重叠,归一化重叠函数为1,回归自相关 ;若探测器位置不同,探测到的信号间将存在相对延迟,若指向不同则各偏振的响应将有区别,最终信号重叠减弱。
D P ( f ) = D a b e a b P ; D a b = 1 2 [ x ^ i x ^ j T ( f , k ^ ⋅ x ^ ) − y ^ i y ^ j T ( f , k ^ ⋅ y ^ ) ] \boxed{D^P(f) = D^{ab} e_{ab}^P; ~~~ D^{ab} = \frac{1}{2}\left[\hat{x}^i \hat{x}^j ~\mathcal{T}(f, \hat{k} \cdot \hat{x}) - \hat{y}^i \hat{y}^j ~\mathcal{T}(f, \hat{k} \cdot \hat{y})\right]}
D P ( f ) = D a b e a b P ; D a b = 2 1 [ x ^ i x ^ j T ( f , k ^ ⋅ x ^ ) − y ^ i y ^ j T ( f , k ^ ⋅ y ^ ) ]
D S S B P ( f , t ) = D P ( f , t ) e − i 2 π f k ^ ⋅ r ⃗ d e t c D_{\rm SSB}^P(f, t) = D^P(f, t) e^{-i2\pi f\hat{k}\cdot\frac{\vec{r}_{\rm det}}{c}}
D S S B P ( f , t ) = D P ( f , t ) e − i 2 π f k ^ ⋅ c r d e t
D P = D a b e a b P ; D a b = 1 2 ( x ^ i x ^ j − y ^ i y ^ j ) D^P = D^{ab} e_{ab}^P; ~~~ D^{ab} = \frac{1}{2}\left(\hat{x}^i \hat{x}^j - \hat{y}^i \hat{y}^j\right)
D P = D a b e a b P ; D a b = 2 1 ( x ^ i x ^ j − y ^ i y ^ j )
D + ( t ) , D × ( t ) D^+(t), D^{\times}(t)
D + ( t ) , D × ( t )
⟨ D I + D J × ∗ ⟩ ↔ ⟨ D I + ∗ D J × ⟩ ? ? ? \left\langle D_I^{+} {D_J^{\times}}^* \right\rangle \leftrightarrow \left\langle {D_I^{+}}^* D_J^{\times} \right\rangle ??? ⟨ D I + D J × ∗ ⟩ ↔ ⟨ D I + ∗ D J × ⟩ ? ? ?
⟨ F I + F J + ∗ ⟩ = cos 2 2 ψ ⟨ D + D + ∗ ⟩ − sin 4 ψ ⟨ D + D × ∗ ⟩ + sin 2 2 ψ ⟨ D × D × ∗ ⟩ ⟨ F I × F J × ∗ ⟩ = sin 2 2 ψ ⟨ D + D + ∗ ⟩ + sin 4 ψ ⟨ D + D × ∗ ⟩ + cos 2 2 ψ ⟨ D × D × ∗ ⟩ ⟨ F + F × ∗ ⟩ = ⟨ F + ∗ F × ⟩ = 1 2 sin 4 ψ ( ⟨ D + D + ∗ ⟩ − ⟨ D × D × ∗ ⟩ ) + cos 4 ψ ⟨ D + D × ∗ ⟩ \begin{aligned}
&\small \left\langle F_I^{+} {F_J^{+}}^*\right\rangle = \cos^2 2\psi \left\langle D^{+}{D^{+}}^*\right\rangle - \sin 4\psi \left\langle D^{+} {D^{\times}}^*\right\rangle + \sin^2 2\psi \left\langle D^{\times} {D^{\times}}^*\right\rangle\\
&\small \left\langle F_I^{\times} {F_J^{\times}}^*\right\rangle = \sin^2 2\psi \left\langle D^{+}{D^{+}}^*\right\rangle + \sin 4\psi \left\langle D^{+} {D^{\times}}^*\right\rangle + \cos^2 2\psi \left\langle D^{\times} {D^{\times}}^*\right\rangle\\
&\small \left\langle F^{+} {F^{\times}}^*\right\rangle = \left\langle {F^{+}}^* F^{\times}\right\rangle = \frac{1}{2}\sin 4\psi \left( \left\langle D^{+}{D^{+}}^*\right\rangle - \left\langle D^{\times}{D^{\times}}^*\right\rangle \right) + \cos 4\psi \left\langle D^{+} {D^{\times}}^*\right\rangle
\end{aligned} ⟨ F I + F J + ∗ ⟩ = cos 2 2 ψ ⟨ D + D + ∗ ⟩ − sin 4 ψ ⟨ D + D × ∗ ⟩ + sin 2 2 ψ ⟨ D × D × ∗ ⟩ ⟨ F I × F J × ∗ ⟩ = sin 2 2 ψ ⟨ D + D + ∗ ⟩ + sin 4 ψ ⟨ D + D × ∗ ⟩ + cos 2 2 ψ ⟨ D × D × ∗ ⟩ ⟨ F + F × ∗ ⟩ = ⟨ F + ∗ F × ⟩ = 2 1 sin 4 ψ ( ⟨ D + D + ∗ ⟩ − ⟨ D × D × ∗ ⟩ ) + cos 4 ψ ⟨ D + D × ∗ ⟩
Γ I J ( f , k ^ ) = ∑ P ⟨ F S S B I P F S S B J P ∗ ⟩ ‾ = ∑ P ⟨ F I P F J P ∗ e − i 2 π f k ^ ⋅ r ⃗ I − r ⃗ J c ⟩ ‾ \boxed{ \Gamma_{IJ}(f, \hat{k}) = \sum_P \overline{\left\langle {F_{\rm \tiny SSB}}_I^P { {F_{\rm \tiny SSB}}_J^P }^*\right\rangle} = \sum_P \overline{\left\langle F_I^P {F_J^P}^* e^{- i 2\pi f\hat{k}\cdot\frac{\vec{r}_{I}-\vec{r}_{J}}{c}}\right\rangle}}
Γ I J ( f , k ^ ) = P ∑ ⟨ F S S B I P F S S B J P ∗ ⟩ = P ∑ ⟨ F I P F J P ∗ e − i 2 π f k ^ ⋅ c r I − r J ⟩
This is because the two independent Michelson interferometers that one can synthesize from the six links of the standard equilateral LISA configuration are rotated at 45° with respect to one another, leading to zero cross correlation for an isotropic gravitational-wave background for frequencies below about c/2L=3×10-2 Hz [17].
通过特解理解基本性质:
当f = 0 f = 0 f = 0 ,探测器平面不重叠时,重叠约化取不到1!!!只有探测器平面重合时低频时才趋向±1!!
当两个探测器指向呈45°,重叠约化会模值比较低吗??
当两个探测器指向呈90°,重叠约化在低频时趋向负值
探测器位置相同,但指向不一致
探测器位置相同,但平面有夹角
探测器保持平行,但有一定距离
γ I J ( f ) = 1 ⟨ R ⟩ ∫ d 2 k ^ 4 π ∫ d ψ 2 π D I a b e a b ( k ^ ) D J c d e c d ( k ^ ) e − i 2 π f k ^ ⋅ r ⃗ I − r ⃗ J c \gamma_{IJ}(f) = \frac{1}{\langle\mathcal{R}\rangle} \int \frac{d^2\hat{k}}{4\pi} \int \frac{d\psi}{2\pi} ~ D_I^{~~ab} e_{ab}(\hat{k}) D_J^{~~cd} e_{cd}(\hat{k}) e^{-i2\pi f \hat{k}\cdot \frac{\vec{r}_I-\vec{r}_J}{c}}
γ I J ( f ) = ⟨ R ⟩ 1 ∫ 4 π d 2 k ^ ∫ 2 π d ψ D I a b e a b ( k ^ ) D J c d e c d ( k ^ ) e − i 2 π f k ^ ⋅ c r I − r J
没有偏振的交叉项
γ I J ( f ) = 1 ⟨ R ⟩ ∑ P ∫ d 2 k ^ 4 π ∫ d ψ 2 π F I P ( k ^ ) F J P ( k ^ ) e − i 2 π f k ^ ⋅ r ⃗ I − r ⃗ J c \gamma_{IJ}(f) = \frac{1}{\langle\mathcal{R}\rangle} \sum_P \int \frac{d^2\hat{k}}{4\pi} \int \frac{d\psi}{2\pi} ~ F^P_I(\hat{k})F^P_J(\hat{k}) e^{-i2\pi f \hat{k}\cdot \frac{\vec{r}_I-\vec{r}_J}{c}}
γ I J ( f ) = ⟨ R ⟩ 1 P ∑ ∫ 4 π d 2 k ^ ∫ 2 π d ψ F I P ( k ^ ) F J P ( k ^ ) e − i 2 π f k ^ ⋅ c r I − r J
γ I J ( f ) = 1 ⟨ R ⟩ ∑ P ∫ d 2 k ^ 4 π ∫ d ψ 2 π D I a b e a b P ( k ^ ) D J c d e c d P ( k ^ ) e − i 2 π f k ^ ⋅ r ⃗ I − r ⃗ J c \gamma_{IJ}(f) = \frac{1}{\langle\mathcal{R}\rangle} \sum_P \int \frac{d^2\hat{k}}{4\pi} \int \frac{d\psi}{2\pi} ~ D_I^{~~ab} e^P_{ab}(\hat{k}) D_J^{~~cd} e^P_{cd}(\hat{k}) e^{-i2\pi f \hat{k}\cdot \frac{\vec{r}_I-\vec{r}_J}{c}}
γ I J ( f ) = ⟨ R ⟩ 1 P ∑ ∫ 4 π d 2 k ^ ∫ 2 π d ψ D I a b e a b P ( k ^ ) D J c d e c d P ( k ^ ) e − i 2 π f k ^ ⋅ c r I − r J
如果不是轨道平均的,为什么不含时?如果是轨道平均过的,探测器方位r ⃗ \vec{r} r 该怎么定义
?
