Sei sulla pagina 1di 11

Notes

Nikos Karnesis, Mauro Pieroni


November 28, 2019

Contents
1 Some general things (Angelo) 1

2 Three channels 3
2.1 A fairly general template . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.1.1 A first approximation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.1.2 An alternative approximation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.1.3 Merging the two approximations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

3 Setting up the data analysis tools 8

4 Some general things (Mauro) 11

1 Some general things (Angelo)


Assuming a continuous time description within each segment represented by s(t), where t belongs to the
interval [−T /2, T /2], we can define the Fourier transform
Z T /2
s̃ (f ) = dt ei 2πf t s (t) . (1.1)
−T /2

The two point correlation of the signals reads


Z T /2 Z T /2
0 0
∗ 0
hs̃(f )s̃ (f )i = dt dt0 ei2π(f t−f t)
hs (t) (t0 )i (1.2)
−T /2 −T /2

We can define τ ≡ t0 − t and write


Z T /2 Z T /2−t
0
∗ 0 −f ) i2πτ
hs̃(f )s̃ (f )i = dt dτ hs (t + τ ) s (t)iei2πt(f e
−T /2 −T /2−t
Z T /2 Z T
0
−f ) i2πf 0 τ
= dt dτ θ (τ − (−T /2 − t)) θ (τ − (T /2 − t)) hs (t + τ ) s (t)iei2πt(f e
−T /2 −T
Z T Z T /2
0
−f ) i2πf 0 τ
= dτ dt θ (t − (−T /2 − τ )) θ (t − (T /2 − τ )) hs (t + τ ) s (t)iei2πt(f e
−T −T /2
Z T Z min[−T /2,−T /2−τ ]
0
−f )
= dτ dt θ (t − (−T /2 − τ )) θ (t − (T /2 − τ )) hs (t + τ ) s (t)iei2πt(f
−T max[−T /2,−T /2−τ ]
Z 0 Z T /2
0
−f ) i2πf 0 τ
= dτ dt hs (t + τ ) s (t)iei2πt(f e
−T −T /2−τ
Z T Z T /2−τ
0
−f ) i2πf 0 τ
+ dτ dt hs (t + τ ) s (t)iei2πt(f e
0 −T /2

1
Nikos Karnesis, Mauro Pieroni Notes 2019

T
sin [(f − f 0 ) π (T − τ )]
Z
= dτ (2 cos [(f + f 0 ) πτ ]) hs (t + τ ) s (t)i
0 (f − f 0 ) π

The last two pieces in the integrand are even functions of τ so we can write the integral on a symmetric
integration domain as
T
sin [(f − f 0 ) π (T − |τ |)]
Z
0
∗ 0
hs̃(f )s̃ (f )i = dτ 0
hs (t + τ ) s (t)ieiπ(f +f )τ (1.4)
−T (f − f ) π

Since τ ∼ 1/∆f  T then the first function can be extracted from the integral which becomes
T
sin [(f − f 0 ) πT ]
Z
0
hs̃(f )s̃∗ (f 0 )i ' dτ hs (t + τ ) s (t)ieiπ(f +f )τ (1.5)
(f − f 0 ) π −T

We can show that the first term is the usual (truncated) delta function δT (f − f 0 ) which for T → ∞ can be
traded with a Dirac delta and so the two point function reads
Z T Z ∞
hs̃(f )s̃∗ (f 0 )i = δT (f − f 0 ) dτ hs (t + τ ) s (t)iei2πf τ = δ (f − f 0 ) dτ hs (t + τ ) s (t)iei2πf τ (1.6)
−T −∞

What is usually done is to re-define the previous integral as


1
hs̃(f )s̃∗ (f 0 )i ≡ δ (f − f 0 ) S (f ) (1.7)
2
This is the power spectrum density of the signal. Since the s(t) are real, the autocorrelation is also real
and taking the complex conjugate we see that S(f )∗ = S(−f ). The invariance under time translations
(stationarity) implies that S(f ) = S(−f ) that means that the power spectrum density is real and an even
function of the frequency. For this reason is conventional to define the “one-side” power spectra S(f ), only
for positive (physical) frequencies with a 1/2 factor and the absolute value.
What has been described can, of course, be extended also to cross-correlation of signals (of different channels
in the LISA case) and to the (cross-)correlation of noises. Assuming that the noise is Gaussian and stationary
we can write it as
1
hñi (f )ñ∗j (f 0 )i = δ (f − f 0 ) Nij (f ) (1.8)
2
(in the case where the noise is considered un-correlated we add a δij ). Nij (f ) (or what is usually called Pn
in the literature) is the noise power spectrum density and it has dimension Hz−1 , since ñ(f ) has dimension
of the inverse of a frequency and δ as well.

