complete reionization constraints from planck 2015 ... · outline • intro: probing reionization...
TRANSCRIPT
Complete Reionization Constraints from Planck 2015 Polarization
Chen He Heinrich University of Chicago
Oct. 2016 UC Berkeley - Cosmology Seminar
Heinrich, Miranda & Hu arXiv:1609.04788
Outline• Intro: Probing reionization with CMB polarization
• Advantages of principal components (PCs)
• Probing high-z ionization with Planck 2015
• Fast model testing with our effective likelihood code
Image credit: https://en.wikipedia.org/wiki/Reionization
Recombination
Inflation
z ~1000
z ~10
z = 0
e� e�
CMB
Reionizatione�
• Astrophysical interest
• Cosmology: (optical depth) error propagates to other cosmological parameters
• leading source of error for neutrino mass from gravitational lensing
• growth of structure and cosmic acceleration [Hu & Jain 04]
Reionization
Image credit: https://en.wikipedia.org/wiki/Reionization
⌧
• Astrophysical interest
• Cosmology: (optical depth) error propagates to other cosmological parameters
• leading source of error for neutrino mass from gravitational lensing
• growth of structure and cosmic acceleration [Hu & Jain 04]
Reionization
Image credit: https://en.wikipedia.org/wiki/Reionization
Image Credit: http://www.staff.uni-mainz.de/wurmm /wurm-home/mass-hierarchy.png
⌧
Probes of reionization• Gunn-Peterson effect (in quasar spectra):
conclude that Universe is fully ionized by z=6.
• CMB anisotropies: signatures on the temperature and polarization power spectra.
• Galaxy luminosity function: well measured for z < 8, ~10s of galaxies at z > 9.
• 21cm experiments (underway): map the distribution of neutral hydrogen with redshift. (PAPER, LOFAR, MWA, MITEoR, HERA, SKA …)
Cosmic microwave background (CMB)
angular wavenumber
Image Credit: http://www.esa.int/spaceinimages/Images/2013/03/Planck_CMB
(mean = 2.7K)
temperature anisotropies
Fourier Transform
Temperature power spectrum
101 102 103
l
102
103
104
l(l+
1)C
TT
l/2
⇡[µ
K2]
Cosmic microwave background (CMB)
angular wavenumber
Image Credit: http://www.esa.int/spaceinimages/Images/2013/03/Planck_CMB
101 102 103
l
10�3
10�2
10�1
100
101
102
l(l+
1)C
EE
l/2
⇡[µ
K2]
E-mode polarization power spectrum
CMB photons Thomson scatter with free electrons
⌧ =
Z ⌘
0d⌘0�Tnea
�T
e� e�
e� + � ! e� + �
ne� �
Optical depth:
e�
⌧ =
Z ⌘
0d⌘0�Tnea
1.Suppress anisotropies (both temp. and pol.)
Thomson scattering optical depth
e�2⌧
angular wavenumber
101 102 103
l
102
103
104
l(l+
1)C
TT
l/2
⇡[µ
K2 ]
⌧
101 102 103
l
10�3
10�2
10�1
100
101
102
l(l+
1)C
EE
l/2
⇡[µ
K2 ]
Power spectrum suppressed as ,
temperature power spectrum
Effects of reionization on the CMB
angular wavenumber
E-mode polarization power spectrum
Effects of reionization on the CMB
Quadrupole Anisotropy
Thomson Scattering
e–
Linear Polarization
ε'
ε'
ε
Image Credit: Wayne Hu’s Tutorials
Temperature quadrupole —> polarization
101 102 103
l
10�3
10�2
10�1
100
101
102
l(l+
1)C
EE
l/2
⇡[µ
K2]
E-mode polarization power spectrum
angular wavenumber
Reionization bump
2. Create polarization anisotropies on large scales
0.00
0.05
0.10
0.15
0.20
CMB measurements of ⌧
⌧
WMAP9Planck dust
cleaned
Planck
Planck
Planckw/ LFI
TT+ lensing
w/ HFIWMAP9
WMAP1TE
Outline• Intro: Probing reionization with CMB polarization
• Advantages of principal components (PCs)
• Probing high-z ionization with Planck 2015
• Fast model testing with our effective likelihood code
Standard approach: tanh
0.0
0.2
0.4
0.6
0.8
1.0
1.2
xe(
z)
tanh ML
5 10 15 20 25 30z
0.00
0.02
0.04
0.06
0.08
⌧(z
,zm
ax)
xe(z) =ne
nH
optical depth
simple tanh
ionization fraction
0.0
0.2
0.4
0.6
0.8
1.0
1.2
xe(
z)
tanh ML
fiducial
5 10 15 20 25 30z
0.00
0.02
0.04
0.06
0.08
⌧(z
,zm
ax)
x
fide = 0.15
Our model independent approach: +
