The Cosmic Microwave Background in High
Definition
Gil Holder
as part of:
SPT collaboration
Outline
• the cosmic microwave background (CMB) – temperature & polarization fluctuations
• Sunyaev-Zeldovich effect – galaxy clusters
• CMB gravitational lensing – chasing neutrino masses
• first detection of “B-modes”
2
• Expanding => cooling
• At earlier times, the universe was hotter
• when atoms formed, universe became
transparent to photons
– special timescale in the universe for photons
Hot Big Bang
http://www.astro.ucla.edu/~wright/BBhistory.html
The Cosmic Microwave Background
CMB according to COBE
(Bennett et al 1996)
Image from COBE science team: http://lambda.gsfc.nasa.gov/product/cobe/
Isotropy
• Cosmic
microwave background is
remarkably isotropic
• Unnaturally
isotropic! +-3.5 mK scale
WMAP science team
The Cosmic Microwave Background
CMB according to COBE
(Bennett et al 1996)
Image from COBE science team: http://lambda.gsfc.nasa.gov/product/cobe/
Nothing too strange within our “horizon”:
40 billion light years
Image from WMAP
8
Planck has higher resolution than WMAP
16 o
9WMAP 60 GHz Planck 143 GHz
South Pole Telescope
10 m mm-wave (3 different wavelengths) telescope
at the south pole
• extremely dry
• very stable
• good support
photo by Keith Vanderlinde
photo by Dana Hrubes
Chicago Colorado UC Berkeley Case Western
McGill Harvard UC Davis Munich +++
The evolution of SPT cameras
2007-2011: SPT 960 detectors
2012-2015: SPTpol
~1600 detectors 2016: SPT-3G
~15,200 detectors
Now with polarization!
slide from S Hoover
12
SPT has higher resolution than Planck
Planck 143 GHz Planck+SPT
4 o
CMB Angular Power Spectrum
CMB Polarization
• CMB fluctuations are relatively strongly
polarized (~10%)
14Q
U
E-modes/B-modes
• E-modes vary spatially parallel or perpedicular to polarization direction
• B-modes vary spatially at 45 degrees
• CMB
• scalar perturbations only generate *only* E
E modes
(a) (b)
Figure 1. A pure E Fourier mode (a), and a pure B mode (b).
example, consider the original COBE detection: although the key science was contained in the two-point correlation function and power spectrum estimates, the actual real-space maps were invaluable in convincing the world of the validity and importance of the results.)
Consideration of issues related to E/B separation is important in experiment design and optimization as well. For example, the ambiguity in E/B separation significantly alters the optimal tradeoff between sky coverage and noise per pixel in a degree-scale B mode experiment [6].
2. Pure and ambiguous modes
The E/B decomposition is easiest to understand in Fourier space. For any given wavevector k, define a coordinate system (x, y) with the x axis parallel to k, and compute the Stokes parameters Q, U . An E mode contains only Q, while a B mode contains only U . In other words, in an E mode, the polarization direction is always parallel or perpendicular to the wavevector, while in a B mode it always makes a 45
◦angle, as shown in Figure 1.
In a map that covers a finite portion of the sky, of course, the Fourier transform cannot be determined with infinite k-space resolution. According to the Heisenberg uncertainty principle, if the observed region has size L, an estimate of an individual Fourier mode with wavevector q will be a weighted average of true Fourier modes k in a region around q of width |k − q| ∼ L
−1. These Fourier modes will all point in slightly different directions, spanning a range of angles
∼ qL. Since the mapping between (Q, U) and (E, B) depends on the angle of the wavevector, we expect the amount of E/B mixing to be of order qL. In particular, this means that the largest scales probed by a given experiment will always have nearly complete E/B mixing. This is unfortunate, since the largest modes probed are generally the ones with highest signal-to-noise ratio. Typically, the noise variance is about the same in all Fourier modes detected by a given experiment, while the signal variance scales as C
l, which decreases as a function of wavenumber.
(Remember, even a “flat” power spectrum is one with l
2C
l∼ constant.)
One way to quantify the amount of information lost in a given experimental setup is to decompose the observed map into a set of orthogonal modes consisting of pure E modes, pure B modes, and ambiguous modes [7]. A pure E mode is orthogonal to all B modes, which means that any power detected in such a mode is guaranteed to come from the E power spectrum.
(a) (b)
Figure 1. A pure E Fourier mode (a), and a pure B mode (b).
example, consider the original COBE detection: although the key science was contained in the two-point correlation function and power spectrum estimates, the actual real-space maps were invaluable in convincing the world of the validity and importance of the results.)