D a b D^{ab} D a b 依赖于k ^ ⋅ x ^ \hat{k}\cdot\hat{x} k ^ ⋅ x ^ ,这里将D a b D^{ab} D a b 从积分中提出来,暗含了低频近似。
γ I J ( f ) = D I a b D J c d ⟨ R ⟩ ∑ P ∫ d 2 k ^ 4 π ∫ d ψ 2 π e a b P ( k ^ ) e c d P ( k ^ ) e − i 2 π f k ^ ⋅ r ⃗ I − r ⃗ J c \gamma_{IJ}(f) = \frac{D_I^{~~ab} D_J^{~~cd} }{\langle\mathcal{R}\rangle} \sum_P \int \frac{d^2\hat{k}}{4\pi} \int \frac{d\psi}{2\pi} ~ e^P_{ab}(\hat{k}) e^P_{cd}(\hat{k}) e^{-i2\pi f \hat{k}\cdot \frac{\vec{r}_I-\vec{r}_J}{c}}
γ I J ( f ) = ⟨ R ⟩ D I a b D J c d P ∑ ∫ 4 π d 2 k ^ ∫ 2 π d ψ e a b P ( k ^ ) e c d P ( k ^ ) e − i 2 π f k ^ ⋅ c r I − r J
利用TT规范下e a b e_{ab} e a b 的对称无迹性质可获得该积分的解析解,具体推导可参考 Flanagan(1993) 以及 Allen & Romano (1999),这里仅做简单介绍。将探测器连线 r ⃗ I J = r ⃗ J − r ⃗ I \vec{r}_{IJ} = \vec{r}_{J} - \vec{r}_{I} r I J = r J − r I 的方向矢量记为z ^ \hat{z} z ^ ,引力波在探测器间传播的相位延迟记为ı ≡ 2 π f r I J / c \imath \equiv 2\pi f r_{\tiny IJ}/c ≡ 2 π f r I J / c ,引入张量
γ a b c d ( ı , z ^ ) ≡ 1 ⟨ R ⟩ ∑ P ∫ d 2 k ^ 4 π e a b P ( k ^ ) e c d P ( k ^ ) e i k ^ ⋅ z ^ ı ; γ I J ( f ) = D I a b D J c d γ a b c d \gamma_{abcd}(\imath, \hat{z}) \equiv \frac{1}{\langle\mathcal{R}\rangle} \sum_P \int \frac{d^2\hat{k}}{4\pi} e^P_{ab}(\hat{k})e^P_{cd}(\hat{k}) e^{i \hat{k}\cdot \hat{z}~\imath}; ~~~ \gamma_{IJ}(f) = D_I^{~~ab} D_J^{~~cd}\gamma_{abcd}
γ a b c d ( , z ^ ) ≡ ⟨ R ⟩ 1 P ∑ ∫ 4 π d 2 k ^ e a b P ( k ^ ) e c d P ( k ^ ) e i k ^ ⋅ z ^ ; γ I J ( f ) = D I a b D J c d γ a b c d
为求解γ a b c d \gamma_{abcd} γ a b c d ,可先将其表示为δ a b \delta_{ab} δ a b 和z a z_a z a 的一般性组合,考虑到γ a b c d \gamma_{abcd} γ a b c d 对于a ↔ b a \leftrightarrow b a ↔ b 、c ↔ d c \leftrightarrow d c ↔ d 、a b ↔ c d ab \leftrightarrow cd a b ↔ c d 的指标交换对称,且对于a , b a, b a , b 、c , d c, d c , d 指标对无迹,最终:
γ a b c d ( ı , z ^ ) = A ( ı ) δ a b δ c d + B ( ı ) ( δ a c δ b d + δ b c δ a d ) + C ( ı ) ( δ a b z c z d + δ c d z a z b ) + E ( ı ) ( δ a c z b z d + δ a d z b z c + δ b c z a z d + δ b d z a z c ) + F ( ı ) z a z b z c z d \begin{aligned}
\gamma_{abcd}(\imath, \hat{z}) =& ~ A(\imath)\delta_{ab}\delta_{cd} + B(\imath)(\delta_{ac}\delta_{bd} + \delta_{bc}\delta_{ad}) + C(\imath)(\delta_{ab}z_c z_d + \delta_{cd}z_a z_b)\\
& + E(\imath)(\delta_{ac}z_b z_d + \delta_{ad}z_b z_c + \delta_{bc}z_a z_d + \delta_{bd}z_a z_c) + F(\imath) z_a z_b z_c z_d
\end{aligned} γ a b c d ( , z ^ ) = A ( ) δ a b δ c d + B ( ) ( δ a c δ b d + δ b c δ a d ) + C ( ) ( δ a b z c z d + δ c d z a z b ) + E ( ) ( δ a c z b z d + δ a d z b z c + δ b c z a z d + δ b d z a z c ) + F ( ) z a z b z c z d
即考虑到对称性,最一般形式下,γ a b c d \gamma_{abcd} γ a b c d 可表示为五个不同张量的组合。将五个张量分别与上式缩并可以得到对应系数的不同组合,而直接与γ a b c d \gamma_{abcd} γ a b c d 定义式缩可得到对应的组合结果,缩并后积分都相对简单,可直接求解:
( 9 6 6 4 1 6 24 4 16 2 6 4 8 8 2 4 16 8 24 4 1 2 2 4 1 ) ( A B C E F ) = ( 0 20 j 0 0 40 ı − 1 j 1 20 ı − 2 j 2 ) \begin{aligned}
\begin{pmatrix}
9 & 6 & 6 & 4 & 1\\
6 & 24 & 4 & 16 & 2\\
6 & 4 & 8 & 8 & 2\\
4 & 16 & 8 & 24 & 4\\
1 & 2 & 2 & 4 & 1
\end{pmatrix}
\begin{pmatrix}
A\\
B\\
C\\
E\\
F
\end{pmatrix} =
\begin{pmatrix}
0\\
20~j_0\\
0\\
40~\imath^{\tiny -1} j_1\\
20~\imath^{\tiny -2} j_2
\end{pmatrix}
\end{aligned} ⎝ ⎜ ⎜ ⎜ ⎜ ⎜ ⎛ 9 6 6 4 1 6 2 4 4 1 6 2 6 4 8 8 2 4 1 6 8 2 4 4 1 2 2 4 1 ⎠ ⎟ ⎟ ⎟ ⎟ ⎟ ⎞ ⎝ ⎜ ⎜ ⎜ ⎜ ⎜ ⎛ A B C E F ⎠ ⎟ ⎟ ⎟ ⎟ ⎟ ⎞ = ⎝ ⎜ ⎜ ⎜ ⎜ ⎜ ⎛ 0 2 0 j 0 0 4 0 − 1 j 1 2 0 − 2 j 2 ⎠ ⎟ ⎟ ⎟ ⎟ ⎟ ⎞
同时,利用探测器张量“对称无迹”的特性,对于γ I J \gamma_{IJ} γ I J 有:
γ I J = D I a b D J c d γ a b c d = D I a b D J c d ( 2 B δ a c δ b d + 4 E δ a c z b z d + F z a z b z c z d ) = 2 B D I a b D J a b + 4 E D I a b D J a c z b z c + F D I a b z a z b D J c d z c z d \begin{aligned}
\gamma_{IJ} &= D_I^{~~ab} D_J^{~~cd}\gamma_{abcd}\\
& = D_I^{~~ab} {D_J}^{cd} (2B ~\delta_{ac}\delta_{bd} + 4E ~ \delta_{ac}z_b z_d + F z_a z_bz_c z_d)\\
& \sout{= 2B ~ D_I^{~~ab} {D_J}_{ab} + 4E ~ D_I^{~~ab} {D_J}_a^{~~c} z_b z_c + F D_I^{~~ab}z_a z_b ~ {D_J}^{cd} z_c z_d}
\end{aligned} γ I J = D I a b D J c d γ a b c d = D I a b D J c d ( 2 B δ a c δ b d + 4 E δ a c z b z d + F z a z b z c z d ) = 2 B D I a b D J a b + 4 E D I a b D J a c z b z c + F D I a b z a z b D J c d z c z d
求解方程组,得到对应系数B , E , F B, E, F B , E , F ,并带入上式,最终:
γ a b c d ′ ≡ 2 B δ a c δ b d + 4 E δ a c z b z d + F z a z b z c z d = ȷ 0 δ a c δ b d + ȷ 1 δ a c z b z d + ȷ 2 z a z b z c z d ( ȷ 0 ȷ 1 ȷ 2 ) = ( 2 B 4 E F ) = 5 ( 1 − 2 1 − 2 8 − 10 1 / 2 − 5 35 / 2 ) ( j 0 ı − 1 j 1 ı − 2 j 2 ) = 1 7 ( 14 − 20 1 0 60 − 10 0 0 35 / 2 ) ( j 0 j 2 j 4 ) \boxed{\begin{aligned}
\gamma'_{abcd} &\equiv 2B ~\delta_{ac}\delta_{bd} + 4E ~ \delta_{ac}z_b z_d + F z_a z_bz_c z_d = \jmath_0 ~\delta_{ac}\delta_{bd} + \jmath_1 ~ \delta_{ac}z_b z_d + \jmath_2 z_a z_bz_c z_d\\
\begin{pmatrix}
\jmath_0 \\
\jmath_1 \\
\jmath_2
\end{pmatrix} &=
\begin{pmatrix}
2B \\
4E \\
F
\end{pmatrix} = 5
