The lensing convergence measurable with future CMB experiments will be highly correlated with the clustering of galaxies that will be observed by deep imaging surveys such as LSST. I will discuss prospects for using that cross-correlation signal to constrain inflation models (fnl) and the sum of neutrino masses. A key limitation of such large-scale structure analyses is that dark matter halos and the galaxies within them only form at peaks of the dark matter density distribution, which is challenging to model. I will present recent work on this, where we found that a new basis of operators that account for large bulk flows successfully describes simulated halos at the field level over a surprisingly wide range of scales.
Talk given by Marcel Schmittfull at IPMU, University of Tokyo, Oct 31st 2019
Formation of low mass protostars and their circumstellar disks
Prospects for CMB lensing-galaxy clustering cross-correlations and modeling biased tracers
1. Prospects for CMB lensing—galaxy
clustering cross-correlations and
modeling biased tracers
Marcel Schmittfull
IPMU, October 2019
2. Cosmic Microwave Background (CMB)
Right after Big Bang
light scatters frequently
—> opaque
As Universe expands,
turns transparent
See surface where light
last scattered —
13.6996 bn yrs ago
This is the CMB
3. Cosmic Microwave Background (CMB)
Perr<latexit sha1_base64="HbsFOTY+FC07lz5Qau+blG/LzoY=">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</latexit>
Hot and cold blobs: Picture of the Universe 13.6996 bn years ago
4. Gravitational lensing of the CMB
4
Figure 1.2: Illustration of gravitational lensing of CMB photons by large-scale
structure (adapted from [10]).
gravitationally deflected by
galaxies and dark matter
CMB
13. Unlensed
Lensed
0 dz Mmin
dM
× |˜yi(M, z)|
2
|˜yj(M, z)|
2
, (8)
where gν is the spectral function of the SZ effect,
V (z) is the comoving volume of the universe integrated
to a redshift of zmax = 4, M is the virial mass such
that [log10(Mmin), log10(Mmax)] = [11, 16], dn/dM is the
FIG. 1: The impact of varying the lensing scaling parameter
on the lensed CMB temperature power spectrum, for AL =
[0,2,5,10].
Smidt+ (2010)
Global power
spectrum
unlensed
lensed
Peaks of global power are smeared out
Statistics of the CMB after lensing
15. Measured lensing magnification
Planck Collaboration: Planck 2018 lensing
Fig. 1. Mollweide projection in Galactic coordinates of the lensing-deflection reconstruction map from our baseline minimum-p
16. Cross-correlate with galaxy catalogs
Galaxy catalog is sensitive to tails of the distribution
of matter (galaxies form at peaks)
Lensing magnification probes entire distribution
(light is deflected by any mass)
Some models of the Big Bang enhance peaks, so
can test them
Cross-correlations particularly well suited: Can
determine ratio of two perfectly correlated random
variables with infinite precision
Planck Collaboration: Planck 2018 lensing
Fig. 1. Mollweide projection in Galactic coordinates of the lensing-deflection reconstruction map from our baseline minimum-
variance (MV) analysis. We show the Wiener-filtered displacement-like scalar field with multipoles ˆ↵MV
LM =
p
L(L + 1)ˆMV
LM , corre-
sponding to the gradient mode (or E mode) of the lensing deflection angle. Modes with L < 8 have been filtered out.
Our baseline lensing reconstruction map is shown in Fig. 1.
In Sect. 2 we explain how this was obtained, and the changes
compared to our analysis in PL2015. We also describe the new
optimal filtering approach used for our best polarization anal-
ysis. In Sect. 3 we present our main results, including power-
spectrum estimates, cosmological parameter constraints, and a
joint estimation of the lensing potential using the CIB. We end
the section by using the estimates of the lensing map to delens
the CMB, reducing the B-mode polarization power and sharpen-
ing the acoustic peaks. In Sect. 4 we describe in detail a number
of null and consistency tests, explaining the motivation for our
data cuts and the limits of our understanding of the data. We also
discuss possible contaminating signals, and assess whether they
are potentially important for our results. In Sect. 5 we briefly de-
scribe the various data products that are made available to the
community, and we end with conclusions in Sect. 6. A series of
appendices describe some technical details of the calculation of
various biases that are subtracted, and derive the error model for
the Monte Carlo estimates.
2. Data and methodology
This final Planck lensing analysis is based on the 2018 Planck
HFI maps as described in detail in Planck Collaboration III
(2018). Our baseline analysis uses the SMICA foreground-
cleaned CMB map described in Planck Collaboration IV (2018),
and includes both temperature and polarization information. We
use the Planck Full Focal Plane (FFP10) simulations, described
in detail in Planck Collaboration III (2018), to remove a num-
ber of bias terms and correctly normalize the lensing power-
spectrum estimates. Our analysis methodology is based on the
previous Planck analyses, as described in PL2013 and PL2015.
After a summary of the methodology, Sect. 2.1 also lists the
changes and improvements with respect to PL2015. Some de-
tails of the covariance matrix are discussed in Sect. 2.2, and de-
tails of the filtering in Sect. 2.3. The main set of codes applying
the quadratic estimators will be made public as part of the CMB
lensing toolbox LensIt.2
2.1. Lensing reconstruction
The five main steps of the lensing reconstruction are as follows.
1. Filtering of the CMB maps. The observed sky maps are cut
by a Galactic mask and have noise, so filtering is applied to
remove the mask and approximately optimally weight for the
noise. The lensing quadratic estimators use as input optimal
Wiener-filtered X = T, E, and B CMB multipoles, as well as
inverse-variance-weighted CMB maps. The latter maps can be
obtained easily from the Wiener-filtered multipoles by divid-
ing by the fiducial CMB power spectra Cfid
` before projecting
onto maps. We write the observed temperature T and polariza-
tion (written as the spin ±2 combination of Stokes parameters
±2P ⌘ Q ± iU) pixelized data as
0
BBBBBBBBBBBB@
Tdat
2Pdat
2Pdat
1
CCCCCCCCCCCCA
= BY
0
BBBBBBBBBBBB@
T
E
B
1
CCCCCCCCCCCCA
+ noise, (1)
2
https://github.com/carronj/LensIt
3
17. Sample variance cancellation
Galaxy number density:
CMB magnification:
Big Bang models are characterized by parameter
Planck Collaboration: Planck 2018 lensing
Fig. 1. Mollweide projection in Galactic coordinates of the lensing-deflection reconstruction map from our baseline minimum-
variance (MV) analysis. We show the Wiener-filtered displacement-like scalar field with multipoles ˆ↵MV
LM =
p
L(L + 1)ˆMV
LM , corre-
sponding to the gradient mode (or E mode) of the lensing deflection angle. Modes with L < 8 have been filtered out.
Our baseline lensing reconstruction map is shown in Fig. 1.