Consideration of issues related to E/B separation is important in experiment design and optimization as well. For example, the ambiguity in E/B separation significantly alters the optimal tradeoff between sky coverage and noise per pixel in a degree-scale B mode experiment [6].
2. Pure and ambiguous modes
The E/B decomposition is easiest to understand in Fourier space. For any given wavevector k, define a coordinate system (x, y) with the x axis parallel to k, and compute the Stokes parameters Q, U . An E mode contains only Q, while a B mode contains only U . In other words, in an E mode, the polarization direction is always parallel or perpendicular to the wavevector, while in a B mode it always makes a 45
◦angle, as shown in Figure 1.
In a map that covers a finite portion of the sky, of course, the Fourier transform cannot be determined with infinite k-space resolution. According to the Heisenberg uncertainty principle, if the observed region has size L, an estimate of an individual Fourier mode with wavevector q will be a weighted average of true Fourier modes k in a region around q of width |k − q| ∼ L
−1. These Fourier modes will all point in slightly different directions, spanning a range of angles
∼ qL. Since the mapping between (Q, U) and (E, B) depends on the angle of the wavevector, we expect the amount of E/B mixing to be of order qL. In particular, this means that the largest scales probed by a given experiment will always have nearly complete E/B mixing. This is unfortunate, since the largest modes probed are generally the ones with highest signal-to-noise ratio. Typically, the noise variance is about the same in all Fourier modes detected by a given experiment, while the signal variance scales as C
l, which decreases as a function of wavenumber.
(Remember, even a “flat” power spectrum is one with l
2C
l∼ constant.)
One way to quantify the amount of information lost in a given experimental setup is to decompose the observed map into a set of orthogonal modes consisting of pure E modes, pure B modes, and ambiguous modes [7]. A pure E mode is orthogonal to all B modes, which means that any power detected in such a mode is guaranteed to come from the E power spectrum.
B modes
Bunn
E-modes/B-modes
• E-modes vary spatially parallel or perpedicular to polarization direction
• B-modes vary spatially at 45 degrees
• CMB
• scalar perturbations only generate *only* E
Image of positive kx/positive ky Fourier transform of a 10x10 deg chunk of
Stokes Q CMB map [simulated; nothing clever done to it]
E modes
• Lensing of CMB is
much more obvious in polarization!
k x
k y
E-mode polarization of ra23h30, dec -55 field (150 GHz)
E
CMB Polarization Angular Power Spectrum
Barkats et al (BICEP)
only
upper
limits
on B
mode
power
Two Expected Sources of B Modes
Gravitational Radiation in Early Universe
(amplitude unknown!) Gravitational lensing of
E modes (amplitude
well-predicted, but no
measured B modes
until later in talk)
E-modes/B-modes
• E-modes vary spatially parallel or perpedicular to polarization direction
• B-modes vary spatially at 45 degrees
• scalar perturbations only generate *only* E
E modes
(a) (b)
Figure 1. A pure E Fourier mode (a), and a pure B mode (b).
example, consider the original COBE detection: although the key science was contained in the two-point correlation function and power spectrum estimates, the actual real-space maps were invaluable in convincing the world of the validity and importance of the results.)
Consideration of issues related to E/B separation is important in experiment design and optimization as well. For example, the ambiguity in E/B separation significantly alters the optimal tradeoff between sky coverage and noise per pixel in a degree-scale B mode experiment [6].
2. Pure and ambiguous modes
The E/B decomposition is easiest to understand in Fourier space. For any given wavevector k, define a coordinate system (x, y) with the x axis parallel to k, and compute the Stokes parameters Q, U . An E mode contains only Q, while a B mode contains only U . In other words, in an E mode, the polarization direction is always parallel or perpendicular to the wavevector, while in a B mode it always makes a 45
◦angle, as shown in Figure 1.
In a map that covers a finite portion of the sky, of course, the Fourier transform cannot be determined with infinite k-space resolution. According to the Heisenberg uncertainty principle, if the observed region has size L, an estimate of an individual Fourier mode with wavevector q will be a weighted average of true Fourier modes k in a region around q of width |k − q| ∼ L
−1. These Fourier modes will all point in slightly different directions, spanning a range of angles
∼ qL. Since the mapping between (Q, U) and (E, B) depends on the angle of the wavevector, we expect the amount of E/B mixing to be of order qL. In particular, this means that the largest scales probed by a given experiment will always have nearly complete E/B mixing. This is unfortunate, since the largest modes probed are generally the ones with highest signal-to-noise ratio. Typically, the noise variance is about the same in all Fourier modes detected by a given experiment, while the signal variance scales as C
l, which decreases as a function of wavenumber.