\begin{pmatrix}
~1 & -2 & ~1\\
-2 & ~8 & -10\\
~1/2 & -5 & ~35/2
\end{pmatrix}
\begin{pmatrix}
j_0\\
\imath^{\tiny -1}j_1\\
\imath^{\tiny -2}j_2
\end{pmatrix} = \frac{1}{7} \begin{pmatrix}
~14 & -20 & ~1\\
0 & ~60 & -10\\
0 & 0 & ~35/2
\end{pmatrix}
\begin{pmatrix}
j_0\\
j_2\\
j_4
\end{pmatrix}
\end{aligned}} γ a b c d ′ ⎝ ⎛ 0 1 2 ⎠ ⎞ ≡ 2 B δ a c δ b d + 4 E δ a c z b z d + F z a z b z c z d = 0 δ a c δ b d + 1 δ a c z b z d + 2 z a z b z c z d = ⎝ ⎛ 2 B 4 E F ⎠ ⎞ = 5 ⎝ ⎛ 1 − 2 1 / 2 − 2 8 − 5 1 − 1 0 3 5 / 2 ⎠ ⎞ ⎝ ⎛ j 0 − 1 j 1 − 2 j 2 ⎠ ⎞ = 7 1 ⎝ ⎛ 1 4 0 0 − 2 0 6 0 0 1 − 1 0 3 5 / 2 ⎠ ⎞ ⎝ ⎛ j 0 j 2 j 4 ⎠ ⎞
其中j n j_n j n 为球贝塞尔函数,自变量为ı = 2 π f r I J / c \imath = 2\pi f r_{\tiny IJ}/c = 2 π f r I J / c ,相应的ȷ n \jmath_n n 也是ı \imath 的函数。
对于和球面相切的迈克尔逊干涉仪(地基探测器),利用对称性,张量缩并可得到简化。具体结果如下(Flanagan 1993),其中β \beta β 为两台干涉仪相对球心的张角,两台探测器可通过大圆连接,σ 1 , σ 2 \sigma_1, \sigma_2 σ 1 , σ 2 是探测器顶角平分线在各自平面内相对该大圆切线的转角(逆时针)。
γ I J ( f ) = Θ 1 ( ı , β ) cos [ 2 ( σ 1 − σ 2 ) ] + Θ 2 ( ı , β ) cos [ 2 ( σ 1 + σ 2 ) ] Θ 1 ( ı , β ) = cos 4 β 2 g 0 ( ı ) Θ 2 ( ı , β ) = − sin 4 β 2 g 0 ( ı ) + cos β g 1 ( ı ) + g 2 ( ı ) ( g 0 g 1 g 2 ) = 5 16 ( − 9 − 6 9 3 1 45 6 − 45 9 3 60 − 16 − 60 36 − 4 ) ( cos ı ı 4 cos ı ı 2 sin ı ı 5 sin ı ı 3 sin ı ı ) = 5 16 ( 1 6 3 3 − 6 − 15 − 4 16 − 20 ) ( j 0 ı − 1 j 1 ı − 2 j 2 ) = 1 112 ( 112 80 3 0 − 120 − 15 0 120 − 20 ) ( j 0 j 2 j 4 ) \begin{aligned}
\gamma_{IJ}(f) &= \Theta_1(\imath, \beta)\cos[2(\sigma_1-\sigma_2)] + \Theta_2(\imath, \beta)\cos[2(\sigma_1+\sigma_2)]\\
\Theta_1(\imath, \beta) &= \cos^4\frac{\beta}{2} g_0(\imath)\\
\Theta_2(\imath, \beta) &= -\sin^4\frac{\beta}{2} g_0(\imath) + \cos\beta g_1(\imath) + g_2(\imath)\\
\begin{pmatrix}
g_0 \\
g_1 \\
g_2
\end{pmatrix}&= \frac{5}{16}
\begin{pmatrix}
-9 & -6 & ~~9 & 3 & 1\\
45 & ~6 & -45 & 9 & 3\\
~60 & -16 & -60 & 36 & -4\\
\end{pmatrix}
\begin{pmatrix}
\frac{\cos\imath}{\imath^4} \\
\frac{\cos\imath}{\imath^2} \\
\frac{\sin\imath}{\imath^5}\\
\frac{\sin\imath}{\imath^3}\\
\frac{\sin\imath}{\imath}
\end{pmatrix}\\
&= \frac{5}{16}
\begin{pmatrix}
~1 & 6 & 3\\
~~3 & -6 & -15\\
-4 & 16 & -20
\end{pmatrix}
\begin{pmatrix}
j_0\\
\imath^{\tiny -1}j_1\\
\imath^{\tiny -2}j_2\\
\end{pmatrix}\\ \\
& = \frac{1}{112}
\begin{pmatrix}
112 & 80 & ~3\\
~0 & -120 & -15\\
~0 & ~120 & -20
\end{pmatrix}
\begin{pmatrix}
j_0\\
j_2\\
j_4\\
\end{pmatrix}
\end{aligned} γ I J ( f ) Θ 1 ( , β ) Θ 2 ( , β ) ⎝ ⎛ g 0 g 1 g 2 ⎠ ⎞ = Θ 1 ( , β ) cos [ 2 ( σ 1 − σ 2 ) ] + Θ 2 ( , β ) cos [ 2 ( σ 1 + σ 2 ) ] = cos 4 2 β g 0 ( ) = − sin 4 2 β g 0 ( ) + cos β g 1 ( ) + g 2 ( ) = 1 6 5 ⎝ ⎛ − 9 4 5 6 0 − 6 6 − 1 6 9 − 4 5 − 6 0 3 9 3 6 1 3 − 4 ⎠ ⎞ ⎝ ⎜ ⎜ ⎜ ⎜ ⎜ ⎛ 4 c o s 2 c o s 5 s i n 3 s i n s i n ⎠ ⎟ ⎟ ⎟ ⎟ ⎟ ⎞ = 1 6 5 ⎝ ⎛ 1 3 − 4 6 − 6 1 6 3 − 1 5 − 2 0 ⎠ ⎞ ⎝ ⎛ j 0 − 1 j 1 − 2 j 2 ⎠ ⎞ = 1 1 2 1 ⎝ ⎛ 1 1 2 0 0 8 0 − 1 2 0 1 2 0 3 − 1 5 − 2 0 ⎠ ⎞ ⎝ ⎛ j 0 j 2 j 4 ⎠ ⎞
Allen & Romano 1999中指出,Flanagan 1993的一般性结果存在笔误,该错误是否被延续到了该结果仍有待验证
所有结论都是在低频近似下,f < c / 2 L ∼ 60 m H z f < c/2L \sim 60 \rm mHz f < c / 2 L ∼ 6 0 m H z
当频率趋于零f → 0 f \rightarrow 0 f → 0 ,g 0 , g 1 , g 2 g_0, g_1, g_2 g 0 , g 1 , g 2 在零点分别取1 , 0 , 0 1, 0, 0 1 , 0 , 0 ,此时γ I J = cos [ 2 ( σ 1 − σ 2 ) ] cos 4 β 2 − cos [ 2 ( σ 1 + σ 2 ) ] sin 4 β 2 \gamma_{IJ} = \cos[2(\sigma_1-\sigma_2)]\cos^4\frac{\beta}{2} - \cos[2(\sigma_1 + \sigma_2)]\sin^4\frac{\beta}{2} γ I J = cos [ 2 ( σ 1 − σ 2 ) ] cos 4 2 β − cos [ 2 ( σ 1 + σ 2 ) ] sin 4 2 β 。
探测器位置相同,但指向不一致
当探测器位于同一位置r I J → 0 r_{IJ} \rightarrow 0 r I J → 0 ,β = 0 \beta = 0 β = 0 ,g 0 , g 1 , g 2 g_0, g_1, g_2 g 0 , g 1 , g 2 在零点分别取1 , 0 , 0 1, 0, 0 1 , 0 , 0 ,此时γ I J = cos [ 2 ( σ 1 − σ 2 ) ] \gamma_{IJ} = \cos[2(\sigma_1-\sigma_2)] γ I J = cos [ 2 ( σ 1 − σ 2 ) ] 。
LISA或太极的A、E通道呈45°,即σ 1 − σ 2 = π 4 \sigma_1-\sigma_2 = \frac{\pi}{4} σ 1 − σ 2 = 4 π ,因此对于各向同性的随机背景,当个探测器A、E通道间重叠为0!这是复合预期的,A、E通道的构建本身就要求噪声独立,而随机背景本身与噪声类似。
探测器距离r ⃗ I J \vec{r}_{IJ} r I J
探测器连线与探测器I、II所张平面的夹角β 1 , β 2 \beta_1, \beta_2 β 1 , β 2
探测器连线在探测器I、II平面投影与探测器顶角平分线夹角σ 1 , σ 2 \sigma_1, \sigma_2 σ 1 , σ 2
探测器连线与探测器I、II平面交线的夹角χ \chi χ
信号探测能力
ρ I J = ⟨ R ⟩ T o b s ∫ d f γ I J 2 ( f ) S h 2 ( f ) N I ( f ) N J ( f ) \boxed{\rho_{\tiny IJ} = \langle \mathcal{R} \rangle \sqrt{T_{\rm obs}} \sqrt{\int df \frac{ \gamma_{IJ}^2(f) S^2_h(f)}{N_I(f) N_J(f)} } }
ρ I J = ⟨ R ⟩ T o b s ∫ d f N I ( f ) N J ( f ) γ I J 2 ( f ) S h 2 ( f )
这里可引入有效功率谱:
S e f f ( f ) ≡ N I ( f ) N J ( f ) ⟨ R ⟩ γ I J ( f ) , Ω e f f ( f ) = 4 π 2 3 H 0 2 f 3 S e f f ( f ) S_{\rm eff}(f) \equiv \frac{\sqrt{N_I(f)N_J(f)}}{\langle\mathcal{R}\rangle \gamma_{IJ}(f)}, ~~~ \Omega_{\rm eff}(f) = \frac{4\pi^2}{3H_0^2} f^3 S_{\rm eff}(f)
S e f f ( f ) ≡ ⟨ R ⟩ γ I J ( f ) N I ( f ) N J ( f ) , Ω e f f ( f ) = 3 H 0 2 4 π 2 f 3 S e f f ( f )