In Sect. 2 we explain how this was obtained, and the changes
compared to our analysis in PL2015. We also describe the new
optimal filtering approach used for our best polarization anal-
ysis. In Sect. 3 we present our main results, including power-
spectrum estimates, cosmological parameter constraints, and a
joint estimation of the lensing potential using the CIB. We end
the section by using the estimates of the lensing map to delens
the CMB, reducing the B-mode polarization power and sharpen-
ing the acoustic peaks. In Sect. 4 we describe in detail a number
of null and consistency tests, explaining the motivation for our
data cuts and the limits of our understanding of the data. We also
discuss possible contaminating signals, and assess whether they
are potentially important for our results. In Sect. 5 we briefly de-
scribe the various data products that are made available to the
community, and we end with conclusions in Sect. 6. A series of
appendices describe some technical details of the calculation of
various biases that are subtracted, and derive the error model for
the Monte Carlo estimates.
2. Data and methodology
This final Planck lensing analysis is based on the 2018 Planck
HFI maps as described in detail in Planck Collaboration III
(2018). Our baseline analysis uses the SMICA foreground-
cleaned CMB map described in Planck Collaboration IV (2018),
and includes both temperature and polarization information. We
use the Planck Full Focal Plane (FFP10) simulations, described
in detail in Planck Collaboration III (2018), to remove a num-
ber of bias terms and correctly normalize the lensing power-
spectrum estimates. Our analysis methodology is based on the
previous Planck analyses, as described in PL2013 and PL2015.
After a summary of the methodology, Sect. 2.1 also lists the
changes and improvements with respect to PL2015. Some de-
tails of the covariance matrix are discussed in Sect. 2.2, and de-
tails of the filtering in Sect. 2.3. The main set of codes applying
the quadratic estimators will be made public as part of the CMB
lensing toolbox LensIt.2
2.1. Lensing reconstruction
The five main steps of the lensing reconstruction are as follows.
1. Filtering of the CMB maps. The observed sky maps are cut
by a Galactic mask and have noise, so filtering is applied to
remove the mask and approximately optimally weight for the
noise. The lensing quadratic estimators use as input optimal
Wiener-filtered X = T, E, and B CMB multipoles, as well as
inverse-variance-weighted CMB maps. The latter maps can be
obtained easily from the Wiener-filtered multipoles by divid-
ing by the fiducial CMB power spectra Cfid
` before projecting
onto maps. We write the observed temperature T and polariza-
tion (written as the spin ±2 combination of Stokes parameters
±2P ⌘ Q ± iU) pixelized data as
0
BBBBBBBBBBBB@
Tdat
2Pdat
2Pdat
1
CCCCCCCCCCCCA
= BY
0
BBBBBBBBBBBB@
T
E
B
1
CCCCCCCCCCCCA
+ noise, (1)
2
https://github.com/carronj/LensIt
3
⇠ N(0, 2
)<latexit sha1_base64="fh2RNmAwU1/7vj3GTJMoGF9vnfw=">AAACInicbVDLSgMxFM3UV62vqks3wSJUkTJTBV24KLhxJRXsAzpjuZOmbWgyMyQZoQzzD36E3+BW1+7EleDGPzF9LGzrgcDhnHO5N8ePOFPatr+szNLyyupadj23sbm1vZPf3aurMJaE1kjIQ9n0QVHOAlrTTHPajCQF4XPa8AfXI7/xSKViYXCvhxH1BPQC1mUEtJHa+RN3AFEE2FVMYFeA7hPgyW1atE+N1BPQngQeysftfMEu2WPgReJMSQFNUW3nf9xOSGJBA004KNVy7Eh7CUjNCKdpzo0VjYAMoEdbhgYgqPKS8Z9SfGSUDu6G0rxA47H6dyIBodRQ+CY5ulrNeyPxP68V6+6ll7AgijUNyGRRN+ZYh3hUEO4wSYnmQ0OASGZuxaQPEog2Nc5s8UVqOnHmG1gk9XLJOSuV784LlatpO1l0gA5RETnoAlXQDaqiGiLoCb2gV/RmPVvv1of1OYlmrOnMPpqB9f0LdyykHw==</latexit>
g = (1 + fNL)<latexit sha1_base64="qSZyT7RHnm+puQb5ztG1Up+wy9s=">AAACEnicbVDLSsNAFJ3UV62vqDvdDBahIpSkCrpQKLhxIVLBPqAJYTKdtENnkjAzEUoI+BF+g1tduxO3/oBL/8Rpm4WtHrhwOOde7r3HjxmVyrK+jMLC4tLySnG1tLa+sbllbu+0ZJQITJo4YpHo+EgSRkPSVFQx0okFQdxnpO0Pr8Z++4EISaPwXo1i4nLUD2lAMVJa8sy9PryEFfs48FJHcHh7kx1BZ4jiGEHPLFtVawL4l9g5KYMcDc/8dnoRTjgJFWZIyq5txcpNkVAUM5KVnESSGOEh6pOupiHiRLrp5IcMHmqlB4NI6AoVnKi/J1LEpRxxX3dypAZy3huL/3ndRAXnbkrDOFEkxNNFQcKgiuA4ENijgmDFRpogLKi+FeIBEggrHdvMFp9nOhN7PoG/pFWr2ifV2t1puX6Rp1ME++AAVIANzkAdXIMGaAIMHsEzeAGvxpPxZrwbH9PWgpHP7IIZGJ8/vOGcLw==</latexit>
fNL<latexit sha1_base64="J3TCtcachvp5KbIs7TN/zj+HhGo=">AAACAHicbVA9SwNBEJ2LXzF+RS1tFoNgFe6ioIVFwMZCJIL5wCSEvc1esmR379jdE8KRxt9gq7Wd2PpPLP0n7iVXmMQHA4/3ZpiZ50ecaeO6305uZXVtfSO/Wdja3tndK+4fNHQYK0LrJOShavlYU84krRtmOG1FimLhc9r0R9ep33yiSrNQPphxRLsCDyQLGMHGSo9BL+koge5uJ71iyS27U6Bl4mWkBBlqveJPpx+SWFBpCMdatz03Mt0EK8MIp5NCJ9Y0wmSEB7RtqcSC6m4yvXiCTqzSR0GobEmDpurfiQQLrcfCt50Cm6Fe9FLxP68dm+CymzAZxYZKMlsUxByZEKXvoz5TlBg+tgQTxeytiAyxwsTYkOa2+CLNxFtMYJk0KmXvrFy5Py9Vr7J08nAEx3AKHlxAFW6gBnUgIOEFXuHNeXbenQ/nc9aac7KZQ5iD8/UL/Y2XJA==</latexit>
18. Sample variance cancellation
Galaxy number density:
CMB magnification:
Big Bang models are characterized by parameter
Planck Collaboration: Planck 2018 lensing
Fig. 1. Mollweide projection in Galactic coordinates of the lensing-deflection reconstruction map from our baseline minimum-
variance (MV) analysis. We show the Wiener-filtered displacement-like scalar field with multipoles ˆ↵MV
LM =
p
L(L + 1)ˆMV
LM , corre-
sponding to the gradient mode (or E mode) of the lensing deflection angle. Modes with L < 8 have been filtered out.
Our baseline lensing reconstruction map is shown in Fig. 1.
