Effects of Survey Geometry
on Power Spectrum Covariance
for Cosmic Shear Survey
Ryuichi Takahashi (Hirosaki U)
with Shunji Soma, Masahiro Takada
and Issha Kayo
RT+, in preparation
What survey shape is the BEST (or WORST) to extract the cosmological information in weak-lensing cosmic shear survey ?
Abstract
Quiz
Which observational survey geometry is better to extract the cosmological information in weak-lensing cosmic shear survey ?
“Square” “Belt” “Elongated Rectangular”
the surface areas are same
VS
“Belt” “Elongated Rectangular”
Quiz
better
the surface areas are same
Which observational survey geometry is better to extract the cosmological information in weak-lensing cosmic shear survey ?
“Square” VS
Quiz2
“Square” “Sparse Sampling”
the surface areas are same
Which observational survey geometry is better to extract the cosmological information in weak-lensing cosmic shear survey ?
VS
Quiz2
“Square” “Sparse Sampling”
the surface areas are same
Which observational survey geometry is better to extract the cosmological information in weak-lensing cosmic shear survey ?
VS better
・ larger-scale fluctuations can be measured @large scales
The above observational shapes have advantages at both large and small scales:
“Sparse Sampling”
“Belt” “Elongated Rectangular”
Summary
・ the non-Gaussian error can be suppressed @small scales
(e.g. Kaiser 1986, 1998; Kilbinger & Schneider 2004; Chiang+ 2013)
・ larger-scale fluctuations can be measured @large scales
The above observational shapes have advantages at both large and small scales:
“Sparse Sampling”
“Belt” “Elongated Rectangular”
Summary
(e.g. Kaiser 1986, 1998; Kilbinger & Schneider 2004; Chiang+ 2013)
・ the non-Gaussian error can be suppressed @small scales
“Sparse Sampling”
“Belt” “Elongated Rectangular”
Summary
Signal-to-Noise in power spectrum measurement can be 3 times higher than the S/N in the square observational shape (→ 9 times larger survey area)
Introduction
Weak-lensing cosmic shear is a powerful observational tool to constrain the cosmological models (including dark energy).
previous surveys : SDSS, CHFTLS, COSMOS, …
on-going & future surveys : Subaru HSC, DES, LSST, …
from LSST homepage
Convergence field 𝜅(𝜃 ) : projected surface mass density at angular position
Convergence power spectrum 𝐶(ℓ)
𝐶 ℓ =1
𝑁ℓ 𝜅 (ℓ)
2
ℓ
: convergence in Fourier domain
: # of mode
𝜃
: multipole ℓ
𝜅 (ℓ)
𝑁ℓ
(=shear power spectrum)
∆𝐶(ℓ)
𝐶(ℓ)=
2
𝑁ℓ
: # of mode
Gaussian error ∝
Measurement error @ large scales
Measurement errors of cosmological parameters
𝑁ℓ
∝ observational area
Gaussian error
(observational area) 1/2
1
(observational area) 1/2
1 ∝
Measurement error = Gaussian error + non-Gaussian error
depends on non-linearity of fluctuations (scale, redshift)
&“survey shape”
“What observational shape is the best to reduce the non-Gaussian error?”
Measurement error @ small scales
(Meiksin & White 1999; Scoccimarro+ 1999)
Small-scale fluctuations are correlated due to non-linear mode coupling
Log-normal Convergence field
Convergence field is assumed to follow log-normal distribution
: minimum convergence ← empty beam
: variance
the mean is zero
𝜅0
𝜎𝜅2
𝜅 0
𝑑𝑃
𝑑𝜅
𝜅0
which is supported by gravitational-lensing ray-tracing simulations (e.g. Jain+ 2000; Taruya+ 2002; Das & Ostriker 2006; RT+ 2011)
Log-normal Convergence Maps
・ 1000 maps prepared
・ each map is a square shape 203x203 deg (= 4π stradian)
・ input power spectrum at redshift 0.9 ← halo-fit nonlinear model
𝜃1
𝜃2
203deg
20
3d
eg
0
cut a region & calculate the power spectrum
・ angular resolution = 1arcmin
2
Log-normal Convergence Maps
We calculate the power spectra & covariance from the 1000 maps
Cumulative SN up to ℓmax
𝑆
𝑁
2= 𝐶𝑖𝑗