(Remember, even a “flat” power spectrum is one with l
2C
l∼ constant.)
One way to quantify the amount of information lost in a given experimental setup is to decompose the observed map into a set of orthogonal modes consisting of pure E modes, pure B modes, and ambiguous modes [7]. A pure E mode is orthogonal to all B modes, which means that any power detected in such a mode is guaranteed to come from the E power spectrum.
(a) (b)
Figure 1. A pure E Fourier mode (a), and a pure B mode (b).
example, consider the original COBE detection: although the key science was contained in the two-point correlation function and power spectrum estimates, the actual real-space maps were invaluable in convincing the world of the validity and importance of the results.)
Consideration of issues related to E/B separation is important in experiment design and optimization as well. For example, the ambiguity in E/B separation significantly alters the optimal tradeoff between sky coverage and noise per pixel in a degree-scale B mode experiment [6].
2. Pure and ambiguous modes
The E/B decomposition is easiest to understand in Fourier space. For any given wavevector k, define a coordinate system (x, y) with the x axis parallel to k, and compute the Stokes parameters Q, U . An E mode contains only Q, while a B mode contains only U . In other words, in an E mode, the polarization direction is always parallel or perpendicular to the wavevector, while in a B mode it always makes a 45
◦angle, as shown in Figure 1.
In a map that covers a finite portion of the sky, of course, the Fourier transform cannot be determined with infinite k-space resolution. According to the Heisenberg uncertainty principle, if the observed region has size L, an estimate of an individual Fourier mode with wavevector q will be a weighted average of true Fourier modes k in a region around q of width |k − q| ∼ L
−1. These Fourier modes will all point in slightly different directions, spanning a range of angles
∼ qL. Since the mapping between (Q, U) and (E, B) depends on the angle of the wavevector, we expect the amount of E/B mixing to be of order qL. In particular, this means that the largest scales probed by a given experiment will always have nearly complete E/B mixing. This is unfortunate, since the largest modes probed are generally the ones with highest signal-to-noise ratio. Typically, the noise variance is about the same in all Fourier modes detected by a given experiment, while the signal variance scales as C
l, which decreases as a function of wavenumber.
(Remember, even a “flat” power spectrum is one with l
2C
l∼ constant.)
One way to quantify the amount of information lost in a given experimental setup is to decompose the observed map into a set of orthogonal modes consisting of pure E modes, pure B modes, and ambiguous modes [7]. A pure E mode is orthogonal to all B modes, which means that any power detected in such a mode is guaranteed to come from the E power spectrum.
B modes
Bunn
Cosmic Microwave Background
• acoustic scale (in cm) set by physics
unrelated to dark energy
– angular scale depends on
expansion history
• provides
normalization of
fluctuation amplitude
at z~1100
23WMAP (all sky)
South Pole Telescope (total 2500 sq deg)
8
oSPT-SZ Survey (completed)
Final survey depths of:
- 100 GHz: < 40 uKCMB-arcmin - 150 GHz: < 18 uKCMB-arcmin - 220 GHz: < 80 uKCMB-arcmin
2500 square degrees
Gravity at work
simulations carried out by the Virgo Supercomputing Consortium using computers based at Computing Centre of the Max-Planck Society in Garching and at the Edinburgh Parallel Computing Centre. The data are publicly available at www.mpa-garching.mpg.de/NumCos
t=400 000 yrs t=20 million yrs t=500 million yrs t=13.7 billion yrs
simulated density contrast at different times
1 billion light years
Zoom in on 2 mm map
~ 4 deg 2 of SPT data
18 uK sensitivity at observing wavelength of 2 mm roughly same resolution as your eye
Image by Will High in recent paper by Williamson et al
One of the heaviest objects in the universe
>10 15 solar masses
Fig. 16.— SPT-CL J0438-5419, also known as ACT-CL J0438-5419, at zrs = 0.45. Blanco/MOSAIC-II irg images are shown in the optical/infrared panel.