In Sect. 2 we explain how this was obtained, and the changes
compared to our analysis in PL2015. We also describe the new
optimal filtering approach used for our best polarization anal-
ysis. In Sect. 3 we present our main results, including power-
spectrum estimates, cosmological parameter constraints, and a
joint estimation of the lensing potential using the CIB. We end
the section by using the estimates of the lensing map to delens
the CMB, reducing the B-mode polarization power and sharpen-
ing the acoustic peaks. In Sect. 4 we describe in detail a number
of null and consistency tests, explaining the motivation for our
data cuts and the limits of our understanding of the data. We also
discuss possible contaminating signals, and assess whether they
are potentially important for our results. In Sect. 5 we briefly de-
scribe the various data products that are made available to the
community, and we end with conclusions in Sect. 6. A series of
appendices describe some technical details of the calculation of
various biases that are subtracted, and derive the error model for
the Monte Carlo estimates.
2. Data and methodology
This final Planck lensing analysis is based on the 2018 Planck
HFI maps as described in detail in Planck Collaboration III
(2018). Our baseline analysis uses the SMICA foreground-
cleaned CMB map described in Planck Collaboration IV (2018),
and includes both temperature and polarization information. We
use the Planck Full Focal Plane (FFP10) simulations, described
in detail in Planck Collaboration III (2018), to remove a num-
ber of bias terms and correctly normalize the lensing power-
spectrum estimates. Our analysis methodology is based on the
previous Planck analyses, as described in PL2013 and PL2015.
After a summary of the methodology, Sect. 2.1 also lists the
changes and improvements with respect to PL2015. Some de-
tails of the covariance matrix are discussed in Sect. 2.2, and de-
tails of the filtering in Sect. 2.3. The main set of codes applying
the quadratic estimators will be made public as part of the CMB
lensing toolbox LensIt.2
2.1. Lensing reconstruction
The five main steps of the lensing reconstruction are as follows.
1. Filtering of the CMB maps. The observed sky maps are cut
by a Galactic mask and have noise, so filtering is applied to
remove the mask and approximately optimally weight for the
noise. The lensing quadratic estimators use as input optimal
Wiener-filtered X = T, E, and B CMB multipoles, as well as
inverse-variance-weighted CMB maps. The latter maps can be
obtained easily from the Wiener-filtered multipoles by divid-
ing by the fiducial CMB power spectra Cfid
` before projecting
onto maps. We write the observed temperature T and polariza-
tion (written as the spin ±2 combination of Stokes parameters
±2P ⌘ Q ± iU) pixelized data as
0
BBBBBBBBBBBB@
Tdat
2Pdat
2Pdat
1
CCCCCCCCCCCCA
= BY
0
BBBBBBBBBBBB@
T
E
B
1
CCCCCCCCCCCCA
+ noise, (1)
2
https://github.com/carronj/LensIt
3
⇠ N(0, 2
)<latexit sha1_base64="fh2RNmAwU1/7vj3GTJMoGF9vnfw=">AAACInicbVDLSgMxFM3UV62vqks3wSJUkTJTBV24KLhxJRXsAzpjuZOmbWgyMyQZoQzzD36E3+BW1+7EleDGPzF9LGzrgcDhnHO5N8ePOFPatr+szNLyyupadj23sbm1vZPf3aurMJaE1kjIQ9n0QVHOAlrTTHPajCQF4XPa8AfXI7/xSKViYXCvhxH1BPQC1mUEtJHa+RN3AFEE2FVMYFeA7hPgyW1atE+N1BPQngQeysftfMEu2WPgReJMSQFNUW3nf9xOSGJBA004KNVy7Eh7CUjNCKdpzo0VjYAMoEdbhgYgqPKS8Z9SfGSUDu6G0rxA47H6dyIBodRQ+CY5ulrNeyPxP68V6+6ll7AgijUNyGRRN+ZYh3hUEO4wSYnmQ0OASGZuxaQPEog2Nc5s8UVqOnHmG1gk9XLJOSuV784LlatpO1l0gA5RETnoAlXQDaqiGiLoCb2gV/RmPVvv1of1OYlmrOnMPpqB9f0LdyykHw==</latexit>
If we measure only :
If we measure and :
g<latexit sha1_base64="lhiJBeb4nMmQQxTZcqh1G/By96k=">AAAB93icbVA9SwNBEJ3zM8avqKXNYRCswl0UtLAI2FgmYD4gOcLeZi5Zsrt37O4J4cgvsNXaTmz9OZb+EzfJFSbxwcDjvRlm5oUJZ9p43rezsbm1vbNb2CvuHxweHZdOTls6ThXFJo15rDoh0ciZxKZhhmMnUUhEyLEdjh9mfvsZlWaxfDKTBANBhpJFjBJjpcawXyp7FW8Od534OSlDjnq/9NMbxDQVKA3lROuu7yUmyIgyjHKcFnupxoTQMRli11JJBOogmx86dS+tMnCjWNmSxp2rfycyIrSeiNB2CmJGetWbif953dREd0HGZJIalHSxKEq5a2J39rU7YAqp4RNLCFXM3urSEVGEGpvN0pZQTG0m/moC66RVrfjXlWrjply7z9MpwDlcwBX4cAs1eIQ6NIECwgu8wpszcd6dD+dz0brh5DNnsATn6xdwBpN/</latexit>
g<latexit sha1_base64="lhiJBeb4nMmQQxTZcqh1G/By96k=">AAAB93icbVA9SwNBEJ3zM8avqKXNYRCswl0UtLAI2FgmYD4gOcLeZi5Zsrt37O4J4cgvsNXaTmz9OZb+EzfJFSbxwcDjvRlm5oUJZ9p43rezsbm1vbNb2CvuHxweHZdOTls6ThXFJo15rDoh0ciZxKZhhmMnUUhEyLEdjh9mfvsZlWaxfDKTBANBhpJFjBJjpcawXyp7FW8Od534OSlDjnq/9NMbxDQVKA3lROuu7yUmyIgyjHKcFnupxoTQMRli11JJBOogmx86dS+tMnCjWNmSxp2rfycyIrSeiNB2CmJGetWbif953dREd0HGZJIalHSxKEq5a2J39rU7YAqp4RNLCFXM3urSEVGEGpvN0pZQTG0m/moC66RVrfjXlWrjply7z9MpwDlcwBX4cAs1eIQ6NIECwgu8wpszcd6dD+dz0brh5DNnsATn6xdwBpN/</latexit>
<latexit sha1_base64="9bJNSyJCqgQXzYZuEwEOn7kVJFo=">AAAB/HicbVA9SwNBEN2LXzF+RS1tFoNgFe6ioIVFwMYygvmA5Ahzm71kze7esrsnhBB/g63WdmLrf7H0n7hJrjCJDwYe780wMy9SnBnr+99ebm19Y3Mrv13Y2d3bPygeHjVMkmpC6yThiW5FYChnktYts5y2lKYgIk6b0fB26jefqDYskQ92pGgooC9ZzAhYJzU6Q1AKusWSX/ZnwKskyEgJZah1iz+dXkJSQaUlHIxpB76y4Ri0ZYTTSaGTGqqADKFP245KENSE49m1E3zmlB6OE+1KWjxT/06MQRgzEpHrFGAHZtmbiv957dTG1+GYSZVaKsl8UZxybBM8fR33mKbE8pEjQDRzt2IyAA3EuoAWtkRi4jIJlhNYJY1KObgoV+4vS9WbLJ08OkGn6BwF6ApV0R2qoToi6BG9oFf05j17796H9zlvzXnZzDFagPf1C1gmlbM=</latexit>
var( ˆfNL) ⇠ 2