X
a
maSa(z)xe(z) = x
fide
optical depth
ionization fraction
(Hu & Holder 03)
Eigenfunctions ranked by contribution to observables
Contribution to from each eigenfunction
~ high vs low z
~ weighted
optical depth
ionization fraction
a = 1
a = 2
⌧
⌧
�2�1
01234
Sa(z
)a = 1
a = 2
a = 3
a = 4
a = 5
10 15 20 25 30z
�0.1
0.0
0.1
0.2
0.3
⌧(z
,zm
ax)
5 10 15 20 25 30 35 40
l
0.00
0.01
0.02
0.03
0.04
0.05
0.06
0.07
0.08l(
l+
1)C
EE
l/2
⇡[µ
K2]
tanh ML
1 PC
5 PCs completely describe E-mode power spectrum
5 10 15 20 25 30 35 40
l
0.00
0.01
0.02
0.03
0.04
0.05
0.06
0.07
0.08l(
l+
1)C
EE
l/2
⇡[µ
K2]
tanh ML
2 PCs
5 PCs completely describe E-mode power spectrum
5 10 15 20 25 30 35 40
l
0.00
0.01
0.02
0.03
0.04
0.05
0.06
0.07
0.08l(
l+
1)C
EE
l/2
⇡[µ
K2]
tanh ML
3 PCs
5 PCs completely describe E-mode power spectrum
5 10 15 20 25 30 35 40
l
0.00
0.01
0.02
0.03
0.04
0.05
0.06
0.07
0.08l(
l+
1)C
EE
l/2
⇡[µ
K2]
tanh ML
4 PCs
5 PCs completely describe E-mode power spectrum
5 10 15 20 25 30 35 40
l
0.00
0.01
0.02
0.03
0.04
0.05
0.06
0.07
0.08l(
l+
1)C
EE
l/2
⇡[µ
K2]
tanh ML
5 PCs
reionization bump matches to cosmic variance precision!
5 PCs completely describe E-mode power spectrum
0.0
0.2
0.4
0.6
0.8
1.0
1.2
1.4
xe(
z)
tanh ML
tanh ML PC
5 10 15 20 25 30z
0.00
0.02
0.04
0.06
0.08
⌧(z
,zm
ax)
Project ionization fraction model onto PCs
5PCs designed for observables, not model reconstruction (remove fiducial)
not well constrained
by data
data reflects better this integrated quantity
ionization fraction
optical depth
Outline• Intro: Probing reionization with CMB polarization
• Advantages of principal components (PCs)
• Probing high-z ionization with Planck 2015
• Fast model testing with our effective likelihood code
⌦bh2,⌦ch
2, ✓MCMC, As, ns
⌧
m1 ,…, m5 LCDM (modify CAMB)PCs
tanh
+(corresponds to step time)
sampled with COSMOMC 10 parameters
Method: Apply MCMC to Planck 2015
Constraints on 5PC
parameters• Constraints on the
PC amplitude in the space of physical models (edge of box)
• First two PCs best constrained, third to fifth still constrained!
• tanh trajectory not favored by data
tanh ML
tanh 68, 95% CL
PC 68, 95% CL
0 0.3
-0.3
0
0.3
-0.3
0
0.3
-0.3
0
0.3
-0.6
-0.3
0
0.6 -0.6 -0.3 0 -0.3 0 0.3 -0.3 0 0.3 -0.3 0 0.3
0 < xe < 1 + fHe
xe(z) = x
fide +
X
a
maSa(z)
CH, Miranda, Hu 2016
Tanh less favoured in PC space
• tanh ML ~2 away from PC mean
−0.1 0.0 0.1 0.2
m1
−0.4
−0.3
−0.2
−0.1
0.0
0.1
0.2
0.3
m2
Gaussian
unph
ysic
al
* PC ML vs tanh ML:
CH, Miranda, Hu 2016
�
2�log Like = 5.3
• shift is 1 up
• consequence on cosmo parameters (show as
• what is the origin of higher ?0.02 0.04 0.06 0.08 0.10 0.12 0.14
⌧ (0, zmax)
P(⌧
|dat
a)
tanh
PC
Model
PC
tanh
⌧(0, zmax
)
0.079± 0.017
0.092± 0.015⌧ � shifts by 1
CH, Miranda, Hu 2016
High redshift ionization shifts tau
0 5 10 15 20 25 30
z
−0.04
−0.02
0.00
0.02
0.04
0.06
0.08
0.10
0.12
0.14
τ(z
,zm
ax)
tanh ML PC
PC mean
PC 68, 95% CL
Planck 2015
• Standard tanh misses this by assumption of form
2 preference for ionization by z = 15
�
CH, Miranda, Hu 2016
Removing effects of lensing
0 5 10 15 20 25 30
z
−0.04
−0.02
0.00
0.02
0.04
0.06
0.08
0.10
0.12
0.14
τ(z
,zm
ax)
tanh ML PC
PC mean
PC 68, 95% CL
Planck 2015 lensing AL marginalized
High redshift ionization is not due to lensing effects
AL
⌧
As
More smoothing
Less initial power
Ase�2⌧
Fixed
goes down
Marginalize over AL
CH, Miranda, Hu 2016
still 2�
High z ionization only found in Planck
• Planck pol —> WMAP pol: preference dropped to 1 • High redshift ionization: origin is Planck polarization (LFI)
�
Data
P15
P13 +WMAP(P)
0.033± 0.016
0.022± 0.018
⌧(15, zmax
)
Extended ionization broadens bump
• PC ML: broader bump —> extended ionization to higher z. • E-mode polarization 8 < l < 20. Tanh fails to pick this out.