同时记频域平均⟨ . . . ⟩ f ≡ ∫ f min f max . . . d f 2 ∫ f min f max d f = ∫ f min f max . . . d f 2 f b a n d \left\langle ... \right\rangle_f \equiv \frac{\int_{f_{\min}}^{f_{\max}} ... df}{ 2\int_{f_{\min}}^{f_{\max}} df} = \frac{\int_{f_{\min}}^{f_{\max}} ... df}{2f_{\rm band}} ⟨ . . . ⟩ f ≡ 2 ∫ f m i n f m a x d f ∫ f m i n f m a x . . . d f = 2 f b a n d ∫ f m i n f m a x . . . d f ,噪比就可表示为:
ρ = T 2 f b a n d ⟨ S h 2 ( f ) S e f f 2 ( f ) ⟩ f = N b i n s ⟨ S h 2 ( f ) S e f f 2 ( f ) ⟩ f \rho = \sqrt{T} \sqrt{2f_{\rm band}} \sqrt{\left\langle\frac{S^2_h(f)}{S^2_{\rm eff}(f)}\right\rangle_f } = \sqrt{N_{\rm bins}} \sqrt{\left\langle\frac{S^2_h(f)}{S^2_{\rm eff}(f)}\right\rangle_f }
ρ = T 2 f b a n d ⟨ S e f f 2 ( f ) S h 2 ( f ) ⟩ f = N b i n s ⟨ S e f f 2 ( f ) S h 2 ( f ) ⟩ f
假设探测器完全重合,则S e f f ( f ) = N ( f ) ⟨ R ⟩ = S n ( f ) S_{\rm eff}(f) = \frac{N(f)}{\langle\mathcal{R}\rangle}= S_n(f) S e f f ( f ) = ⟨ R ⟩ N ( f ) = S n ( f ) ,从而
ρ = T 2 f b a n d ρ s i n g l e = 8 × 1 0 3 ( T 1 y r ) 1 / 2 ( f b a n d 1 H z ) 1 / 2 ρ s i n g l e \rho = \sqrt{T} \sqrt{2f_{\rm band}} ~ \rho_{\rm single} = 8 \times 10^3 \left(\frac{T}{1\rm yr}\right)^{1/2} \left(\frac{f_{\rm band}}{1\rm Hz}\right)^{1/2} \rho_{\rm single}
ρ = T 2 f b a n d ρ s i n g l e = 8 × 1 0 3 ( 1 y r T ) 1 / 2 ( 1 H z f b a n d ) 1 / 2 ρ s i n g l e
空间探测器频带宽度大概就是1 H z 1\rm Hz 1 H z ,一年观测信噪比提升约8 × 1 0 3 8 \times 10^3 8 × 1 0 3 倍。地基探测器频带约150 H z 150\rm Hz 1 5 0 H z ,一年观测信噪比提升约1 × 1 0 5 1 \times 10^5 1 × 1 0 5 倍
跟Maggiore结果差了2 \sqrt{2} 2 倍 ,提升是1 × 1 0 5 1 \times 10^5 1 × 1 0 5 倍
考虑正负频率,频率带宽为只考虑正频率的2倍!!
S e f f ( f ) S_{\rm eff}(f) S e f f ( f ) 对应于单台探测器的S n ( f ) S_n(f) S n ( f ) ,可作为与单台
但真正的提升在于前面的时间和频带宽度项。
假设背景信号的功率密度谱服从幂率分布Ω G W = Ω α ( f f c ) α \Omega_{\rm \tiny GW} = \Omega_\alpha \left(\frac{f}{f_c}\right)^\alpha Ω G W = Ω α ( f c f ) α ,f c f_c f c 为特征频率。给定信噪比阈值,根据上面的最优信噪比,可以给出对Ω α \Omega_\alpha Ω α 的下限:
Ω α t h r e s h o l d = ρ t h r e s h o l d ⟨ R ⟩ T o b s f c α 4 π 2 3 H 0 2 [ ∫ d f γ I J 2 ( f ) f 2 α − 6 N I ( f ) N J ( f ) ] − 1 / 2 \Omega_{\alpha_{\rm threshold}} = \frac{\rho_{\rm threshold}}{\langle R\rangle \sqrt{T_{\rm obs}} } f_c^{\alpha} ~ \frac{4\pi^2}{3H_0^2} \left[\int df \frac{\gamma_{IJ}^2(f)f^{2\alpha-6} }{N_I(f)N_J(f)}\right]^{-1/2}
Ω α t h r e s h o l d = ⟨ R ⟩ T o b s ρ t h r e s h o l d f c α 3 H 0 2 4 π 2 [ ∫ d f N I ( f ) N J ( f ) γ I J 2 ( f ) f 2 α − 6 ] − 1 / 2
Ω α t h r e s h o l d = ρ t h r e s h o l d ⟨ R ⟩ T o b s f c α [ ∫ d f ( γ I J ( f ) f α Ω n I J ( f ) ) 2 ] − 1 / 2 \begin{aligned}\Omega_{\alpha_{\rm threshold}} = \frac{\rho_{\rm threshold}}{\langle R\rangle \sqrt{T_{\rm obs}} } f_c^{\alpha} ~ \left[\int df \left( \frac{ \gamma_{IJ}(f) f^{\alpha} }{ \Omega_{n_{IJ}}(f) } \right)^2 \right]^{-1/2}
\end{aligned} Ω α t h r e s h o l d = ⟨ R ⟩ T o b s ρ t h r e s h o l d f c α [ ∫ d f ( Ω n I J ( f ) γ I J ( f ) f α ) 2 ] − 1 / 2
对任意谱指数,通过上述公式都可得到Ω α \Omega_\alpha Ω α 的下限,进而绘制出对应的Ω G W \Omega_{\rm \tiny GW} Ω G W 的谱线,遍历所有谱指数α \alpha α 可以得到一系列幂率直线。这些幂率谱的上包络就可以作为探测器,在给定信噪比下,对任意 幂率噪声探测极限的综合指标,被称为幂率积分 (power-law integrated, PLI) 灵敏度曲线(Thrane & Romano 2013)。
Ω P L I ( f ) ≡ max α [ Ω α t h r e s h o l d ( f f c ) α ] = ρ t h r e s h o l d ⟨ R ⟩ T o b s 4 π 2 3 H 0 2 max α { [ ∫ d f γ I J 2 ( f ) f 2 α − 6 N I ( f ) N J ( f ) ] − 1 / 2 f α } \begin{aligned}
\Omega_{\rm \tiny PLI}(f) &\equiv \max_\alpha\left[ \Omega_{\alpha_{\rm threshold}} \left(\frac{f}{f_c}\right)^\alpha \right]\\
&= \frac{\rho_{\rm threshold}}{\langle R\rangle\sqrt{T_{\rm obs}} } ~ \frac{4\pi^2}{3H_0^2} ~ \max_\alpha \left\{ \left[\int df \frac{\gamma_{IJ}^2(f)f^{2\alpha-6} }{N_I(f)N_J(f)}\right]^{-1/2} f^\alpha \right\}
\end{aligned} Ω P L I ( f ) ≡ α max [ Ω α t h r e s h o l d ( f c f ) α ] = ⟨ R ⟩ T o b s ρ t h r e s h o l d 3 H 0 2 4 π 2 α max { [ ∫ d f N I ( f ) N J ( f ) γ I J 2 ( f ) f 2 α − 6 ] − 1 / 2 f α }
Ω P L I ( f ) ≡ max α [ Ω α t h r e s h o l d ( f f c ) α ] = ρ t h r e s h o l d ⟨ R ⟩ T o b s max α { [ ∫ d f ( γ I J ( f ) f α Ω n I J ( f ) ) 2 ] − 1 / 2 f α } \begin{aligned}
\Omega_{\rm \tiny PLI}(f) &\equiv \max_\alpha\left[ \Omega_{\alpha_{\rm threshold}} \left(\frac{f}{f_c}\right)^\alpha \right]\\
&= \frac{\rho_{\rm threshold}}{\langle R\rangle\sqrt{T_{\rm obs}} } \max_\alpha \left\{ \left[\int df \left( \frac{ \gamma_{IJ}(f) f^{\alpha} }{ \Omega_{n_{IJ}}(f) } \right)^2 \right]^{-1/2} f^\alpha \right\}
\end{aligned} Ω P L I ( f ) ≡ α max [ Ω α t h r e s h o l d ( f c f ) α ] = ⟨ R ⟩ T o b s ρ t h r e s h o l d α max ⎩ ⎪ ⎨ ⎪ ⎧ [ ∫ d f ( Ω n I J ( f ) γ I J ( f ) f α ) 2 ] − 1 / 2 f α ⎭ ⎪ ⎬ ⎪ ⎫
原则上,对S h P I L ( f ) S_{h_{\rm PIL}}(f) S h P I L ( f ) 也可以定义类似的幂率积分灵敏度曲线,这条曲线与上面的Ω P I L \Omega_{\rm \tiny PIL} Ω P I L 曲线直接转换得到的结果一致吗?相差多少?