Fig. 17.— SPT-CL J0549-6204 at zrs = 0.32. Blanco/MOSAIC-II irg images are shown in the optical/infrared panel.
patch of
isolated
cosmic fog
Thermal Sunyaev-Zel’dovich Effect
CMB Hot
electrons
CMB+ ν
I
Optical depth: τ ~ 0.01
Fractional energy gain per scatter: ~ 0.01
Typical cluster signal: ~500 uK
SPT Cluster Images
0658-5556 (z=0.30) (Bullet)
2344-4243 (z=0.62)
2106-5844 (z=1.13) 2337-5942 (z=0.78)
SZ
IR-Optical
12 ’ 6 ’
Williamson et al 2011, arXiv:1101.1290
SPT Cluster Sample Properties
• Optically confirmed >300 clusters, ~80% newly discovered
• High redshift: < z> = 0.55 and ~20-25% of clusters at z > 0.8
• Optical measurements also confirm ~95% purity at S/N = 5
• Mass threshold flat/falling w/ redshift: M
500(z=0.6) > ~3x10
14M
sol/h
70Redshift Histogram SZ Mass vs Redshift
! "#
$%&&'()*!+,!-./0!12%&/3/3'4!5%&!-'!$'%42('67!/0'!8'(%53/9!.,!/034!(':%/3.&!5%&!-'!50'5;'67!
permitting a test of the underlying GR theory*!
!
<3=2('!+>?@!3::24/(%/'4!/0'!',,'5/!.,!6%(;!'&'(=9!.&!/0'!634/%&5'?('6403,/!%&6!=(.A/0?
('6403,/!(':%/3.&47!03=0:3=0/3&=!/0'!&''6!,.(!B'(5'&/?:'8':!B('5343.&!3&!/0'4'!12%&/3/3'4!3,!
A'!%('!/.!5.&4/(%3&!/0'!6%(;?'&'(=9!'12%/3.&!.,!4/%/'!/.!%-.2/!C*#!%552(%59*!
!
!!! !
Fig. VI-2: The primary observables for dark-energy – the distance-redshift relation DDz)!
and the growth-redshift relation gDz) – are plotted vs. redshift for three cosmological models. The green curve is an open-Universe model with no dark energy at all. The black curve is the “concordance” / CDM model, which is flat and has a cosmological constant, i.e., w . This model is consistent with all reliable present-day data. The red curve is a dark-energy model with w , for which other parameters have been adjusted to match WMAP data. At left one sees that dark-energy models are easily
distinguished from non-dark-energy models. At right, we plot the ratios of each model to the / CDM model, and it is apparent that distinguishing the w model from / CDM requires percent-level precision on the diagnostic quantities.!
Four Astrophysical Approaches to Dark Energy Measurements
1. Type Ia Supernovae
E9B'!+%!42B'(&.8%'!%('!-':3'8'6!/.!-'!/0'!'FB:.438'!6343&/'=(%/3.&4!.,!A03/'?6A%(,!4/%(4!
/0%/!%55('/'!$%/'(3%:!/.!'F5''6!/0'!4/%-3:3/9!:3$3/!.,!#*G!4.:%(!$%44'4!6'(38'6!-9!
H0%&6(%4';0%(*!!!I'5%24'!/0'!$%44'4!.,!/0'4'!.-J'5/4!%('!&'%(:9!%::!/0'!4%$'7!/0'3(!
'FB:.43.&4!%('!'FB'5/'6!/.!4'(8'!%4!4/%&6%(6!5%&6:'4!.,!;&.A&!:2$3&.43/9!L7!3&!A0350!
5%4'!/0'!(':%/3.&!f = LKGSd
L!5%&!-'!24'6!/.!3&,'(!/0'!:2$3&.43/9!634/%&5'!d
L*!!LB'5/(%:!
:3&'4!3&!/0'!42B'(&.8%!:3=0/!$%9!-'!24'6!/.!36'&/3,9!/0'!('6403,/7!%4!5%&!4B'5/(%:!,'%/2('4!
.,!/0'!=%:%F9!0.4/3&=!/0'!'FB:.43.&*!!
!
Characterizing Dark Energy
from Dark Energy Task Force report w=-0.9
w=-1
33 10
Fig. 4.— Constraints on the Ωm, σ8, and w parameters using the SPT and CMB+BAO+SNe data sets assuming a wCDM cosmology.
Contours show the 68% and 95% confidence regions for the CMB+BAO+SNe (gray, dashed), and CMB+BAO+SNe+SPT (green, solid) data sets. The SPT data most significantly improves the constraints on σ8 and w, improving the constraints on each, by factors of 1.4 and 1.3, respectively.
Figure 4, we show the constraints of the combined CMB+BAO+SNe data set, before and after including the SPT cluster data. The SPT data most significantly improves the constraints on σ8 and w; reducing their 2-d likelihood area by a factor of ∼1.8. In Table 6, we give the marginalized constraints for several param- eters before and after the inclusion of the SPT data.