5 10 15 20 25 30 35 40l
0.00
0.01
0.02
0.03
0.04
0.05
0.06
0.07
0.08
l(l+
1)C
EE
l/2
π[µ
K2]
tanh ML
tanh ML PC
PC ML
High redshift ionization
CH, Miranda, Hu 2016
Sources
• Foregrounds or systematics (Planck 2017 will clarify)
• Possible high redshift reionization sources:
• DM annihilation
• POP II + POP III stars (metal-poor stars), etc …
• Running MCMC vs PC effective likelihood
Outline• Intro: Probing reionization with CMB polarization
• Advantages of principal components (PCs)
• Probing high-z ionization with Planck 2015
• Fast model testing with our effective likelihood code
Effective Likelihood Code
• Easily tests any models of ionization model xe(z) between 6 < z < 30.
• Projection unto PCs:
• Kernel density estimate:
Gaussian kernel (zero mean, covariance a fraction f of the chain covariance)
P (⌧ |data) / LPC [data|m(⌧)]P (⌧)
p ! xe(z) ! m
LPC (data|m) =NX
i=1
wiKf (m�mi)
Example: tanh
• Cutoff due to full ionization by z = 6
• f = 0.14 smoothing suffices for tanh (should work better for models favoured by data)
0.03 0.06 0.09 0.12τ (tanh model)
P(τ|d
ata)
tanhtanh LPC
Gaussian
P (⌧ |data) / LPC[data|m(⌧)]P (⌧)
5min vs 24 hours!
⌧ ! xe(z) ! {ma} ! LPC
CH+16, submitted to PRD
High z ionization: Pop-III stars?
Type of star Metal Content Cooling Host halos
Pop-III Metal-free Molecular hydrogen Minihalos
Pop-II Metal-poor Atomic line emission More massive halos
105 � 106M�
Tanh
Pop-II
Pop-III
Pop-III, self-regulated
Ionization fraction
Miranda, Lidz, CH, Hu 2016
Regularization mechanism needed
plateau from regularization
Tanh
P15 68%, 95% CL
Pop-II
Pop-III
Pop-III, self-regulated
regularization mechanism needed
Miranda, Lidz, CH, Hu 2016
Pop-III, self-regulated
Miranda, Lidz, CH, Hu 2016
Conclusion• We probed reionization using CMB polarization data
• With the principal component analysis, we can extract all information available in the observable
• Planck 2015 polarization data allows us to constrain an additional mode: high redshift polarization.
• Optical depth shifts by 1 (compared to tanh)
• z >15 optical depth preferred at ~ 2
• Use PC analysis!
�
�
Effective Likelihood Code
• Use our effective likelihood code for efficient and unbiased testing of any ionization history models.(tanh: 5min vs 24 hours MCMC ).
• When applied on Planck 2017 polarization data — better constraints on high redshift ionization component.
(Code available on request)
Thank you!
Back-up slides
⌧ = 0.089± 0.014 (WMAP9)
⌧ = 0.075± 0.013 (WMAP9 dust cleaned with Planck 353)
⌧ = 0.078± 0.019 (Planck LFI low l + high l TT )
⌧ = 0.070± 0.024 (Planck TT + lensing)
⌧ = 0.17± 0.04 (WMAP1, TE)
⌧ = 0.055± 0.009 (Planck HFI low l)
⌧CMB measurements of
• shifts up by 1
• consequence on other cosmo parameters
0.02 0.04 0.06 0.08 0.10 0.12 0.14
⌧ (0, zmax)
P(⌧
|dat
a)
tanh
PC
65 66 67 68 69 70 71
H0
0.78
0.80
0.82
0.84
0.86
0.88
�8
PC
tanh
Model
PC
tanh
⌧(0, zmax
)
0.079± 0.017
0.092± 0.015
⌧ �
Parameter shifts
End of Back-up slides