2
Nikos Karnesis, Mauro Pieroni Notes 2019

2 Three channels
The standard LISA output channels A, E and T are constructed as
1 ˜ 
d˜A = 2dX − d˜Y − d˜Z ,
3
1  
˜
dE = √ d˜Z − d˜Y ,
3
1  
d˜T = d˜X + d˜Y + d˜Z . (2.1)
3
We can thus write the conversion from d˜i = (d˜X , d˜Y , d˜Z ) with a rotation matrix c of form
2
− 13 − 31

3
cOi =  0 − √13 √13  . (2.2)
 
1 1 1
3 3 3

where the O index labels the rows and the i index labels the columns.

With this, we can now express the two point correlations in A, E, T as:
1 h ˜ ˜∗ i
hd˜A d˜∗A i = 4hdX dX i + hd˜Y d˜∗Y i + hd˜Z d˜∗Z i − 2(hd˜X d˜∗Y i + hd˜∗X d˜Y i) − 2(hd˜X d˜∗Z i + hd˜∗X d˜Z i) + (hd˜Y d˜∗Z i + hd˜∗Y d˜Z i) ,
9
˜ ˜ 1 h ˜ ˜∗ i

hdE dE i = hdZ dZ i + hd˜Y d˜∗Y i − (hd˜Z d˜∗Y i + hd˜∗Z d˜Y i) ,
3
˜ ˜ 1 h ˜ ˜∗ i

hdT dT i = hdX dX i + hd˜Y d˜∗Y i + hd˜Z d˜∗Z i + (hd˜X d˜∗Y i + hd˜∗X d˜Y i) + (hd˜X d˜∗Z i + hd˜∗X d˜Z i) + (hd˜Y d˜∗Z i + hd˜∗Y d˜Z i) .
9
1 h i
˜ ˜
hdA dE i = √ −2hd˜X d˜∗Y i + 2hd˜X d˜∗Z i + hd˜Y d˜∗Y i − hd˜Y d˜∗Z i + hd˜Z d˜∗Y i − hd˜Z d˜∗Z i ,

3 3
1 h ˜ ˜∗ i
hd˜A d˜∗T i = 2hdX dX i − hd˜Y d˜∗Y i − hd˜Z d˜∗Z i + 2hd˜X d˜∗Y i − hd˜Y d˜∗X i + 2hd˜X d˜∗Z i − hd˜Z d˜∗X i − (hd˜Y d˜∗Z i + hd˜Z d˜∗Y i) ,
9
1 h i
hd˜E d˜∗T i = √ hd˜Z d˜∗Z i − hd˜Y d˜∗Y i − hd˜Y d˜∗X i + hd˜Z d˜∗X i + hd˜Z d˜∗Y i − hd˜Y d˜∗Z i .
3 3
(2.3)
Up to this point we have only assumed signal and noise to be Gaussian and signal and noise to be uncorrelated.

2.1 A fairly general template


In the following we assume:
1. The two point signal correlation for XYZ to be:

hw̃i (f )w̃j∗ (f 0 )i = δ(f − f 0 )R̃ij (f )Ph (f ) (2.4)

where R̃ij is (frequency dependent) response function for my TDI channels.


2. The two point noise correlation for XYZ to be:

hñi (f )ñ∗j (f 0 )i = δ(f − f 0 ) Ñij



(f ) (2.5)

(Notice that I have no off-diagonal noise correlations!) where

Ω 4π 2 f 3
Ñij (f ) = Ñij (f ) (2.6)
3(H/h)2

with H denoting the Hubble parameter, h the dimensionless Hubble parameter and Ñij (f ) are the
TDI X Y Z and correlations as labeled by the indexes i, j of the instrument which is connected to the
strain as Sn ≡ N/R.