The combined constraints are w = −0.973 ± 0.063 and σ8 = 0.793 ± 0.028, a factor of 1.3 and 1.4 improvement, respectively, over the constraints without clusters. The combined data set also constrains Ωm = 0.273 ± 0.015 and h = 0.697 ± 0.018. These constraints are consistent with previous cluster-based results from Vikhlinin et al.
(2009b); Mantz et al. (2010c); Rozo et al. (2010), which used X-ray and optical selected samples of typically lower redshift clusters. The sensitivity of the SPT cluster data to the amplitude of structure, or σ8, is primarily what gives it the ability to break degeneracies with the distance-relation based constraints from the BAO and SNe data sets. We note the slight tension with the H0 constraints from (Riess et al. 2011) of h = 0.738 ± 0.024.
While this tension is not significant, it helps to intuitively
explain some constraints on neutrino mass in Section 5.2, so we make note of it here.
5.1.2. ζ− M500 Constraints
Given the work of V10 and other cluster results (e.g.
Vikhlinin et al. 2009b; Mantz et al. 2010c; Rozo et al.
2010), we expect the cluster mass-calibration to be the dominant systematic uncertainty limiting our results. In Figure 5, we show the constraints on ASZ and σ8. The SPT cluster data has a significant degeneracy between the constraints on ASZ and σ8. From the SPT data, we constrain the fractional uncertainty, δASZ/ASZ, to be 27%, which is effectively only constrained by the 14 clusters that have both X-ray and SZ measurements.
This constraint is not significantly better than the un- certainty in the simulation based prior of 30%. With enough X-ray observations, we expect the ζ − M500 cal- ibration to track the YX − M500 relation, because the latter is currently better observationally constrained. In this limit, we would expect a fractional uncertainty on ASZ of BSZ(δAX/AX) ∼ 14%. In Section 4, for a ΛCDM cosmology, we nearly achieved this limit, constraining
Constraints on Dark Energy
w=−0.973 ± 0.063
SPT-clusters: −1.087 ± 0.363
Benson et al 2011
Gravitational deflection
• CMB is a unique source for lensing
• Gaussian, with well-understood power spectrum (contains all info)
• At redshift which is (a) unique, (b) known, and (c) highest
T L (ˆn) = T U (ˆn + ∇φ(ˆn))
CMB Lensing
∇φ(ˆn) = −2
� χ
�0
dχ χ � − χ
χ � χ ∇ ⊥ Φ(χˆn, χ),
Broad kernel, peaks at z ~ 2 In WL limit, add many
deflections along line of sight
Photons get shifted n ˆ
T
n + ∇φ ˆ
Lensing simplified
• gravitational potentials
distort shapes by stretching, squeezing,
shearing
Gravity
Lensing simplified
• gravitational potentials
distort shapes by stretching, squeezing,
shearing
Gravity
Lensing simplified
• where gravity
stretches, gradients become smaller
• where gravity compresses,
gradients are larger
• shear changes direction
Gravity
• We extract ϕ by taking a suitable average over CMB multipoles
separated by a distance L
• We use the standard Hu quadratic estimator.
Mode Coupling from Lensing
T L (ˆn) = T U (ˆn + ∇φ(ˆn))
= T U (ˆn) + ∇T U (ˆn) · ∇φ(ˆn) + O(φ 2 ),
• Non-gaussian mode coupling for l 1 �= −l 2 :
l x
l y
L
l CMB1
l CMB2
SPT Lensing Mass Map
+-0.05 color bar
(noise ~0.01)
Planck
(all-sky)
SPT
(2500 sq deg)
CMB Lensing Power Spectrum
• well
measured with
Planck, SPT, ACT
Planck XVII 2013
Massive Neutrinos in Cosmology
–Below free-streaming scale, neutrinos act like radiation
• drag on growth
–Above free-streaming scale, neutrinos act like matter
€
Ω ν ≈ (m i
i
∑ /0.1 eV ) 0.0022 h 0.7 −2
Neutrinos & CMB Lensing
44
Neutrino masses
• Perturbations are washed out on
scales smaller than neutrino free-
streaming scale
• current upper bounds from CMB are WMAP: m nu < 1.3 eV ; WMAP+BAO+H 0 : m nu <
0.56 eV
d ∼ T ν /m ν × 1/H
Neutrino masses
• Perturbations are washed out on
scales smaller than neutrino free-
streaming scale
• current upper bounds from CMB are WMAP: m nu < 1.3 eV ; WMAP+BAO+H 0 : m nu <
0.56 eV
d ∼ T ν /m ν × 1/H
• Peak at l=40 (k eq =[300 Mpc] -1 at z = 2): coherent over degree scales
• RMS deflection angle is only ~2.7’
Upper limits on neutrino masses
• CMB experiments closing in on
interesting
neutrino mass range
• CMB lensing adds new information
– forecast ~0.05 eV sensitivity in ~4
yrs Planck collaboration 2013
45Not everything makes total sense
• combining all cosmological
information leads to a preference for a high neutrino