<latexit sha1_base64="q7aORK2uPQFWV9/VujOi1K0JE98=">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</latexit>
ˆfNL =
g
1 ) var( ˆfNL) = 0
<latexit sha1_base64="acTMnBQP2Js1IiQWFlfVXRtc34c=">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</latexit>
g = (1 + fNL)<latexit sha1_base64="qSZyT7RHnm+puQb5ztG1Up+wy9s=">AAACEnicbVDLSsNAFJ3UV62vqDvdDBahIpSkCrpQKLhxIVLBPqAJYTKdtENnkjAzEUoI+BF+g1tduxO3/oBL/8Rpm4WtHrhwOOde7r3HjxmVyrK+jMLC4tLySnG1tLa+sbllbu+0ZJQITJo4YpHo+EgSRkPSVFQx0okFQdxnpO0Pr8Z++4EISaPwXo1i4nLUD2lAMVJa8sy9PryEFfs48FJHcHh7kx1BZ4jiGEHPLFtVawL4l9g5KYMcDc/8dnoRTjgJFWZIyq5txcpNkVAUM5KVnESSGOEh6pOupiHiRLrp5IcMHmqlB4NI6AoVnKi/J1LEpRxxX3dypAZy3huL/3ndRAXnbkrDOFEkxNNFQcKgiuA4ENijgmDFRpogLKi+FeIBEggrHdvMFp9nOhN7PoG/pFWr2ifV2t1puX6Rp1ME++AAVIANzkAdXIMGaAIMHsEzeAGvxpPxZrwbH9PWgpHP7IIZGJ8/vOGcLw==</latexit>
fNL<latexit sha1_base64="J3TCtcachvp5KbIs7TN/zj+HhGo=">AAACAHicbVA9SwNBEJ2LXzF+RS1tFoNgFe6ioIVFwMZCJIL5wCSEvc1esmR379jdE8KRxt9gq7Wd2PpPLP0n7iVXmMQHA4/3ZpiZ50ecaeO6305uZXVtfSO/Wdja3tndK+4fNHQYK0LrJOShavlYU84krRtmOG1FimLhc9r0R9ep33yiSrNQPphxRLsCDyQLGMHGSo9BL+koge5uJ71iyS27U6Bl4mWkBBlqveJPpx+SWFBpCMdatz03Mt0EK8MIp5NCJ9Y0wmSEB7RtqcSC6m4yvXiCTqzSR0GobEmDpurfiQQLrcfCt50Cm6Fe9FLxP68dm+CymzAZxYZKMlsUxByZEKXvoz5TlBg+tgQTxeytiAyxwsTYkOa2+CLNxFtMYJk0KmXvrFy5Py9Vr7J08nAEx3AKHlxAFW6gBnUgIOEFXuHNeXbenQ/nc9aac7KZQ5iD8/UL/Y2XJA==</latexit>
19. 3Sample variance cancellation
Planck Collaboration: Planck 2018 lensing
Fig. 1. Mollweide projection in Galactic coordinates of the lensing-deflection reconstruction map from our baseline minimum-
variance (MV) analysis. We show the Wiener-filtered displacement-like scalar field with multipoles ˆ↵MV
LM =
p
L(L + 1)ˆMV
LM , corre-
sponding to the gradient mode (or E mode) of the lensing deflection angle. Modes with L < 8 have been filtered out.
Our baseline lensing reconstruction map is shown in Fig. 1.
In Sect. 2 we explain how this was obtained, and the changes
compared to our analysis in PL2015. We also describe the new
optimal filtering approach used for our best polarization anal-
ysis. In Sect. 3 we present our main results, including power-
spectrum estimates, cosmological parameter constraints, and a
joint estimation of the lensing potential using the CIB. We end
the section by using the estimates of the lensing map to delens
the CMB, reducing the B-mode polarization power and sharpen-
ing the acoustic peaks. In Sect. 4 we describe in detail a number
of null and consistency tests, explaining the motivation for our
data cuts and the limits of our understanding of the data. We also
discuss possible contaminating signals, and assess whether they
are potentially important for our results. In Sect. 5 we briefly de-
scribe the various data products that are made available to the
community, and we end with conclusions in Sect. 6. A series of
appendices describe some technical details of the calculation of
various biases that are subtracted, and derive the error model for
the Monte Carlo estimates.
2. Data and methodology
This final Planck lensing analysis is based on the 2018 Planck
HFI maps as described in detail in Planck Collaboration III
(2018). Our baseline analysis uses the SMICA foreground-
cleaned CMB map described in Planck Collaboration IV (2018),
and includes both temperature and polarization information. We
use the Planck Full Focal Plane (FFP10) simulations, described
in detail in Planck Collaboration III (2018), to remove a num-
ber of bias terms and correctly normalize the lensing power-
spectrum estimates. Our analysis methodology is based on the
previous Planck analyses, as described in PL2013 and PL2015.
After a summary of the methodology, Sect. 2.1 also lists the
changes and improvements with respect to PL2015. Some de-
tails of the covariance matrix are discussed in Sect. 2.2, and de-
tails of the filtering in Sect. 2.3. The main set of codes applying
the quadratic estimators will be made public as part of the CMB
lensing toolbox LensIt.2
2.1. Lensing reconstruction
The five main steps of the lensing reconstruction are as follows.
1. Filtering of the CMB maps. The observed sky maps are cut
by a Galactic mask and have noise, so filtering is applied to
remove the mask and approximately optimally weight for the
noise. The lensing quadratic estimators use as input optimal
Wiener-filtered X = T, E, and B CMB multipoles, as well as
inverse-variance-weighted CMB maps. The latter maps can be
obtained easily from the Wiener-filtered multipoles by divid-
ing by the fiducial CMB power spectra Cfid
` before projecting
onto maps. We write the observed temperature T and polariza-
tion (written as the spin ±2 combination of Stokes parameters
±2P ⌘ Q ± iU) pixelized data as
0
BBBBBBBBBBBB@
Tdat
2Pdat
2Pdat
1
CCCCCCCCCCCCA
= BY
0
BBBBBBBBBBBB@
T
E
B
1
CCCCCCCCCCCCA
+ noise, (1)
2
https://github.com/carronj/LensIt
3
Std. Big BangStd. Big Bang
Enhanced tails
(Angular scale)-1
Dalal+ (2008), Seljak (2009), McDonald & Seljak (2009), MS & Seljak (2018)
21. Forecast for future experiments
Compute Fisher information matrix assuming Gaussian likelihood
(curvature of the likelihood near maximum = Hessian)
Inverse gives lower bound on parameter error bars
shot
mber
er-
alize
nal-
t in
bias
mple
rove
and
esti-
and
de-
rove
e in
VII.