3
Nikos Karnesis, Mauro Pieroni Notes 2019

Notice that if I substitute eq. (2.4) and eq. (2.5) into eq. (4.1) I get (at the same frequency and zero elsewhere):

hdi d˜∗j i = R̃ij (f )Ph (f ) + Ñij



(f ) . (2.7)

With this I can thus substitute into eq. (2.3) to get:


1 nh i
hd˜A d˜∗A i = 4R̃XX + R̃Y Y + R̃ZZ Ph + 4ÑXX Ω
+ ÑYΩY + ÑZZ Ω
+
9 h i h i h i
∗ ∗
−2 R̃XY + R̃XY Ph − 2 R̃XZ + R̃XZ Ph + R̃Y Z + R̃Y∗ Z Ph +
h i h i h io
Ω Ω∗ Ω Ω∗
−2 ÑXY + ÑXY − 2 ÑXZ + ÑXZ + ÑYΩZ + ÑYΩ∗Z ,
1 nh i h i
hd˜E d˜∗E i = R̃Y Y + R̃ZZ Ph − R̃Y Z + R̃Y∗ Z Ph +
3 h io
+ÑYΩY + ÑZZ Ω
− ÑYΩZ + ÑYΩ∗Z ,
1 nh i
hd˜T d˜∗T i = R̃XX + R̃Y Y + R̃ZZ Ph + ÑXX Ω
+ ÑYΩY + ÑZZΩ
+
9 h i h i h i
∗ ∗
+ R̃XY + R̃XY Ph + R̃XZ + R̃XZ Ph + R̃Y Z + R̃Y∗ Z Ph +
h i h i h io
Ω Ω∗ Ω Ω∗
+ ÑXY + ÑXY + ÑXZ + ÑXZ + ÑYΩZ + ÑYΩ∗Z .
1 nh i
hd˜A d˜∗E i = √ R̃Y Y − R̃ZZ − 2R̃XY + 2R̃XZ − R̃Y Z + R̃Y∗ Z Ph +
3 3
o
+ÑYΩY − ÑZZ Ω
− 2ÑXY + 2ÑXZ − ÑY Z + ÑY∗ Z ,
1 nh i
hd˜A d˜∗T i = 2R̃XX − R̃Y Y − R̃ZZ Ph + 2ÑXX Ω
− ÑYΩY − ÑZZ Ω
+
9 h i
∗ ∗
+ 2R̃XY − R̃XY + 2R̃XZ − R̃XZ − R̃Y Z − R̃Y∗ Z Ph +
h io
∗ ∗
+ 2ÑXY − ÑXY + 2ÑXZ − ÑXZ − ÑY Z − ÑY∗ Z ,
1 nh i
hd˜E d˜∗T i = √ R̃ZZ − R̃Y Y − R̃XY ∗ ∗
+ R̃XZ + R̃Y∗ Z − R̃Y Z Ph
3 3
o
∗ ∗
+ ÑZZΩ
− ÑYΩY − ÑXY + ÑXZ + ÑY∗ Z − ÑY Z . (2.8)

Notice that R̃XX , R̃Y Y and R̃ZZ are real and positive for all frequencies while R̃XY , R̃XZ and R̃Y Z are real
and negative at small frequency and real and oscillating at larger frequencies! This is what will make the
response function in T equal to zero at small frequency!!

2.1.1 A first approximation


In the following we proceed with a first approximation on the cross correlation for XYZ:

hñi (f )ñ∗j (f 0 )i = δ(f − f 0 ) δij Ñ(i)



(f ) (2.9)

Substituting eq. (2.4) and eq. (2.9) into eq. (2.7) I get (at the same frequency and zero elsewhere):

hdi d˜∗j i = R̃ij (f )Ph (f ) + δij Ñ(i)



(f ) . (2.10)

Again, this can be substituted into eq. (2.3) to get:


1 nh i
hd˜A d˜∗A i = 4R̃XX + R̃Y Y + R̃ZZ Ph + 4Ñ(XX)Ω Ω
+ Ñ(Y Ω
Y ) + Ñ(ZZ)
9 h i h i h i o
∗ ∗
−2 R̃XY + R̃XY Ph − 2 R̃XZ + R̃XZ Ph + R̃Y Z + R̃Y∗ Z Ph ,
1 nh i h i o
hd˜E d˜∗E i = R̃ZZ + R̃Y Y Ph + Ñ(YΩ Ω ∗
Y ) + Ñ(ZZ) − R̃Y Z + R̃Y Z Ph ,
3