mass and some form of new light particle in the
universe
Wyman et al 2013
SPT Lensing Mass Map
+-0.05 color bar
(noise ~0.01)
Cosmic Infrared
Background Traces Mass
48 SPT TT Lensing map 100 sq deg Herschel 500 um
Two Expected Sources of B Modes
Gravitational Radiation in Early Universe
(amplitude unknown!) Gravitational lensing of
E modes (amplitude
well-predicted, but no
measured B modes
until later in talk)
E-mode polarization of ra23h30, dec -55 field (150 GHz)
E
Predicting B-Modes
3
FIG. 1: (Left panel): Wiener-filtered E-mode polarization measured by SPTpol at 150 GHz. (Center panel): Wiener-filtered CMB lensing potential inferred from CIB fluctuations measured by Herschel at 500 µm. (Right panel): Gravitational lensing B-mode estimate synthesized using Eq. (1). The lower left corner of each panel indicates the blue(-)/red(+) color scale.
[23] onboard the Herschel space observatory [24] as a tracer of the CMB lensing potential φ. The CIB has been established as a well-matched tracer of the lens- ing potential [22, 25, 26] and currently provides a higher signal-to-noise estimate of φ than is available with CMB lens reconstruction. Its use in cross-correlation with the SPTpol data also makes our measurement less sensitive to instrumental systematic effects [27]. We focus on the Herschel 500 µm map, which has the best overlap with the CMB lensing kernel [22].
Post-Map Analysis: We obtain Fourier-domain CMB temperature and polarization modes using a Wiener filter (e.g. [28] and refs. therein), derived by maximizing the likelihood of the observed I, Q, and U maps as a function of the fields T (�l), E(�l), and B(�l). The filter simultaneously deconvolves the two- dimensional transfer function due to beam, TOD filter- ing, and map pixelization while down-weighting modes that are “noisy” due to either atmospheric fluctuations, extragalactic foreground power, or instrumental noise.
We place a prior on the CMB auto-spectra, using the best-fit cosmological model given by [29]. We use a sim- ple model for the extragalactic foreground power in tem- perature [19]. We use jackknife difference maps to deter- mine a combined atmosphere+instrument noise model, following [30]. We set the noise level to infinity for any pixels within 5� of sources detected at > 5σ in [31]. We extend this mask to 10� for all sources with flux greater than 50 mJy, as well as galaxy clusters detected using the Sunyaev-Zel’dovich effect in [32]. These cuts remove approximately 5 deg2 of the total 100 deg2 survey area.
We remove spatial modes close to the scan direction with an �x < 400 cut, as well as all modes with l > 3000. For these cuts, our estimated beam and filter map transfer functions are within 20% of unity for every unmasked mode (and accounted for in our analysis in any case).
The Wiener filter naturally separates E and B con-
tributions, although in principle this separation depends on the priors placed on their power spectra. To check that we have successfully separated E and B, we also form a simpler estimate using the χB formalism advo- cated in [33]. This uses numerical derivatives to estimate a field χB(�x) which is proportional to B in harmonic space. This approach cleanly separates E and B, al- though it can be somewhat noisier due to mode-mixing induced by point source masking. We therefore do not mask point sources when applying the χB estimator.
We obtain Wiener-filtered estimates ˆφCIB of the lensing potential from the Herschel 500 µm maps by applying an apodized mask, Fourier transforming, and then multiply- ing by CCIB-φ
l (CCIB-CIB
l Clφφ)−1. We limit our analysis to modes l ≥ 150 of the CIB maps. We model the power spectrum of the CIB following [34], with CCIB-CIB
l =
3500(l/3000)−1.25Jy2/sr. We model the cross-spectrum CCIB-φ
l between the CIB fluctuations and the lensing po- tential using the SSED model of [35], which places the peak of the CIB emissivity at redshift zc = 2 with a broad redshift kernel of width σz = 2. We choose a linear bias parameter for this model to agree with the results of [22, 26]. More realistic multi-frequency CIB models are available (for example, [36]); however, we only require a reasonable template. The detection significance is inde- pendent of errors in the amplitude of the assumed CIB-φ correlation.