oss-
e 14
III.
and
wer
one-dimensional data vector that starts with all spectra
at `min, continues with all spectra at `min + 1, etc:
d = d`min
, d`min+1, . . . , d`max
. (15)
At each `,
d` = C11
` , C12
` , . . . , CNN
` (16)
contains N(N + 1)/2 spectra Cij
` with j i. Assum-
ing Eq. (14), the covariance cov(d, d) is then a block-
diagonal matrix with `max `min + 1 blocks of size
N(N + 1)/2 ⇥ N(N + 1)/2, which is easily inverted if
the number of fields is N . 100. The Fisher matrix at
the power spectrum level is then
Fab =
`maxX
`=`min
@d`
@✓a
[cov(d`, d`)] 1 @d`
@✓b
. (17)
We evaluate this without binning in `.2
for Fisher forecasts with Gaussian covariances, this may be
more challenging for actual data analyses. In that case one
may want to compress the observations before forming power
spectra (see Appendix D). Although many of the cross-spectra
have zero signal in the Limber approximation because they
Fab =
⌧
@2
ln L(d|✓)
@✓a@✓b<latexit sha1_base64="ta6ogSoEc9qH8kFkQwG7srESlKI=">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</latexit>
22. If models work and systematics under control, what can we hope for?
Optimistic setting
23. 0 1 2 3 4 5 6 7
z
10 3
10 2
10 1
1
101
102dn/dz[arcmin2
]
0
0.2
0.4
0.6
0.8
1.0
W
W
LSST Optimistic
LSST Gold
FIG. 1: The redshift distribution of the CMB lensing convergence
(red curve, normalized to a unit maximum) and LSST galaxy sam-
ples, both Optimistic (light gray) and Gold (dark gray). We as-
sume 16 tomographic redshift bins in the range 0 < z < 7, cross-
FIG. 2:
masses w
configura
primordi
LSST number density
Gold sample
CMB lensing kernel
Optimistic sample
At low z, use clustering redshifts (Gorecki+ 2014)
At high z, add Lyman-break galaxies (dropouts; extrapolated from HSC observations)
Total of 66 galaxies per arcmin2 (MS & Seljak 2018)
24. Lyman-break galaxies (LBGs) at z>3
Photons blueward of 912 Å
ionize neutral hydrogen in
young star-forming galaxies
Lyman break at (1+z) 912 Å
Blue band ‘dropout’
0.5 million z=4-7 LBGs found
by HSC/Goldrush in 100 deg2
Expect ~100 million in LSST
Good for CMB lensing Xcorrel
Ellis 1998
Ono, Ouchi+ (2018) 1704.06004
25. di↵erent fsky, maybe? MS: We do not include photo-z
errors. Hope this is not too bad because lensing kernel is
very broad. Could matter for tracer cross tracer though.
Maybe include photo z errors in forecasts?
FIG. 3. Redshift distribution and tomographic redshift bins.
For CMB lensing (solid black), we plot the dn/dz that would
yield C
if integrated over, with arbitrary normalization. It
at high
smaller
the low
also ha
at low `
For a
contrib
satisfyi
ples, wh
the cor
black li
ples are
coe cie
(see Re
combin
the CM
peak co
motivat
niques.
but is s
Dropout
galaxies
Tomographic redshift bins
MS & Seljak (2018)
26. Power spectra: CMB-S4 & LSST
5
FIG. 4. Left panel: Angular auto-power spectra of CMB-S4 lensing map (black) and LSST galaxy density g (colored). Solid
ines show the signal power (not including lensing noise or shot noise), and shaded regions show 1 error bars assuming the
Gaussian covariance (27), fsky = 0.5, minimum variance lensing noise expected for CMB-S4, and LSST number density shown
n Fig. 2. Dashed lines show lensing reconstruction noise (black) and shot noise (colored). Right panel: Angular cross-spectra
between CMB lensing and LSST galaxy density.
MS & Seljak (2018)
gg
κκ
κg
noise
CXX0
` =
Z
z
P X X0 (k = `/ (z), z)
⇥ WX (z)bX(z)WX0 (z)bX0 (z).
CXX0
` =
2
⇡
Z 1
0
dk
k
X,`(k) X0,`(k)k3
P X X0 (k, z = 0),
X,`(k) ⌘
Z 1
0
d ¯WX ( )j`(k ).
Low ell: High ell:
27. SNR of auto-power spectra
MS & Seljak (2018)
7
the combined
relation coef-
be as high as
g very broad
100. The rea-
ical distances
a very low k
0 2
hMpc 1
).
power spec-
wer spectrum
projection in-
lues of and
scales smaller
the contribu-
re important.
urces at z > 4
noise reduces
variance can-
re the tracers
`max
SNR of CXX
500 1000 2000
CMB 233 406 539
BOSS LRG z=0-0.9 140 187 230
SDSS r < 22 z=0-0.5 247 487 936
SDSS r < 22 z=0.5-0.8 247 487 936
DESI BGS z=0-0.5 230 417 665
DESI ELG z=0.6-0.8 158 210 256
DESI ELG z=0.8-1.7 150 194 225
DESI LRG z=0.6-1.2 184 267 349
DESI QSO z=0.6-1.9 44.8 48.8 50.8
LSST i < 27 (3yr) z=0-0.5 250 496 982
LSST i < 27 (3yr) z=0.5-1 250 496 979
LSST i < 27 (3yr) z=1-2 249 492 956
LSST i < 27 (3yr) z=2-3 245 469 830
LSST i < 27 (3yr) z=3-4 239 444 724
LSST i < 27 (3yr) z=4-7 224 387 555
TABLE I. Total signal-to-noise of auto-power spectra CXX
`
28. SNR of kg cross-power spectra
Gaussian covariance (14), fsky = 0.5, minimum variance lensing noise expected
in Fig. 4. Dashed lines show lensing reconstruction noise (black) and shot nois
between CMB lensing and LSST galaxy density.
`max
SNR of CCMBX
500 1000 2000
BOSS LRG z=0-0.9 77.3 117 159
SDSS r < 22 z=0-0.5 88.3 167 284
SDSS r < 22 z=0.5-0.8 88.3 167 284
DESI BGS z=0-0.5 50.1 93.5 144
DESI ELG z=0.6-0.8 50.7 73.5 97
DESI ELG z=0.8-1.7 103 148 185
DESI LRG z=0.6-1.2 86.7 133 182
DESI QSO z=0.6-1.9 74.9 94.5 108
LSST i < 27 (3yr) z=0-0.5 78.1 150 258
LSST i < 27 (3yr) z=0.5-1 112 202 338
LSST i < 27 (3yr) z=1-2 144 259 406
LSST i < 27 (3yr) z=2-3 121 219 324
LSST i < 27 (3yr) z=3-4 101 182 261
LSST i < 27 (3yr) z=4-7 94 167 229
TABLE II. Like Table I but for CMB-lensing–clustering cross-
spectra Cg
` .