4
Nikos Karnesis, Mauro Pieroni Notes 2019

1 nh i
hd˜T d˜∗T i = Ω
R̃XX + R̃Y Y + R̃ZZ Ph + Ñ(XX) Ω
+ Ñ(Y Ω
Y ) + Ñ(ZZ) +
9 h i h i h i o
∗ ∗
+ R̃XY + R̃XY Ph + R̃XZ + R̃XZ Ph + R̃Y Z + R̃Y∗ Z Ph .
1 nh i o
hd˜A d˜∗E i = √ −2R̃XY + 2R̃XZ − R̃Y Z + R̃∗Y Z + R̃Y Y − R̃ZZ Ph + Ñ(Y Ω Ω
Y ) − Ñ(ZZ) ,
3 3
1 nh i
hd˜A d˜∗T i = Ω
2R̃XX − R̃Y Y − R̃ZZ Ph + 2Ñ(XX) Ω
− Ñ(Y Ω
Y ) − Ñ(ZZ) +
9 h i o
∗ ∗
+ 2R̃XY − R̃XY + 2R̃XZ − R̃XZ − R̃Y Z − R̃Y∗ Z Ph ,
1 nh i
hd˜E d˜∗T i = √ Ω
R̃ZZ − R̃Y Y Ph + Ñ(ZZ) Ω
− Ñ(Y Y) +
3 3
h i o
∗ ∗
+ −R̃XY + R̃XZ + R̃Y∗ Z − R̃Y Z Ph . (2.11)

Again, the signs of the response functions in X, Y, Z should make in T equal to zero at small frequency, as
expected.

2.1.2 An alternative approximation


An alternative set of approximations we can do is:
• ÑXX = ÑY Y = ÑZZ .

• ÑXY = ÑXZ = ÑY Z .


• R̃XX = R̃Y Y = R̃ZZ .
• R̃XY = R̃XZ = R̃Y Z .

With these approximations eq. (2.3) reduces to:


1n h i h io
hd˜A d˜∗A i = 6R̃XX Ph + 6ÑXXΩ
− 3 R̃XY + R̃XY∗
Ph − 3 ÑXYΩ Ω∗
+ ÑXY ,
9
1 n h i h io
hd˜E d˜∗E i = 2R̃XX Ph − R̃XY + R̃XY ∗
Ph + 2ÑXXΩ
− ÑXYΩ Ω∗
+ ÑXY ,
3
1 n h i h io
hd˜T d˜∗T i = 3R̃XX Ph + 3ÑXXΩ
+ 3 R̃XY + R̃XY∗
Ph + 3 ÑXYΩ Ω∗
+ ÑXY .
9
1 nh i o
hd˜A d˜∗E i = √ −R̃XY + R̃XY∗
Ph − ÑXY + ÑXY ∗
,
3 3
1 nh i h io
hd˜A d˜∗T i = ∗
R̃XY − R̃XY Ph + ÑXY − ÑXY∗
,
3
1 nh i o
hd˜E d˜∗T i = √ −R̃XY + R̃XY∗ ∗
Ph + ÑXY − ÑXY . (2.12)
3 3
Again, the signs of the response functions in X, Y, Z should make in T equal to zero at small frequency, as
expected. Moreover, since the off diagonal noise correlation functions are imaginary, they are NOT going to
cancel out in the cross correlations of different channels! This actually happens only at fairly large frequencies
(f & 10−3 Hz), for very small frequencies ÑXY is real and thus we expect the cross correlations to be zero at
very small frequencies (I tested this increasing the number of chunks!). Some correlated data are shown in
fig. 2.1. One interesting comment: at small frequencies hd˜T d˜∗T i is slightly shifted with respect to the orange

curve. This shift can be predicted theoretically using that (again at small frequencies) ÑXX = ÑXY + ÑXY
˜ ˜∗ 6
so that, since the signal is expected to cancel out at small frequencies, hdT dT i ' 9 ÑXX (I checked that
introducing a 3/2 factor to compensate for this the two curves match)! Similarly we can check that for
hd˜A d˜∗A i, hd˜E d˜∗E i at small frequency the noise is suppressed by a factor 3.

5
Nikos Karnesis, Mauro Pieroni Notes 2019

Figure 2.1: My correlated data on XYZ and AET

2.1.3 Merging the two approximations


In the following we proceed with three additional assumptions with respect to section 2.1.1:
• Ñ(XX) = Ñ(Y Y ) = Ñ(ZZ) .

• R̃XX = R̃Y Y = R̃ZZ .