显然,幂率积分灵敏度只适用于背景能量密度谱服从幂率分布的情况,对于单峰分布等其他常见能谱并不具有意义。于是最近有人提出了另一种指标——峰值积分 (peak-integrated, PI)灵敏度曲线(Schmitz 2020)。
其基本想法也很简单:幂率谱的优势是谱的形状和强度是天然分离的,而一般性能谱,谱形也会受模型参数影响,与强度是耦合的。在对信噪比贡献最多的峰值附近,谱形近似不依赖模型参数,通常可近似为分段幂率:
Ω G W ( f ) = Ω p e a k ( p ) S ( f ; f p e a k ) \Omega_{\rm \tiny GW}(f) = \Omega_{\rm peak}(\bm{p}) \mathcal{S}(f; f_{\rm peak})
Ω G W ( f ) = Ω p e a k ( p ) S ( f ; f p e a k )
这里Ω p e a k ( p ) \Omega_{\rm peak}(\bm{p}) Ω p e a k ( p ) 是峰值谱值,取决于随机背景的物理模型参数p \bm{p} p ,S \mathcal{S} S 是峰值附近的谱形函数,与模型参数无关(仅由基本的产生机制决定),一般可用分段幂率函数近似。
Ω p e a k t h r e s h o l d = ρ t h r e s h o l d ⟨ R ⟩ T o b s [ ∫ d f ( γ I J ( f ) S ( f ; f p e a k ) Ω n I J ( f ) ) 2 ] − 1 / 2 \Omega_{\rm peak_{\rm threshold}} = \frac{\rho_{\rm threshold}}{\langle R\rangle \sqrt{T_{\rm obs}} } ~ \left[\int df \left( \frac{ \gamma_{IJ}(f) \mathcal{S}(f; f_{\rm peak}) }{ \Omega_{n_{IJ}}(f) } \right)^2 \right]^{-1/2}
Ω p e a k t h r e s h o l d = ⟨ R ⟩ T o b s ρ t h r e s h o l d [ ∫ d f ( Ω n I J ( f ) γ I J ( f ) S ( f ; f p e a k ) ) 2 ] − 1 / 2
最后,对所有可能的峰值频率取值计算上述阈值,就可以得到其作为频率的函数:
Ω P I ( f ) ≡ Ω p e a k t h r e s h o l d ( f p e a k ) \Omega_{\rm PI}(f) \equiv \Omega_{\rm peak_{\rm threshold}}(f_{\rm peak})
Ω P I ( f ) ≡ Ω p e a k t h r e s h o l d ( f p e a k )
注意,虽然S \mathcal{S} S 不依赖模型参数,但显然是依赖具体模型选择的,因此峰值积分灵敏度针对谱形不同的宇宙学背景理论需要分别计算对应的灵敏度曲线。作为对比,幂率积分灵敏度同样也只是适用于预言幂率谱的理论。
对于多种背景混叠情况可以计算总体灵敏度吗,纵轴取什么??
2个探测器N段数据
ρ I J 2 ( i ) = T ( i ) ∫ d f Γ I J 2 ( f ) S h 2 ( f ) N I ( i ) N J ( i ) ( f ) {\rho_{\tiny IJ}^2}^{(i)} = T^{(i)} \int df ~ \frac{\Gamma_{IJ}^2(f)S^2_h(f)}{N_I^{(i)}N_J^{(i)}(f)}
ρ I J 2 ( i ) = T ( i ) ∫ d f N I ( i ) N J ( i ) ( f ) Γ I J 2 ( f ) S h 2 ( f )
ρ 2 = ∑ i ρ I J ( i ) 2 = ∫ d f Γ I J 2 ( f ) S h 2 ( f ) ∑ i T ( i ) N I ( i ) ( f ) N J ( i ) ( f ) ∼ N ρ 2 ( i ) \rho^2 = \sum_i {\rho_{\tiny IJ}^{(i)}}^2 = \int df ~ \Gamma_{IJ}^2(f)S^2_h(f) \sum_{i} \frac{T^{(i)}}{N_I^{(i)}(f)N_J^{(i)}(f)} \sim N {\rho^2}^{(i)}
ρ 2 = i ∑ ρ I J ( i ) 2 = ∫ d f Γ I J 2 ( f ) S h 2 ( f ) i ∑ N I ( i ) ( f ) N J ( i ) ( f ) T ( i ) ∼ N ρ 2 ( i )
N个探测器联合观测
对于多个探测器可以两两互相关,最优的整体信噪比为全部组合信噪比平方和开根号:
ρ 2 = ∑ I ∑ J < I ρ I J 2 , ρ = ∑ I ∑ J < I ρ I J 2 ∼ N ( N − 1 ) 2 ρ I J ∝ N ρ I J \boxed{\rho^2 = \sum_I \sum_{J<I} \rho_{\tiny IJ}^2, ~~~ \rho = \sqrt{\sum_I \sum_{J<I} \rho_{\tiny IJ}^2} \sim \sqrt{\frac{N(N-1)}{2}} \rho_{\tiny IJ} \propto N \rho_{\tiny IJ}}
ρ 2 = I ∑ J < I ∑ ρ I J 2 , ρ = I ∑ J < I ∑ ρ I J 2 ∼ 2 N ( N − 1 ) ρ I J ∝ N ρ I J
注意,根据前面的讨论,对于随机背景信号,互相关信噪比≫ \gg ≫ 自相关的信噪比,因此这里不用考虑探测器自相关。同时由于N个探测器,有N ( N − 1 ) 2 \frac{N(N-1)}{2} 2 N ( N − 1 ) 个独立组合,信噪比提升接近N倍,而非前面确定性信号N \sqrt{N} N 倍。直觉上如何理解?从信息角度理解,对于模板已知的信号,单探测器匹配滤波所提供的信息已足够多,增加探测器提升有限;对于随机背景,信号本身形式未知,增加探测器提供的信息更多。
带入两个探测器信噪比,展开有:
ρ = [ ∑ I ∑ I < J T o b s I J ∫ d f Γ I J 2 ( f ) S h 2 ( f ) N I ( f ) N J ( f ) ] − 1 / 2 = [ ∫ d f S h 2 ( f ) ∑ I ∑ I < J T o b s I J Γ I J 2 ( f ) N I ( f ) N J ( f ) ] − 1 / 2 = ⟨ R ⟩ T o b s [ ∫ d f S h 2 ( f ) ∑ I ∑ I < J γ I J 2 ( f ) N I ( f ) N J ( f ) ] − 1 / 2 \begin{aligned}
\rho &= \left[\sum_{I}\sum_{I<J} T_{ {\rm obs}_{IJ} } \int df ~ \frac{\Gamma_{IJ}^2(f) S^2_h(f) }{N_I(f)N_J(f)}\right]^{-1/2} \\
&= \left[\int df ~ S^2_h(f) \sum_{I}\sum_{I<J} \frac{T_{ {\rm obs}_{IJ} }\Gamma_{IJ}^2(f)}{N_I(f)N_J(f)}\right]^{-1/2} \\
&= \langle\mathcal{R}\rangle \sqrt{T_{\rm obs}} \left[\int df ~ S^2_h(f) \sum_{I}\sum_{I<J} \frac{\gamma_{IJ}^2(f)}{N_I(f)N_J(f)}\right]^{-1/2}
\end{aligned} ρ = [ I ∑ I < J ∑ T o b s I J ∫ d f N I ( f ) N J ( f ) Γ I J 2 ( f ) S h 2 ( f ) ] − 1 / 2 = [ ∫ d f S h 2 ( f ) I ∑ I < J ∑ N I ( f ) N J ( f ) T o b s I J Γ I J 2 ( f ) ] − 1 / 2 = ⟨ R ⟩ T o b s [ ∫ d f S h 2 ( f ) I ∑ I < J ∑ N I ( f ) N J ( f ) γ I J 2 ( f ) ] − 1 / 2
ρ = ⟨ R ⟩ T o b s [ ∫ d f Ω G W 2 ( f ) ∑ I ∑ I < J ( γ I J ( f ) Ω n I J ( f ) ) 2 ] − 1 / 2 \begin{aligned}
\rho &= \langle\mathcal{R}\rangle \sqrt{T_{\rm obs}} \left[\int df ~ \Omega_{\rm \tiny GW}^2(f) \sum_{I}\sum_{I<J} \left(\frac{\gamma_{IJ}(f)}{\Omega_{n_{IJ}}(f)}\right)^2\right]^{-1/2}
\end{aligned} ρ = ⟨ R ⟩ T o b s [ ∫ d f Ω G W 2 ( f ) I ∑ I < J ∑ ( Ω n I J ( f ) γ I J ( f ) ) 2 ] − 1 / 2
其中T I J T_{IJ} T I J 为I , J I, J I , J 探测器同时观测的时间,这里假设所有探测器观测时间都保持相同。对应的幂率积分灵敏度曲线为:
Ω P L I ( f ) = max α [ Ω α t h r e s h o l d ( f f c ) α ] = ρ t h r e s h o l d ⟨ R ⟩ T 4 π 2 3 H 0 2 max α { [ ∫ d f ∑ I ∑ I < J γ I J 2 ( f ) f 2 α − 6 N I ( f ) N J ( f ) ] − 1 / 2 f α } \boxed{\begin{aligned}