Results: In Fig. 1, we plot Wiener-filtered estimates Eˆ150 and ˆφCIB using the CMB measured by SPTpol at 150 GHz and the CIB fluctuations traced by Herschel. In addition, we plot our estimate of the lensing B modes, Bˆlens, obtained by applying Eq. (1) to these measure- ments. In Fig. 2 we show the cross-spectrum between this lensing B-mode estimate and the B modes mea- sured directly by SPTpol. The data points are a good fit to the expected cross-correlation, with a χ2/dof of 3.5/4 and a corresponding probability-to-exceed (PTE) of 48%.
measured E modes estimated predicted B
Hanson, Hoover, Crites et al 2013
Many Ways of Predicting B-Modes
3
FIG. 1: (Left panel): Wiener-filtered E-mode polarization measured by SPTpol at 150 GHz. (Center panel): Wiener-filtered CMB lensing potential inferred from CIB fluctuations measured by Herschel at 500 µm. (Right panel): Gravitational lensing B-mode estimate synthesized using Eq. (1). The lower left corner of each panel indicates the blue(-)/red(+) color scale.
[23] onboard the Herschel space observatory [24] as a tracer of the CMB lensing potential φ. The CIB has been established as a well-matched tracer of the lens- ing potential [22, 25, 26] and currently provides a higher signal-to-noise estimate of φ than is available with CMB lens reconstruction. Its use in cross-correlation with the SPTpol data also makes our measurement less sensitive to instrumental systematic effects [27]. We focus on the Herschel 500 µm map, which has the best overlap with the CMB lensing kernel [22].
Post-Map Analysis: We obtain Fourier-domain CMB temperature and polarization modes using a Wiener filter (e.g. [28] and refs. therein), derived by maximizing the likelihood of the observed I, Q, and U maps as a function of the fields T (�l), E(�l), and B(�l). The filter simultaneously deconvolves the two- dimensional transfer function due to beam, TOD filter- ing, and map pixelization while down-weighting modes that are “noisy” due to either atmospheric fluctuations, extragalactic foreground power, or instrumental noise.
We place a prior on the CMB auto-spectra, using the best-fit cosmological model given by [29]. We use a sim- ple model for the extragalactic foreground power in tem- perature [19]. We use jackknife difference maps to deter- mine a combined atmosphere+instrument noise model, following [30]. We set the noise level to infinity for any pixels within 5� of sources detected at > 5σ in [31]. We extend this mask to 10� for all sources with flux greater than 50 mJy, as well as galaxy clusters detected using the Sunyaev-Zel’dovich effect in [32]. These cuts remove approximately 5 deg2 of the total 100 deg2 survey area.
We remove spatial modes close to the scan direction with an �x < 400 cut, as well as all modes with l > 3000. For these cuts, our estimated beam and filter map transfer functions are within 20% of unity for every unmasked mode (and accounted for in our analysis in any case).
The Wiener filter naturally separates E and B con-
tributions, although in principle this separation depends on the priors placed on their power spectra. To check that we have successfully separated E and B, we also form a simpler estimate using the χB formalism advo- cated in [33]. This uses numerical derivatives to estimate a field χB(�x) which is proportional to B in harmonic space. This approach cleanly separates E and B, al- though it can be somewhat noisier due to mode-mixing induced by point source masking. We therefore do not mask point sources when applying the χB estimator.
We obtain Wiener-filtered estimates ˆφCIB of the lensing potential from the Herschel 500 µm maps by applying an apodized mask, Fourier transforming, and then multiply- ing by CCIB-φ
l (CCIB-CIB
l Clφφ)−1. We limit our analysis to modes l ≥ 150 of the CIB maps. We model the power spectrum of the CIB following [34], with CCIB-CIB
l =
3500(l/3000)−1.25Jy2/sr. We model the cross-spectrum CCIB-φ
l between the CIB fluctuations and the lensing po- tential using the SSED model of [35], which places the peak of the CIB emissivity at redshift zc = 2 with a broad redshift kernel of width σz = 2. We choose a linear bias parameter for this model to agree with the results of [22, 26]. More realistic multi-frequency CIB models are available (for example, [36]); however, we only require a reasonable template. The detection significance is inde- pendent of errors in the amplitude of the assumed CIB-φ correlation.