For galaxy
by more than
z & 2). At `
but less than
of Fig. 7 com
ance of each
ments or exp
shows that t
0.1 per mo
lower-redshif
ample, the f
0.05 per mo
The g cro
nal than gg a
quence of com
than with th
signal-to-nois
sponding gg
the cross-cor
and each ind
Fig. 5 above,MS & Seljak (2018)
29. Correlation of CMB lensing and galaxies
FIG. 3. Correlation r = Cg
(C
Cgg
) 1/2
of CMB-S4 lensing
A. Local primordia
1. Motivation to
The large-scale structure of
primordial density fluctuation
universe. Measuring the stat
scale structures using galaxy
can thus constrain the physics
fluctuations. A non-Gaussian
tribution function (pdf) can o
inflation models, involving for
higher-derivative operators.
local type of primordial non-
the primordial potential is th
sian field and its square, (x)+
has a non-Gaussian pdf. If w
amplitude, fNL & 1, it will r
of the inflationary expansion
robust way [25, 26]. This is o
tional means to rule out a who
early-universe models.
In practice the measurement
threshold signal fNL = 1 sepa
and multi-field models has a v
ables. The best constraints, f
Planck’s CMB temperature a
ments [27].
does not improve our forecasts because we cros
against CMB lensing, so that redshift accurac
less important than number density. We assu
of b(z) = 1.7 ¯D 1
(z) where ¯D(z = 0) = 1.
For DESI, we use five redshift samples, wit
densities from Table 2.3 in Ref. [47]: The lo
BGS sample at 0 z 0.5 with 9.6 ⇥ 106
o
bias b(z) = 1.34 ¯D 1
(z), the LRG sample at 0.6
with 3.9 ⇥ 106
objects and bias b(z) = 1.7 ¯D
ELG sample at 0.6 z 0.8 with 3.5 ⇥ 1
and bias b(z) = 0.84 ¯D 1
(z), a second ELG
0.8 z 1.7 with 1.3 ⇥ 107
objects and the
and a QSO sample at 0.6 z 1.9 with 1.4⇥1
and bias 1.2 ¯D 1
(z). In each case, the number
refers to a survey area of 14, 000 deg2
.
D. CMB lensing–LSS correlation coe
The performance of the cross-correlation an
pends on the cross-correlation coe cient
r` =
Cg
`
q
ˆC
`
ˆCgg
`
between the measured CMB lensing converge
the observed galaxy density g, where the pow
ˆC include lensing reconstruction noise and s
CMB lensing and
galaxy maps are
up to 95%
correlated
MS & Seljak (2018)
30. MS & Seljak (2018); Forecast papers for Simons Observatory, CMB-S4, NASA PICO satellite
(Largest angular scale)-1
—> Potential to rule out all Big Bang models driven by a single field
Forecast for local non-Gaussianity
12
FIG. 9. MS: Date: 20 Aug 2017
Constraints on primordial non-Gaussianity amplitude fNL as
a function of minimum wavenumber `min, for di↵erent LSS
surveys (colors), with full sky overlap between observations
Threshold needed to
distinguish Big Bang
models
Forecastfor
Different experiments
`min ⇠<latexit sha1_base64="JpqnNPjmjQoSX2eihOhMzZtQsZ0=">AAACDXicbVDLSsNAFJ3UV62vqLhyM1gEVyWpgi5cFNy4rGAf0IQwmU7aoTOTMDMRSsg3+A1ude1O3PoNLv0TJ20WtvXAhcM593IPJ0wYVdpxvq3K2vrG5lZ1u7azu7d/YB8edVWcSkw6OGax7IdIEUYF6WiqGeknkiAeMtILJ3eF33siUtFYPOppQnyORoJGFCNtpMA+8QhjgceRHkuecSpyT1Ee2HWn4cwAV4lbkjoo0Q7sH28Y45QToTFDSg1cJ9F+hqSmmJG85qWKJAhP0IgMDBWIE+Vns/g5PDfKEEaxNCM0nKl/LzLElZry0GwWOdWyV4j/eYNURzd+RkWSaiLw/FGUMqhjWHQBh1QSrNnUEIQlNVkhHiOJsDaNLXwJeW46cZcbWCXdZsO9bDQfruqt27KdKjgFZ+ACuOAatMA9aIMOwCADL+AVvFnP1rv1YX3OVytWeXMMFmB9/QJGU5zP</latexit>
With/without sample
variance cancellation
(fNL)<latexitsha1_base64="SjRPaUwNSZsk4HkulZr/qbH5CbU=">AAACCnicbVDLSsNAFJ34rPUV69LNYBHqpiRV0IWLghsXIhXsA5oQJtNJO3RmEmYmYgn9A7/Bra7diVt/wqV/4rTNwrYeuHA4517O5YQJo0o7zre1srq2vrFZ2Cpu7+zu7dsHpZaKU4lJE8cslp0QKcKoIE1NNSOdRBLEQ0ba4fB64rcfiVQ0Fg96lBCfo76gEcVIGymwS56ifY4qUZB5ksO72/FpYJedqjMFXCZuTsogRyOwf7xejFNOhMYMKdV1nUT7GZKaYkbGRS9VJEF4iPqka6hAnCg/m/4+hidG6cEolmaEhlP170WGuFIjHppNjvRALXoT8T+vm+ro0s+oSFJNBJ4FRSmDOoaTImCPSoI1GxmCsKTmV4gHSCKsTV1zKSEfm07cxQaWSatWdc+qtfvzcv0qb6cAjsAxqAAXXIA6uAEN0AQYPIEX8ArerGfr3fqwPmerK1Z+cwjmYH39AhnnmmM=</latexit>
31. Growth of structure as function of time
Expansion history / geometry / dark energy
Sum of neutrino masses
Galaxy formation?