−1ℓ𝑖,ℓ𝑗<ℓmax
𝐶𝑊 ℓ𝑖 𝐶𝑊(ℓ𝑖)
SN(Signal-to-Noise ratio) in power spectrum measurement
𝐶𝑊(ℓ)
𝐶𝑖𝑗
: power spectrum
: covariance
Results
rectangular observational shape
𝜃𝐴
𝜃𝐵
𝜃𝐴 × 𝜃𝐵 = 100 deg2
with an area of 100deg^2
Cumulative SN for rectangular observational shapes
Survey area = 100deg^2
𝑧𝑠 = 0.9
Cumulative SN up to l_max
Maximum multipole
Cumulative SN for rectangular observational shapes
Survey area = 100deg^2
𝑧𝑠 = 0.9
Cumulative SN up to l_max
Maximum multipole
2 times better
3 times better
“What observational survey shape is the best (or worst) to suppress the non-Gaussian error in the power spectrum measurement”
Problem setting :
1. prepare 100 small patches, each patch has an area of 1x1 deg^2
2. place these 100 patches on the all sky (203x203 deg^2)
→ survey area is 100deg^2 in total
100 deg^2 area in total
Cumulative SN up to l_max
3 times better
3 times better
Maximum multipole
𝜎𝑊2 =
𝑑2𝑞
2𝜋 2𝑊 𝑞
2𝐶(𝑞)
=1
𝑆𝑊2 𝑑2𝜃1𝑑
2𝜃2𝑊 𝜃 1 𝑊(𝜃 2)𝜉( 𝜃 1 − 𝜃 2 )
Two-point correlation function
The non-Gaussian error depends on variance of convergence in the survey region:
Small non-Gaussian error Small variance 𝜎𝑊2
(Hilbert+ 2011; Takada & Hu 2013)
Two
-po
int
corr
ela
tio
n f
un
ctio
n o
f co
nve
rge
nce
ξ has a minimum at θ=15deg
Best separation among the patches is ~15deg
Separation angle
100 deg^2 area in total
Each patch is separated by ~15deg separation
Sparse sampling
1x1 deg^2 patches are placed in a configuration 10x10 with a separation θsep
Cumulative SN up to l_max
Maximum multipole
Cumulative SN up to l_max
Maximum multipole
more sparse distribution is better
・ larger-scale fluctuations can be measured @large scales
The above observational shapes have advantages at both large and small scales:
“Sparse Sampling”
“Belt” “Elongated Rectangular”
Summary
・ the non-Gaussian error can be suppressed @small scales
(e.g. Kaiser 1986, 1998; Kilbinger & Schneider 2004; Chiang+ 2013)
Power Spectrum Covariance for Log-normal Field
𝑊 𝜃 = 1 0
𝑆𝑊 = 𝑑2𝜃 𝑊(𝜃 )
Total survey area
𝐶𝑊(ℓ) =1
𝑆𝑊 𝑑2𝑞 𝑊 ℓ − 𝑞
2𝐶(𝑞)
Survey window function
Circular-averaged band power spectrum
𝐶 𝑊 ℓ =1
𝑆𝑊
𝑑2ℓ
𝑁ℓℓ ∈ℓ
𝑑2𝑞
2𝜋 2𝑊 ℓ − 𝑞
2𝐶(𝑞)
𝑁ℓ = 2𝜋ℓ∆ℓ : # of modes in (ℓ − ∆ℓ 𝟐 , ℓ + ∆ℓ 𝟐 )
inside the survey region otherwise
Power spectrum convolved with the window function
Power spectrum covariance
𝐶𝑖𝑗 = 𝐶 𝑊 ℓ𝑖 , 𝐶 𝑊 ℓ𝑗 =1
𝑆𝑊
2
𝑁ℓ𝐶 𝑊(ℓ𝑖)
2𝛿𝑖𝑗 + 𝑇𝑊(ℓ𝑖 , ℓ𝑗)
𝐶𝐺𝑖𝑗 =
1
𝑆𝑊
2
𝑁ℓ𝐶 𝑊(ℓ𝑖)
2𝛿𝑖𝑗
𝐶𝑇0𝑖𝑗 =
1
𝑆𝑊
𝑑2ℓ
𝑁ℓ𝑖ℓ ∈ℓ𝑖
𝑑2ℓ′
𝑁ℓ𝑗ℓ ∈ℓ𝑗
𝑇(ℓ, −ℓ, ℓ′, −ℓ′)
(Takada & Hu 2013)
𝐶𝜎𝑊𝑖𝑗 =
4
𝜅02 𝜎𝑊
2 𝐶 ℓ𝑖 𝐶(ℓ𝑗) 𝜎𝑊2 =
𝑑2𝑞
2𝜋 2 𝑊 𝑞 2𝐶(𝑞)
Variance of convergence in the survey region
= 𝐶𝐺𝑖𝑗+ 𝐶𝑇0
𝑖𝑗 + 𝐶𝜎𝑊𝑖𝑗
Gaussian non-Gaussian terms
∝ 1 / (survey area) depends on the survey geometry
(beat-coupling term)
The trispectrum can be written in term of power spectra for the log-normal field
Hilbert+ (2011) showed that a four point correlation function of log-normal field can be written in terms of its two point function
→ Power spectrum covariance can be written in terms of its power spectrum
Gaussian
→ three independent data
→ single data
Non-Gaussian Cls are fully correlated at different k
Non-Gaussian error depends on survey geometry
“Optimal” survey geometry can reduce the non-Gaussian errors
シミュレーションマップを使った解析
: # of realizations
𝐶 𝑊,𝑟 ℓ =1
𝑆𝑊
1
𝑁ℓ 𝜅 𝑊,𝑟(ℓ
′)2
ℓ′ ∈ℓ
Cylindrically-averaged power spectrum estimator
Mean power spectrum
𝐶𝑊 ℓ =1
𝑁𝑟 𝐶 𝑊,𝑟(ℓ)
𝑁𝑟
𝑟=1
𝐶𝑖𝑗 = 𝐶 𝑊 ℓ𝑖 , 𝐶 𝑊 ℓ𝑗 =1
𝑁𝑟 − 1 𝐶 𝑊,𝑟 ℓ𝑖 − 𝐶𝑊 ℓ𝑖 𝐶 𝑊,𝑟(ℓ𝑖) −𝐶𝑊 ℓ𝑗
𝑁𝑟
𝑟=1
Covariance matrix
𝑁𝑟 = 1000
Cumulative SN(Signal-to-Noise ratio) up to ℓmax
𝑆
𝑁
2= 𝐶𝑖𝑗
−1ℓ<ℓmax
𝐶𝑊 ℓ𝑖 𝐶𝑊(ℓ𝑖)
(Taruya+ 2002) Ray-tracing simulation
Covariance matrix (diagonal elements)
Multipole
Variance
Multipole
𝐶𝑖𝑗
𝐶𝑖𝑖𝐶𝑗𝑗
Covariance matrix (off-diagonal elements)