Results: In Fig. 1, we plot Wiener-filtered estimates Eˆ150 and ˆφCIB using the CMB measured by SPTpol at 150 GHz and the CIB fluctuations traced by Herschel. In addition, we plot our estimate of the lensing B modes, Bˆlens, obtained by applying Eq. (1) to these measure- ments. In Fig. 2 we show the cross-spectrum between this lensing B-mode estimate and the B modes mea- sured directly by SPTpol. The data points are a good fit to the expected cross-correlation, with a χ2/dof of 3.5/4 and a corresponding probability-to-exceed (PTE) of 48%.
measured E modes estimated predicted B
E 150 GHz E 90 GHz
E from Temperature
CIB
TT
EE
TE
( Spitzer cat)
X B 150 GHz X B 90 GHz
Hanson, Hoover, Crites et al 2013
4
FIG. 2: (Black, center bars): Cross-correlation of the lens- ing B modes measured by SPTpol at 150 GHz with lensing B modes inferred from CIB fluctuations measured by Herschel and E modes measured by SPTpol at 150 GHz; as shown in Fig. 1. (Green, left-offset bars): Same as black, but using E modes measured at 95 GHz, testing both foreground contam- ination and instrumental systematics. (Orange, right-offset bars): Same as black, but with B modes obtained using the χB procedure described in the text rather than our fiducial Wiener filter. (Gray bars): Curl-mode null test as described in the text. (Dashed black curve): Lensing B-mode power spectrum in the fiducial cosmological model.
We determine the uncertainty and normalization of the cross-spectrum estimate using an ensemble of simulated, lensed CMB+noise maps and simulated Herschel maps.
We obtain comparable uncertainties if we replace any of the three fields involved in this procedure with observed data rather than a simulation, and the normalization we determine for each bin is within 15% of an analytical prediction based on approximating the Wiener filtering procedure as diagonal in Fourier space.
In addition to the cross-correlation Eφ×B, it is also interesting to take a “lensing perspective” and rear- range the fields to measure the correlation EB ×φ. In this approach, we perform a quadratic “EB” lens re- construction [13] to estimate the lensing potential ˆφEB, which we then cross-correlate with CIB fluctuations. The observed cross-spectrum can be compared to previous temperature-based lens reconstruction results [22, 26].
This cross-correlation is plotted in Fig. 3. Again, the shape of the cross-correlation which we observe is in good agreement with the fiducial model, with a χ2/dof of 2.2/4 and a PTE of 70%.
Both the Eφ×B and EB ×φ cross-spectra discussed above are probing the three-point correlation function (or bispectrum) between E, B, and φ that is induced by lensing. We assess the overall significance of the measure- ment by constructing a minimum-variance estimator for the amplitude ˆA of this bispectrum, normalized to have
FIG. 3: “Lensing view” of the EBφ correlation plotted in Fig. 2, in which we cross-correlate an EB lens reconstruc- tion from SPTpol data with CIB intensity fluctuations mea- sured by Herschel. Left green, center black, and right or- ange bars are as described in Fig. 2. Previous analyses using temperature-based lens reconstruction from Planck [26] and SPT-SZ [22] are shown with boxes. The results of [26] are at a nominal wavelength of 550 µm, which we scale to 500 µm with a factor of 1.22 [37]. The dashed black curve gives our fiducial model for CCIB-φ
l as described in the text.
a value of unity for the fiducial cosmology+CIB model (analogous to the analyses of [38, 39] for the T T φ bis- pectrum). This estimator can be written as a weighted sum over either of the two cross-spectra already dis- cussed. Use of ˆA removes an arbitrary choice between the “lensing” or “B-mode” perspectives, as both are sim- ply collapsed faces of the EBφ bispectrum. Relative to our fiducial model, we measure a bispectrum amplitude A = 1.092ˆ ± 0.141, non-zero at approximately 7.7σ.
We have tested that this result is insensitive to analy- sis choices. Replacement of the B modes obtained using the baseline Wiener filter with those determined using the χB estimator causes a shift of 0.2σ. Our standard B-mode estimate incorporates a mask to exclude bright point sources, while the χB estimate does not. The good agreement between them indicates the insensitivity of po- larization lensing measurements to point-source contam- ination. If we change the scan direction cut from lx< 400 to 200 or 600, the measured amplitude shifts are less than 1.2σ, consistent with the root-mean-squared (RMS) shifts seen in simulations. If we repeat the analysis with- out correcting for I → Q, U leakage, the measured ampli- tude shifts by less than 0.1σ. A similar shift is found if we rotate the map polarization vectors by one degree to mimic an error in the average PSB angle.
We have produced estimates of ˆBlens using alterna- tive estimators of E. When we replace the E modes measured at 150 GHz with those measured at 95 GHz, we measure an amplitude ˆA = 1.225± 0.164, indicating