Other science from cross-correlations
33. Yu, Knight et al. (arXiv:1809.02120)
0 1 2 3 4 5 6 7
z
0.98
0.99
1.0
1.01
8(z)/8(z),fiducial
a 3f /5
Marginalize over CDM parameters and Bi; S4 + DESI added
Fixing CDM parameters
FIG. 4: 1 constraints on the matter amplitude 8 in 6 tomo-
graphic redshift bins, z = 0 0.5, 0.5 1, 1 2, 2 3, 3 4, 4 7,
60 meV neutrinos
Massless neutrinos
Example: Sum of neutrino masses
Varianceoffluctuations
(Size of the Universe)-1 - 1
Fix other cosmological params
Marginalize over other cosmological params
34. Uses z<7
redshift lever
arm
Measures
neutrino mass
without optical
depth to CMB
Independent &
competitive to
usual probes
Yu, Knight, Sherwin+ (1809.02120)
3
0.2 1 2 3 4 5 6 7
zmax
20
30
40
50
100
(
P
m)[meV]
S4Lens + LSST (Optimistic)
S4Lens + LSST (Optimistic) + Planck(l > 30)
S4Lens + LSST (Optimistic) + S4(l > 30)
Planck/S4(l > 30) + DESI
S4Lens + Planck/S4(l > 30) + DESI
S4Lens + LSST (Gold) + Planck/S4(l > 30) + DESI
S4Lens + LSST (Optimistic) +Planck/S4(l > 30) + DESI
FIG. 2: Forecasted 1 constraints on the sum of the neutrino
100
30
20
50
3
0.2 1 2 3 4 5 6 7
zmax
20
30
40
50
100
(
P
m)[meV]
S4Lens + LSST (Optimistic)
S4Lens + LSST (Optimistic) + Planck(l > 30)
S4Lens + LSST (Optimistic) + S4(l > 30)
Planck/S4(l > 30) + DESI
S4Lens + Planck/S4(l > 30) + DESI
S4Lens + LSST (Gold) + Planck/S4(l > 30) + DESI
S4Lens + LSST (Optimistic) +Planck/S4(l > 30) + DESI
FIG. 2: Forecasted 1 constraints on the sum of the neutrino
masses without optical depth information, for di↵erent experiment
configurations: CMB-S4 lensing and LSST clustering (black) +
primordial CMB data (green dotted for Planck and green solid for
Example: Sum of neutrino masses
35. 33
Sensitive to
kmax, but not
too much
when all data
combined
(
P
m⌫ ) [meV] (Gold/Optimistic)
kmax Lens + LSST + Planck/S4 T&P + DESI
0.05 307 / 243 94 / 68 32 / 29
0.1 176 / 129 68 / 53 31 / 27
0.2 107 / 71 47 / 38 28 / 25
0.3 84 / 55 40 / 33 27 / 24
0.4 79 / 49 38 / 31 26 / 23
TABLE I: Forecasts of the neutrino mass constraints without op-
tical depth information, for di↵erent LSST number densities and
redshift distributions, kmax limits, and lensing reconstruction noise
levels. For di↵erent combinations of data, constraints provided on
the left assume the LSST Gold sample, and those on the right
assume the LSST Optimistic sample.
spectra is given by
n
Len
kmax (⌧) =
0.3 25
(
P
Lens +
kmax (⌧) =
0.3 22 /
TABLE II: Forecasts of t
flat priors on the optica
S4 lensing, S4 primary
BAO information. Botto
numbers on the left assu
assume the Optimistic s
Yu, Knight, Sherwin+ (1809.02120)
Ignores scale-dependent bias due to neutrinos
LoVerde 2014, LoVerde 2016, Muñoz & Dvorkin 2018, MS &
Seljak 2018, Chiang, LoVerde & Villaescusa-Navarro 2019
Example: Sum of neutrino masses
36. Modeling all power spectra for nonlinear scales (high L)
Photometric redshift errors
Relationship between galaxies and dark matter (galaxy bias)
Challenges for growth measurements
37. Other cool things to do with CMB lensing
(1) Joint analysis of CMB and CMB magnification, with covariance
(2) Estimate unlensed CMB
(3) Biases of the magnification estimator
Figure 1.2: Illustration of gravitational lensing of CMB photons by large-scale
structure (adapted from [10]).
reconstructing these lenses from the observed CMB we can obtain crucial infor-
mation on e.g. dark energy or the geometry of the universe. Several experiments
will provide high quality CMB data in the next few years, e.g. full-mission Planck
data including polarization, ACT/ACTPol, Polarbear and SPT/SPTpol.
The contribution of this thesis, presented in Chapter 5, is a thorough analysis
of how the reconstructed lensing information can be combined with the primary
CMB data to perform reliably a joint parameter estimation. Such joint analy-
ses are important to break degeneracies that limit the information that can be
extracted from the CMB fluctuations laid down at recombination. The joint anal-
ysis is non-trivial because the part of the lensing information that is present in
the primary CMB power spectrum as well as in the lensing reconstruction is po-
tentially double-counted. We quantify the temperature-lensing cross-correlation
analytically, finding two physical contributions and confirming the results with
simulated lensed CMB maps. This cross-correlation has not been considered be-
fore and it could have turned out to be anywhere between zero and unity. We
15
(a) Theoretical matter cosmic variance contribution (D9) (b) Simulations
FIG. 5. (a) Theoretical matter cosmic variance contribution (D9) to the correlation of the unbinned power spectra of recon-
structed lensing potential and lensed temperature. The covariance (D9) is converted to a correlation using the same conversion
factor as in (38). (b) Measured correlation ˆcorrel( ˆC
ˆin
ˆin
L , ˆC
˜T ˜T
LT
ˆCT T
LT
) of the input lensing potential power and the di↵erence
of noise-free lensed and unlensed temperature powers in 1000 simulations.
FIG. 6. Left: The approximate contribution to the covariance between [L(L + 1)]2 ˆC
ˆˆ
L /(2⇡) and L0
(L0
+ 1) ˆC
˜T ˜T
L0 /(2⇡) from
cosmic variance of the lenses, derived from Eq. (D9). Right: The rank-one approximation to the matrix on the left from
retaining only the largest singular value.TODOO: discuss this in main text or appendix
CT
correlations in all calculations we try to eliminate correlations between the lensing potential and the unlensed
temperature in the simulations by considering
ˆcov( ˆC
ˆˆ
L , ˆC
˜T ˜T
L0 )|CV( ) = ˆcov( ˆC
ˆin
ˆin
L , ˆC
˜T ˜T
L0
ˆCT T
L0 ), (52)
where ˆC
ˆin
ˆin
is the empirical power of the input lensing potential and ˆCT T
is the empirical power of the unlensed
temperature. Subtracting the unlensed from the lensed empirical power spectrum also reduces the noise of the
covariance estimate because it eliminates the scatter due to cosmic variance of the unlensed temperature. We estimate
the covariance in simulations similarly to (48). As shown in Fig. 5b these measurements agree with the theoretical
expectation from (D9).