• R̃XY = R̃XZ = R̃Y Z .
With these assumptions my correlations in section 2.1 reduce to
1n h i o
hd˜A d˜∗A i = Ω
6R̃XX Ph (f ) + 6Ñ(XX) ∗
(f ) − 3 R̃XY + R̃XY Ph (f ) ,
9

6
Nikos Karnesis, Mauro Pieroni Notes 2019

1n h i o
hd˜E d˜∗E i = Ω
2R̃XX Ph (f ) + 2Ñ(XX) ∗
(f ) − R̃XY + R̃XY Ph (f ) ,
3
1n h i o
hd˜T d˜T i =
∗ Ω
3R̃XX Ph (f ) + 3Ñ(XX) ∗
(f ) + 3 R̃XY + R̃XY Ph (f ) .
9
1 nh i o
hd˜A d˜E i =

√ −R̃XY + R̃XY ∗
Ph ,
3 3
1 nh i o
hd˜A d˜∗T i = ∗
R̃XY − R̃XY Ph (f ) = 0 ,
3
1 nh ∗ i o
hd˜E d˜∗T i = √ R̃XY − R̃XY Ph (f ) = 0 . (2.13)
3 3
In this approximation the noise cancels out in the cross correlations of different channels, moreover, since
the responses are real, all the cross correlations are zero! A test of this is shown in fig. 2.2.

Figure 2.2: My uncorrelated data on XYZ and AET

7
Nikos Karnesis, Mauro Pieroni Notes 2019

3 Setting up the data analysis tools


From now on we work under the assumptions of section 2.1.2. Moreover, to have a simpler connection with
theory what we will call our data is:
" #
0 4π 2 f 3 hd˜i d˜∗j i 0 R̃ij 2 2
Dij δ(f − f ) ≡ = δ(f − f ) (f ) h ΩGW (f ) + h ΩN,ij (f ) =
3(H/h)2 RXX R̃XX (3.1)
0 2 2
 
≡ δ(f − f ) Rij h ΩGW (f ) + h ΩN,ij (f ) ,

notice that in the case of a single channel that we treated in [?], ΩN is precisely the sensitivity curve.
Matching this definition with the system in section 3 we get:
∗ ∗
2 1 R̃XY + R̃XY 2 1 R̃XY + R̃XY
RAA = − , REE = − ,
3 3 R̃XX 3 3 R̃XX
∗ ∗
1 1 R̃XY + R̃XY 1 R̃XY − R̃XY
RT T = + . RAE =− √ , (3.2)
3 3 R̃XX 3 3 R̃XX
∗ ∗
1 R̃XY − R̃XY 1 R̃XY − R̃XY
RAT = , RET =− √ ,
3 R̃XX 3 3 R̃XX
With these approximations eq. (2.3) reduces to:
1n h i h io
hd˜A d˜∗A i = 6R̃XX Ph + 6ÑXXΩ
− 3 R̃XY + R̃XY∗
Ph − 3 ÑXYΩ Ω∗
+ ÑXY ,
9
1 n h i h io
hd˜E d˜∗E i = 2R̃XX Ph − R̃XY + R̃XY ∗
Ph + 2ÑXXΩ
− ÑXYΩ Ω∗
+ ÑXY ,
3
1 n h i h io
hd˜T d˜∗T i = 3R̃XX Ph + 3ÑXXΩ
+ 3 R̃XY + R̃XY∗
Ph + 3 ÑXYΩ Ω∗
+ ÑXY .
9
1 nh i o
hd˜A d˜∗E i = √ −R̃XY + R̃XY∗
Ph − ÑXY + ÑXY ∗
,
3 3
1 nh i h io
hd˜A d˜∗T i = ∗
R̃XY − R̃XY Ph + ÑXY − ÑXY∗
,
3
1 nh i o
hd˜E d˜∗T i = √ −R̃XY + R̃XY∗ ∗
Ph + ÑXY − ÑXY . (3.3)
3 3
notice that since Rij is real, Rij is zero off-diagonal, and also:
" #
2 3 Ω Ω Ω∗
4π f 2 Ñ 1 Ñ + Ñ
h2 ΩN,AA = h2 ΩN,EE ≡ XX
− XY XY
, (3.4)
3(H/h)2 3 R̃XX 3 R̃XX
" #
2 4π 2 f 3 Ω
1 ÑXX Ω
1 ÑXY Ω∗
+ ÑXY
h ΩN,T T ≡ + , (3.5)
3(H/h)2 3 R̃XX 3 R̃XX
" #
2 2 4π 2 f 3 1 ÑXYΩ
− ÑXYΩ∗
h ΩN,AE = h ΩN,ET ≡ − √ (3.6)
3(H/h)2 3 3 R̃XX
" #
2
√ 2 4π 2 f 3 Ω
1 ÑXY Ω∗
− ÑXY
h ΩN,AT = − 3 h ΩN,AE ≡ (3.7)
3(H/h)2 3 R̃XX
(3.8)