\Omega_{\rm PLI}(f) &= \max_\alpha\left[ \Omega_{\alpha_{\rm threshold}} \left(\frac{f}{f_c}\right)^\alpha \right]\\
&= \frac{\rho_{\rm threshold}}{\langle R\rangle\sqrt{T}} ~ \frac{4\pi^2}{3H_0^2} ~ \max_\alpha \left\{ \left[ \int df \sum_{I}\sum_{I<J} \frac{\gamma_{IJ}^2(f) f^{2\alpha-6}}{N_I(f)N_J(f)}\right]^{-1/2} f^\alpha \right\}
\end{aligned}} Ω P L I ( f ) = α max [ Ω α t h r e s h o l d ( f c f ) α ] = ⟨ R ⟩ T ρ t h r e s h o l d 3 H 0 2 4 π 2 α max ⎩ ⎪ ⎨ ⎪ ⎧ [ ∫ d f I ∑ I < J ∑ N I ( f ) N J ( f ) γ I J 2 ( f ) f 2 α − 6 ] − 1 / 2 f α ⎭ ⎪ ⎬ ⎪ ⎫
Ω P L I ( f ) = max α [ Ω α t h r e s h o l d ( f f c ) α ] = ρ t h r e s h o l d ⟨ R ⟩ T o b s max α { [ ∫ d f ∑ I ∑ I < J ( γ I J ( f ) f α Ω n I J ( f ) ) 2 ] − 1 / 2 f α } \boxed{\begin{aligned}
\Omega_{\rm PLI}(f) &= \max_\alpha\left[ \Omega_{\alpha_{\rm threshold}} \left(\frac{f}{f_c}\right)^\alpha \right]\\
&= \frac{\rho_{\rm threshold}}{\langle R\rangle\sqrt{T_{\rm obs}}} ~\max_\alpha \left\{ \left[ \int df \sum_{I}\sum_{I<J} \left(\frac{\gamma_{IJ}(f) f^{\alpha}}{\Omega_{n_{IJ}}(f)}\right)^2\right]^{-1/2} f^\alpha \right\}
\end{aligned}} Ω P L I ( f ) = α max [ Ω α t h r e s h o l d ( f c f ) α ] = ⟨ R ⟩ T o b s ρ t h r e s h o l d α max ⎩ ⎪ ⎨ ⎪ ⎧ [ ∫ d f I ∑ I < J ∑ ( Ω n I J ( f ) γ I J ( f ) f α ) 2 ] − 1 / 2 f α ⎭ ⎪ ⎬ ⎪ ⎫
ln Ω P L I ( f ) = max α [ ln Ω α t h r e s h o l d − α ln f c + α ln f ] = ln ( ρ t h r e s h o l d ⟨ R ⟩ T 4 π 2 3 H 0 2 ) + max α [ α ln f − 1 2 ln ( ∫ d f ∑ I ∑ I < J γ I J 2 ( f ) f 2 α − 6 N I ( f ) N J ( f ) ) ] \small\begin{aligned}
\ln \Omega_{\rm PLI}(f) &= \max_\alpha\Big[ \ln \Omega_{\alpha_{\rm threshold}} - \alpha \ln f_c + \alpha \ln f \Big]\\
&= \ln \left(\frac{\rho_{\rm threshold}}{\langle R\rangle\sqrt{T}} ~ \frac{4\pi^2}{3H_0^2}\right) + \max_\alpha \left[\alpha \ln f - \frac{1}{2} \ln\left(\int df \sum_{I}\sum_{I<J} \frac{\gamma_{IJ}^2(f) f^{2\alpha-6}}{N_I(f)N_J(f)}\right)\right]
\end{aligned} ln Ω P L I ( f ) = α max [ ln Ω α t h r e s h o l d − α ln f c + α ln f ] = ln ( ⟨ R ⟩ T ρ t h r e s h o l d 3 H 0 2 4 π 2 ) + α max [ α ln f − 2 1 ln ( ∫ d f I ∑ I < J ∑ N I ( f ) N J ( f ) γ I J 2 ( f ) f 2 α − 6 ) ]
ln Ω P L I ( f ) = max α [ ln Ω α t h r e s h o l d − α ln f c + α ln f ] = ln ( ρ t h r e s h o l d ⟨ R ⟩ T o b s ) + max α { α ln f − 1 2 ln [ ∫ d f ∑ I ∑ I < J ( γ I J ( f ) f α Ω n I J ( f ) ) 2 ] } \small\begin{aligned}
\ln \Omega_{\rm PLI}(f) &= \max_\alpha\Big[ \ln \Omega_{\alpha_{\rm threshold}} - \alpha \ln f_c + \alpha \ln f \Big]\\
&= \ln \left(\frac{\rho_{\rm threshold}}{\langle R\rangle\sqrt{T_{\rm obs}}} \right) + \max_\alpha \left\{\alpha \ln f - \frac{1}{2} \ln\left[\int df \sum_{I}\sum_{I<J} \left(\frac{\gamma_{IJ}(f) f^{\alpha}}{\Omega_{n_{IJ}}(f)}\right)^2\right]\right\}
\end{aligned} ln Ω P L I ( f ) = α max [ ln Ω α t h r e s h o l d − α ln f c + α ln f ] = ln ( ⟨ R ⟩ T o b s ρ t h r e s h o l d ) + α max { α ln f − 2 1 ln [ ∫ d f I ∑ I < J ∑ ( Ω n I J ( f ) γ I J ( f ) f α ) 2 ] }
除了上面的两两关联,多探测器情况,还可以直接全部一起关联,计算最佳滤波。比如四个探测器的“四点”关联,以及进一步扩展到任意偶数探测器的整体关联。其中的关键是信号作为零均值高斯随机变量,相互关联(交叉中心矩)可以拆解为两两关联相乘,并对所有置换组合求和。
⟨ s 1 s 2 . . . s 2 N ⟩ = ⟨ s 1 s 2 ⟩ ⟨ s 3 s 4 ⟩ . . . ⟨ s 2 N − 1 s 2 N ⟩ + all possible permutations ⏟ ( 2 N − 1 ) ( 2 N − 3 ) . . . 1 terms \langle s_1 s_2 ... s_{2N}\rangle = \underbrace{\langle s_1 s_2\rangle \langle s_3 s_4\rangle ... \langle s_{2N-1} s_{2N}\rangle + \text{all possible permutations}}_{(2N-1)(2N-3)...1 ~ \text{terms}}
⟨ s 1 s 2 . . . s 2 N ⟩ = ( 2 N − 1 ) ( 2 N − 3 ) . . . 1 terms ⟨ s 1 s 2 ⟩ ⟨ s 3 s 4 ⟩ . . . ⟨ s 2 N − 1 s 2 N ⟩ + all possible permutations
对于仪器噪声,由于不同探测器不相关仍只有自相关项。从而计算信噪比时,分子是互功率谱乘积对所有组合求和,分子则是所有噪声谱乘积,最终信噪比平方就是两两信噪比平方乘积对所有组合求和。
ρ 2 ≈ ρ 12 2 ρ 34 2 . . . ρ 2 N − 1 , 2 N 2 + all possible permutations ⏟ ( 2 N − 1 ) ( 2 N − 3 ) . . . 1 terms \rho^2 \approx \underbrace{\rho^2_{12}~\rho^2_{34}...\rho^2_{2N-1, 2N} + \text{all possible permutations}}_{(2N-1)(2N-3)...1 ~ \text{terms}}
ρ 2 ≈ ( 2 N − 1 ) ( 2 N − 3 ) . . . 1 terms ρ 1 2 2 ρ 3 4 2 . . . ρ 2 N − 1 , 2 N 2 + all possible permutations
注意这里的信噪比是不能与两两相关情况直接对比的,多变量相关与两变量相关对应的信号是不同的,前者是后者的N次方,开方后才可对比,或者统一转换为Ω α t h r e s h o l d \Omega_{\alpha_{\rm threshold}} Ω α t h r e s h o l d 对比探测能力。对比可发现,多探测器整体关联的探测能力总是低于探测器两两组合情况 (Allen & Romano 1999),因此整体关联并非好选择。最后,原则上奇数个探测器也可以关联,但这里的拆分只针对偶数,奇数要如何进行拆分?
高阶效应
非高斯
非各向同性:主要由与本地物质分布耦合的源造成,如银河系的双白矮星
辐射计(点源):不进行立体角积分,计算不同方向的信号强度
球谐展开:
e i k ^ ⋅ z ^ = 4 π ∑ ℓ = 0 ∞ ∑ m = − ℓ ℓ i ℓ j ℓ ( k r ) Y ℓ m ( z ^ ) Y ℓ m ∗ ( k ^ ) e^{i\hat{k}\cdot\hat{z}} = 4\pi\sum_{\ell=0}^\infty \sum_{m=-\ell}^{\ell} i^{\ell} ~ j_{\ell}(k r) ~ Y_\ell^m(\hat{z}) ~ {Y_\ell^m}^*(\hat{k})
e i k ^ ⋅ z ^ = 4 π ℓ = 0 ∑ ∞ m = − ℓ ∑ ℓ i ℓ j ℓ ( k r ) Y ℓ m ( z ^ ) Y ℓ m ∗ ( k ^ )
球谐展开中,指标l l l 代表对角度的划分,取值越大划分的角尺度越细,对应越高阶的多极矩,如l = 0 l=0 l = 0 为各向同性成分(单极),l = 1 l=1 l = 1 为偶极分量;m m m 代表了多极矩的方位指向,或者说在特定方向的投影情况,考虑到方位角共划分了l + 1 l+1 l + 1 份,因此m m m 只需要取− l , . . . , 0 , . . . , l -l,...,0,...,l − l , . . . , 0 , . . . , l ,基于这2 l + 1 2l+1 2 l + 1 种正交解即可组合出所有其余方位。γ I J ( f ) = 1 ⟨ R ⟩ ∫ d 2 k ^ 4 π ∫ d ψ 2 π D I a b e a b ( k ^ ) D J c d e c d ( k ^ ) e i k ^ ⋅ z ^ ı \gamma_{IJ}(f) = \frac{1}{\langle\mathcal{R}\rangle} \int \frac{d^2\hat{k}}{4\pi} \int \frac{d\psi}{2\pi} ~ D_I^{~~ab} e_{ab}(\hat{k}) D_J^{~~cd} e_{cd}(\hat{k}) e^{i\hat{k}\cdot \hat{z}~\imath}
γ I J ( f ) = ⟨ R ⟩ 1 ∫ 4 π d 2 k ^ ∫ 2 π d ψ D I a b e a b ( k ^ ) D J c d e c d ( k ^ ) e i k ^ ⋅ z ^
γ ℓ m ( f , k ^ ) = 1 ⟨ R ⟩ D I a b e a b ( k ^ ) D J c d e c d ( k ^ ) e i k ^ ⋅ z ^ ı \gamma_{\ell m}(f, \hat{k}) = \frac{1}{\langle\mathcal{R}\rangle} ~ D_I^{~~ab} e_{ab}(\hat{k}) D_J^{~~cd} e_{cd}(\hat{k}) e^{i\hat{k}\cdot \hat{z}~\imath}
γ ℓ m ( f , k ^ ) = ⟨ R ⟩ 1 D I a b e a b ( k ^ ) D J c d e c d ( k ^ ) e i k ^ ⋅ z ^
存在圆偏振:e R = e + + i e × 2 , e L = e + − i e × 2 I ( f ) = S R ( f ) + S L ( f ) , V ( f ) = S R ( f ) − S L ( f ) , χ ( f ) = V ( f ) I ( f ) R I ( f ) = R R ( f ) + R L ( f ) , V ( f ) = S R ( f ) − S L ( f ) \begin{gathered}
\bm{e}^R = \frac{\bm{e}^+ + i\bm{e}^\times}{\sqrt{2}}, ~~~~~ \bm{e}^L = \frac{\bm{e}^+ - i\bm{e}^\times}{\sqrt{2}}\\
I(f) = S_R(f) + S_L(f), ~~~ V(f) = S_R(f) - S_L(f), ~~ \chi(f) = \frac{V(f)}{I(f)}\\
\mathcal{R}^I(f) = R_R(f) + R_L(f), ~~~~~ V(f) = S_R(f) - S_L(f)
\end{gathered} e R = 2 e + + i e × , e L = 2 e + − i e × I ( f ) = S R ( f ) + S L ( f ) , V ( f ) = S R ( f ) − S L ( f ) , χ ( f ) = I ( f ) V ( f ) R I ( f ) = R R ( f ) + R L ( f ) , V ( f ) = S R ( f ) − S L ( f )
噪声存在相关时可借助维纳滤波、贝叶斯参数估计等进行改善
非稳态、非高斯会导致统计量非最优,但仍有效;非各向同性会引入偏置
这里的正负号有待确认
γ I J V ( f ) = 1 ⟨ R ⟩ ∫ d 2 k ^ 4 π i ( F I + F J × − F I × F J + ) e i k ^ ⋅ z ^ ı \gamma_{IJ}^{V}(f) = \frac{1}{\langle\mathcal{R}\rangle} \int \frac{d^2\hat{k}}{4\pi} ~ i\left(F^+_I F^\times_J - F^\times_I F^+_J\right) e^{i\hat{k}\cdot \hat{z}~\imath}
γ I J V ( f ) = ⟨ R ⟩ 1 ∫ 4 π d 2 k ^ i ( F I + F J × − F I × F J + ) e i k ^ ⋅ z ^
γ I J V ( f ) = 1 ⟨ R ⟩ ∫ d 2 k ^ 4 π i ( D I a b e a b + D J c d e c d × − D I a b e a b × D J c d e c d + ) e i k ^ ⋅ z ^ ı \gamma_{IJ}^{V}(f) = \frac{1}{\langle\mathcal{R}\rangle} \int \frac{d^2\hat{k}}{4\pi} ~ i\left(D_I^{~~ab} e^+_{ab} D_J^{~~cd} e^\times_{cd} - D_I^{~~ab} e^\times_{ab} D_J^{~~cd} e^+_{cd}\right) e^{i\hat{k}\cdot \hat{z}~\imath}
γ I J V ( f ) = ⟨ R ⟩ 1 ∫ 4 π d 2 k ^ i ( D I a b e a b + D J c d e c d × − D I a b e a b × D J c d e c d + ) e i k ^ ⋅ z ^
γ I J V ( f ) = i D I a b D J c d 1 ⟨ R ⟩ ∫ d 2 k ^ 4 π ( e a b + e c d × − e a b × e c d + ) e i k ^ ⋅ z ^ ı \gamma_{IJ}^{V}(f) = i D_I^{~~ab} D_J^{~~cd} \frac{1}{\langle\mathcal{R}\rangle} \int \frac{d^2\hat{k}}{4\pi} ~ \left(e^+_{ab} e^\times_{cd} - e^\times_{ab} e^+_{cd}\right) e^{i\hat{k}\cdot \hat{z}~\imath}
γ I J V ( f ) = i D I a b D J c d ⟨ R ⟩ 1 ∫ 4 π d 2 k ^ ( e a b + e c d × − e a b × e c d + ) e i k ^ ⋅ z ^
参照之前的做法,引入张量
γ a b c d V ( ı , z ^ ) ≡ i ⟨ R ⟩ ∫ d 2 k ^ 4 π ( e a b + e c d × − e a b × e c d + ) e i k ^ ⋅ z ^ ı γ I J V ( f ) = D I a b D J c d γ a b c d V \begin{aligned}
\gamma^V_{abcd}(\imath, \hat{z}) &\equiv \frac{i}{\langle\mathcal{R}\rangle} \int \frac{d^2\hat{k}}{4\pi} \left(e^+_{ab} e^\times_{cd} - e^\times_{ab} e^+_{cd}\right) e^{i \hat{k}\cdot \hat{z}~\imath}\\
\gamma^V_{IJ}(f) &= D_I^{~~ab} D_J^{~~cd} \gamma^V_{abcd}
\end{aligned} γ a b c d V ( , z ^ ) γ I J V ( f ) ≡ ⟨ R ⟩ i ∫ 4 π d 2 k ^ ( e a b + e c d × − e a b × e c d + ) e i k ^ ⋅ z ^ = D I a b D J c d γ a b c d V
之前的γ a b c d V \gamma^V_{abcd} γ a b c d V 只考虑的是e a b + e c d + + e a b × e c d × e^+_{ab}e^+_{cd} + e^\times_{ab}e^\times_{cd} e a b + e c d + + e a b × e c d × ,这里是交叉相减。最核心的是这里γ a b c d V \gamma^V_{abcd} γ a b c d V 中a b ↔ c d ab \leftrightarrow cd a b ↔ c d 反对称,借助反称张量w a b = ϵ a b c z c w_{ab} = \epsilon_{abc}z^c w a b = ϵ a b c z c 有:
γ a b c d V ( ı , s ^ ) = B ^ ( ı ) ( w a c δ b d + w a d δ b c + w b c δ a d + w b d δ a c ) + C ^ ( ı ) ( δ a b z c z d − δ c d z a z b ) + D ^ ( ı ) ( w a c z b z d + w a d z b z c + w b c z a z d + w b d z a z c ) \begin{aligned}
\gamma^V_{abcd}(\imath, \hat s) = & \hat{B}(\imath)\ (w_{ac} \delta_{bd}+ w_{ad}\delta_{bc}+w_{bc}\delta_{ad}+ w_{bd}\delta_{ac})
+\hat{C}(\imath)\ (\delta_{ab}z_c z_d - \delta_{cd}z_a z_b)\\