Given some analytical expressions for the response functions we can thus write a simple Gaussian likeli-
hood to describe our data:
!2
NChunks X X Dij,k − Rij h2 ΩGW (fk , θ) ~ − h2 ΩN,ij (fk , ~n)
− ln L = , (3.9)
2 i,j
σij,k
k

8
Nikos Karnesis, Mauro Pieroni Notes 2019

where the index k runs over the data points, the indexes i, j run over the different channels and σij,k is the
estimated error on the data point k in the combination of channels i, j.

9
Nikos Karnesis, Mauro Pieroni Notes 2019

Figure 3.1: My data compared with theoretical expectations. What I am plotting is abs of the data after
coarse graining. Notice that doing abs and then coarse grain is not equivalent because that loses track of the
relative phases!!

10
Nikos Karnesis, Mauro Pieroni Notes 2019

4 Some general things (Mauro Old)


At this point we could express d˜i as w̃i + ñi where h̃i is the signal convolved with the response function of
the detector w̃i = Riab hab (the ’real’ signal is denoted with s̃i ) and ñi is the noise measured on each channel.
Let us express the two point correlation functions of data in the X, Y, Z channels as:

hd˜i d˜∗j i = hw̃i w̃j∗ i + hñi ñ∗j i . (4.1)

By defining Fourier transform (using only positive frequencies) as:


Z ∞h i
d(t) = ˜ ) + e−if t d(−f
eif t d(f ˜ ) df (4.2)
0

together with the fact that since d(t) is real, which implies d(f ˜ ) = d˜∗ (−f ) (which also holds for signal and
noise). These imply that the two point correlation function in real space is:
Z ∞Z ∞h i
0
hd(t)d(t)i = ˜ 0 )i + ei(f −f 0 )t hd(f
˜ )d(f
ei(f +f )t hd(f ˜ )d˜∗ (f 0 )i + e−i(f −f 0 )t hd˜∗ (f )d(f
˜ 0 )i + e−i(f +f 0 )t hd˜∗ (f )d˜∗ (f 0 )i df df 0
0 0
(4.3)
and recalling that assuming a statistical isotropic SGWB and a noise which is uncorrelated between different
frequencies we can define the power spectra1 as something like:

hs̃(f )s̃∗ (f 0 )i = δ(f − f 0 )Ph (f ) hñ(f )ñ∗ (f 0 )i = δ(f − f 0 )N Ω (f ) (4.4)

we can immediately see2 that the only terms which contribute to the correlators in time domain are:
Z ∞Z ∞h i Z ∞
i(f −f 0 )t ˜ ˜∗ 0 −i(f −f 0 )t ˜∗ ˜ 0 0
hd(t)d(t)i = e hd(f )d (f )i + e hd (f )d(f )i df df = 2 [N (f ) + R(f )S(f )] df ,
0 0 0
(4.5)
as it should be. Notice that in eq. (4.4) I assumed both the signal and the noise spectra to be real. For
example, this is not the case for correlations between different channels! However, this is not going to be a
problem since, given the correlators:

hñi (f )ñ∗j (f 0 )i = δ(f − f 0 )Nij



(f ) , (4.6)
Ω Ω∗
the only terms contributing to hdi (t)dj (t)i are proportional to Nij (f ) + Nij (f )!

To conclude this section let me stress one more thing:


Z ∞Z ∞h i Z ∞
0
hd(t)d(−t)i = ei(f +f )t hd(f ˜ 0 )i + e−i(f +f 0 )t hd˜∗ (f )d˜∗ (f 0 )i df df 0 = 2
˜ )d(f [N (f ) + R(f )S(f )] df ,
0 0 0
(4.7)
were I have used:

hs̃(f )s̃(f 0 )i = δ(f + f 0 )Ph (f ) hñ(f )ñ(f 0 )i = δ(f + f 0 )N Ω (f ) , (4.8)

which is fine.

1 N.B. what I called N (f ) is the TDI X in units of Ω! More on this is said in section 2.1.1
2 Since both f and f 0 are positive, for the two terms without * or with two * the condition set by the δ is never satisfied.

11

Potrebbero piacerti anche