& + \hat{D}(\imath)\ (w_{ac}z_b z_d+w_{ad}z_b z_c+w_{bc}z_a z_d+w_{bd}z_a z_c)
\end{aligned} γ a b c d V ( , s ^ ) = B ^ ( ) ( w a c δ b d + w a d δ b c + w b c δ a d + w b d δ a c ) + C ^ ( ) ( δ a b z c z d − δ c d z a z b ) + D ^ ( ) ( w a c z b z d + w a d z b z c + w b c z a z d + w b d z a z c )
将三个张量基分别与上式缩并可以得到对应系数的不同组合,而直接与γ a b c d \gamma_{abcd} γ a b c d 定义式缩可得到对应的组合结果,缩并后积分都相对简单,可直接求解:
( 40 0 8 0 4 0 8 0 8 ) ( B C D ) = − 40 ( j 1 0 ı − 1 j 2 ) \begin{aligned}
\begin{pmatrix}
40 & 0 & 8\\
0 & 4 & 0\\
8 & 0 & 8\\
\end{pmatrix}
\begin{pmatrix}
B\\
C\\
D\\
\end{pmatrix} = -40
\begin{pmatrix}
j_1\\
0\\
\imath^{\tiny -1} j_2
\end{pmatrix}
\end{aligned} ⎝ ⎛ 4 0 0 8 0 4 0 8 0 8 ⎠ ⎞ ⎝ ⎛ B C D ⎠ ⎞ = − 4 0 ⎝ ⎛ j 1 0 − 1 j 2 ⎠ ⎞
( B D ) = − 5 4 ( 1 − 1 − 1 5 ) ( j 1 ı − 1 j 2 ) = − 1 4 ( 4 − 1 0 5 ) ( j 1 j 3 ) \begin{aligned}
\begin{pmatrix}
B \\
D
\end{pmatrix} &= -\frac{5}{4}
\begin{pmatrix}
~1 & -1 \\
-1 & ~5
\end{pmatrix}
\begin{pmatrix}
j_1\\
\imath^{\tiny -1}j_2
\end{pmatrix} = -\frac{1}{4} \begin{pmatrix}
~4 & -1\\
~0 & ~5
\end{pmatrix}
\begin{pmatrix}
j_1\\
j_3
\end{pmatrix}
\end{aligned} ( B D ) = − 4 5 ( 1 − 1 − 1 5 ) ( j 1 − 1 j 2 ) = − 4 1 ( 4 0 − 1 5 ) ( j 1 j 3 )
γ I J V ( f ) = [ ( cos β − 1 ) j 1 ( ı ) + ( 7 8 + 3 8 cos β ) j 3 ( ı ) ] sin β 2 sin [ 2 ( σ 1 + σ 2 ) ] \gamma_{IJ}^{V}(f) = \left[ (\cos\beta-1)j_1(\imath) + \left(\frac{7}{8} + \frac{3}{8}\cos\beta\right)j_3(\imath)\right] \sin\frac{\beta}{2} \sin[2(\sigma_1 + \sigma_2)]
γ I J V ( f ) = [ ( cos β − 1 ) j 1 ( ) + ( 8 7 + 8 3 cos β ) j 3 ( ) ] sin 2 β sin [ 2 ( σ 1 + σ 2 ) ]
另一方面,利用探测器张量“对称无迹”的特性,对于γ I J \gamma_{IJ} γ I J 有:
γ I J = D I a b D J c d γ a b c d V = D I a b D J c d ( 4 B w a c δ b d + 4 D w a c z b z d ) \begin{aligned}
\gamma_{IJ} &= D_I^{~~ab} D_J^{~~cd}\gamma^V_{abcd} = D_I^{~~ab} {D_J}^{cd} \left(4B ~w_{ac}\delta_{bd} + 4D ~ w_{ac} z_b z_d\right)\\
\end{aligned} γ I J = D I a b D J c d γ a b c d V = D I a b D J c d ( 4 B w a c δ b d + 4 D w a c z b z d )
对于等边三角构型的空间引力波探测器网络(LISA-Taiji),考虑到A、E通道,总体(平方)响应为:
Y I J = γ A A ′ 2 + γ A E ′ 2 + γ E A ′ 2 + γ E E ′ 2 Y_{IJ} = \gamma^2_{AA'} + \gamma^2_{AE'} + \gamma^2_{EA'} + \gamma^2_{EE'}
Y I J = γ A A ′ 2 + γ A E ′ 2 + γ E A ′ 2 + γ E E ′ 2
带入上面γ I J \gamma_{IJ} γ I J 表达式,记4 B , 4 D 4B, 4D 4 B , 4 D 为b 0 , b 1 b_0, b_1 b 0 , b 1 有:
Y I J = b 0 b 0 X 00 + b 0 b 1 X 01 + b 1 b 0 X 10 + b 1 b 1 X 11 X 00 = ∑ I = A , E ∑ J = A ′ , E ′ ( D I a b D J c d w a c δ b d ) 2 X 01 = X 10 = ∑ I = A , E ∑ J = A ′ , E ′ ( D I a b D J c d w a c δ b d ) ( D I i j D J k l w i k z j z l ) X 11 = ∑ I = A , E ∑ J = A ′ , E ′ ( D I a b D J c d w a c z b z d ) 2 \begin{aligned}
Y_{IJ} &= b_0b_0X_{00} + b_0b_1X_{01} + b_1b_0X_{10} + b_1b_1X_{11}\\
X_{00} &= \sum_{I=A,E} \sum_{J=A',E'} \left(D_I^{~~ab} {D_J}^{cd} w_{ac}\delta_{bd}\right)^2 \\
X_{01} = X_{10} &= \sum_{I=A,E} \sum_{J=A',E'} \left(D_I^{~~ab} {D_J}^{cd} w_{ac}\delta_{bd}\right) \left(D_I^{~~ij} {D_J}^{kl} w_{ik} z_j z_l\right)\\
X_{11} &= \sum_{I=A,E} \sum_{J=A',E'} \left(D_I^{~~ab} {D_J}^{cd} w_{ac} z_b z_d\right)^2
\end{aligned} Y I J X 0 0 X 0 1 = X 1 0 X 1 1 = b 0 b 0 X 0 0 + b 0 b 1 X 0 1 + b 1 b 0 X 1 0 + b 1 b 1 X 1 1 = I = A , E ∑ J = A ′ , E ′ ∑ ( D I a b D J c d w a c δ b d ) 2 = I = A , E ∑ J = A ′ , E ′ ∑ ( D I a b D J c d w a c δ b d ) ( D I i j D J k l w i k z j z l ) = I = A , E ∑ J = A ′ , E ′ ∑ ( D I a b D J c d w a c z b z d ) 2
事实可以证明Y I J Y_{IJ} Y I J 以及X i j X_{ij} X i j 不依赖引力波探测器在探测平面内的具体指向,旋转不变。从而探测器张量所提供信息完全由探测器平面法矢量 e ^ I , e ^ J \hat{e}_I, \hat{e}_J e ^ I , e ^ J 决定(长波近似),进而 Y I J , X i j Y_{IJ}, X_{ij} Y I J , X i j 完全由 e ^ I , e ^ J , z ^ \hat{e}_I, \hat{e}_J, \hat{z} e ^ I , e ^ J , z ^ 决定,总共六个自由度。考虑到随机背景的各向同性,探测器网络的整体的三维旋转不影响探测,因此影响整体响应只有三个自由度,简单起见可选择用三个单位矢量的内积(夹角)表示。
值得注意的
探测器连线矢量反向时,V分量的辅助张量会反号,但V分量的重叠约化不受影响?对应探测器的互换??
探测器平面法矢量反号,交叉项乘积会反号,但因为有平方,等效ORF不受影响,只与平面夹角有关
超过π / 2 \pi/2 π / 2 的矢量夹角以及负的夹角都与0 ∼ π / 2 0 \sim \pi/2 0 ∼ π / 2 的矢量夹角效果等价
当频率趋于零f → 0 f \rightarrow 0 f → 0 ,j 1 , j 3 j_1, j_3 j 1 , j 3 在零点取0,此时γ I J V = 0 \gamma^V_{IJ} = 0 γ I J V = 0 ;当探测器位于同一位置r I J → 0 r_{IJ} \rightarrow 0 r I J → 0 ,同样有γ I J V = 0 \gamma^V_{IJ} = 0 γ I J V = 0 。
σ 1 = σ 2 = 0 \sigma_1 = \sigma_2 = 0 σ 1 = σ 2 = 0
A-A, E-E
A-E, E-A
χ ( f ) = V ( f ) I ( f ) = Ω V ( f ) Ω I ( f ) \chi(f) = \frac{V(f)}{I(f)} = \frac{\Omega_V(f)}{\Omega_I(f)}
χ ( f ) = I ( f ) V ( f ) = Ω I ( f ) Ω V ( f )
根据误差传递公式,c = a b c = \frac{a}{b} c = b a ,若a , b a, b a , b 不相关,有σ c μ c = ( σ a μ a ) 2 + ( σ b μ b ) 2 \frac{\sigma_c}{\mu_c} = \sqrt{\left(\frac{\sigma_a}{\mu_a}\right)^2 + \left(\frac{\sigma_b}{\mu_b}\right)^2} μ c σ c = ( μ a σ a ) 2 + ( μ b σ b ) 2 :???
ρ χ 2 = ρ V 2 ρ I 2 ρ V 2 + ρ I 2 \rho^2_{\chi} = \frac{\rho^2_V~\rho^2_I}{\rho^2_V + \rho^2_I}
ρ χ 2 = ρ V 2 + ρ I 2 ρ V 2 ρ I 2
前景及噪声扣除
Pieroni & Barausse 2020
Adams & Cornish 2014, Adams & Cornish 2010
Tinto et al. 2001, Hogan & Bender 2001