`0
<latexit sha1_base64="X1y07k9ZmLbjfpS1W4qyfTp9Ng0=">AAAB+3icbVA9SwNBEJ3zM8avqKXNYhCtwl0UtLAI2FhG8JJAcoS9zV6yZHfv2N0TwnG/wVZrO7H1x1j6T9wkV5jEBwOP92aYmRcmnGnjut/O2vrG5tZ2aae8u7d/cFg5Om7pOFWE+iTmseqEWFPOJPUNM5x2EkWxCDlth+P7qd9+pkqzWD6ZSUIDgYeSRYxgYyW/Rzm/6Feqbs2dAa0SryBVKNDsV356g5ikgkpDONa667mJCTKsDCOc5uVeqmmCyRgPaddSiQXVQTY7NkfnVhmgKFa2pEEz9e9EhoXWExHaToHNSC97U/E/r5ua6DbImExSQyWZL4pSjkyMpp+jAVOUGD6xBBPF7K2IjLDCxNh8FraEIreZeMsJrJJWveZd1eqP19XGXZFOCU7hDC7BgxtowAM0wQcCDF7gFd6c3Hl3PpzPeeuaU8ycwAKcr18nCZUA</latexit>
`<latexitsha1_base64="bfqfNP2WZfAkR6k1rfkz29U6stI=">AAAB+nicbVA9SwNBEJ3zM8avqKXNYhCswl0UtLAI2FhGMB+QHGFvM0mW7O0du3tCOPMXbLW2E1v/jKX/xL3kCpP4YODx3gwz84JYcG1c99tZW9/Y3Nou7BR39/YPDktHx00dJYphg0UiUu2AahRcYsNwI7AdK6RhILAVjO8yv/WESvNIPppJjH5Ih5IPOKMmk7ooRK9UdivuDGSVeDkpQ456r/TT7UcsCVEaJqjWHc+NjZ9SZTgTOC12E40xZWM6xI6lkoao/XR265ScW6VPBpGyJQ2ZqX8nUhpqPQkD2xlSM9LLXib+53USM7jxUy7jxKBk80WDRBATkexx0ucKmRETSyhT3N5K2IgqyoyNZ2FLEE5tJt5yAqukWa14l5Xqw1W5dpunU4BTOIML8OAaanAPdWgAgxG8wCu8Oc/Ou/PhfM5b15x85gQW4Hz9AsMDlM8=</latexit>
15
(a) Theoretical matter cosmic variance contribution (D9) (b) Simulations
FIG. 5. (a) Theoretical matter cosmic variance contribution (D9) to the correlation of the unbinned power spectra of recon-
structed lensing potential and lensed temperature. The covariance (D9) is converted to a correlation using the same conversion
factor as in (38). (b) Measured correlation ˆcorrel( ˆC
ˆin
ˆin
L , ˆC
˜T ˜T
LT
ˆCT T
LT
) of the input lensing potential power and the di↵erence
of noise-free lensed and unlensed temperature powers in 1000 simulations.
FIG. 6. Left: The approximate contribution to the covariance between [L(L + 1)]2 ˆC
ˆˆ
L /(2⇡) and L0
(L0
+ 1) ˆC
˜T ˜T
L0 /(2⇡) from
cosmic variance of the lenses, derived from Eq. (D9). Right: The rank-one approximation to the matrix on the left from
retaining only the largest singular value.TODOO: discuss this in main text or appendix
CT
correlations in all calculations we try to eliminate correlations between the lensing potential and the unlensed
temperature in the simulations by considering
ˆcov( ˆC
ˆˆ
L , ˆC
˜T ˜T
L0 )|CV( ) = ˆcov( ˆC
ˆin
ˆin
L , ˆC
˜T ˜T
L0
ˆCT T
L0 ), (52)
where ˆC
ˆin
ˆin
is the empirical power of the input lensing potential and ˆCT T
is the empirical power of the unlensed
temperature. Subtracting the unlensed from the lensed empirical power spectrum also reduces the noise of the
covariance estimate because it eliminates the scatter due to cosmic variance of the unlensed temperature. We estimate
the covariance in simulations similarly to (48). As shown in Fig. 5b these measurements agree with the theoretical
expectation from (D9).
`0
<latexit sha1_base64="X1y07k9ZmLbjfpS1W4qyfTp9Ng0=">AAAB+3icbVA9SwNBEJ3zM8avqKXNYhCtwl0UtLAI2FhG8JJAcoS9zV6yZHfv2N0TwnG/wVZrO7H1x1j6T9wkV5jEBwOP92aYmRcmnGnjut/O2vrG5tZ2aae8u7d/cFg5Om7pOFWE+iTmseqEWFPOJPUNM5x2EkWxCDlth+P7qd9+pkqzWD6ZSUIDgYeSRYxgYyW/Rzm/6Feqbs2dAa0SryBVKNDsV356g5ikgkpDONa667mJCTKsDCOc5uVeqmmCyRgPaddSiQXVQTY7NkfnVhmgKFa2pEEz9e9EhoXWExHaToHNSC97U/E/r5ua6DbImExSQyWZL4pSjkyMpp+jAVOUGD6xBBPF7K2IjLDCxNh8FraEIreZeMsJrJJWveZd1eqP19XGXZFOCU7hDC7BgxtowAM0wQcCDF7gFd6c3Hl3PpzPeeuaU8ycwAKcr18nCZUA</latexit>
MS, Challinor et al. (2013)
Peloton, MS, et al. (2017)
Planck collab. (2013, 2018)
e.g. Sherwin & MS (2015)
Böhm, MS & Sherwin (2016)
Beck, Fabbian & Errard (2018)
Böhm, Sherwin et al. (2018)
Planck Collaboration: Gravitational lensing by large-scale structures with Planck
0.12 0.08 0.04 0.00 0.04
K
40
48
56
64
72
80
H0
0.30
0.35
0.40
0.45
0.50
0.55
0.60
0.65
0.70
0.75
Fig. 15. Two views of the geometric degeneracy in curved ⇤CDM models which is partially broken by lensing. Left: the degeneracy
in the ⌦m-⌦⇤ plane, with samples from Planck+WP+highL colour coded by the value of H0. The contours delimit the 68% and 95%
confidence regions, showing the further improvement from including the lensing likelihood. Right: the degeneracy in the ⌦K-H0
plane, with samples colour coded by ⌦⇤. Spatially-flat models lie along the grey dashed lines.
6.2. Correlation with the ISW Effect
As CMB photons travel to us from the last-scattering surface,
the gravitational potentials they traverse may undergo a non-
negligible amount of evolution. This produces a net redshift or
blueshift of the photons concerned, as they fall into and then
escape from the evolving potentials. The overall result is a con-
tribution to the CMB temperature anisotropy known as the late-
time ISW e↵ect, or the Rees-Sciama (R-S) e↵ect depending on
whether the evolution of the potentials concerned is in the linear
(ISW) or non-linear (R-S) regime of structure formation (Sachs
& Wolfe 1967; Rees & Sciama 1968). In the epoch of dark en-
ergy domination, which occurs after z ⇠ 0.5 for the concor-
dance ⇤CDM cosmology, large-scale potentials tend to decay
over time as space expands, resulting in a net blueshifting of the
CMB photons which traverse these potentials.
In the concordance ⇤CDM model, there is significant over-
lap between the large-scale structure which sources the CMB
lensing potential and the ISW e↵ect (greater than 90% at
L < 100), although it should be kept in mind that we cannot
observe the ISW component by itself, and so the e↵ective cor-
relation with the total CMB temperature is much smaller, on the
order of 20%.
The correlation between the lensing potential and the ISW
e↵ect results in a non-zero bispectrum or three-point function
for the observed CMB fluctuations. This bispectrum is peaked
for “squeezed” configurations, in which one short leg at low-`
Following Lewis et al. (2011), we begin with an estimator for
the cross-spectrum of the lensing potential and the ISW e↵ect as
ˆCT
L =
f 1
sky
2L + 1
X
M
ˆTLM ˆ⇤
LM, (45)
where ˆT`m = CTT
`
¯T`m is the Wiener-filtered temperature map
and ˆ is given in Eq. (13). In Fig. 16 we plot the measured cross-
spectra for our individual frequency reconstructions at 100, 143,
and 217 GHz as well as the MV reconstruction. We also plot the
mean and scatter expected in the fiducial ⇤CDM model.
To compare quantitatively the overall level of the measured
CT
L correlation to the value in ⇤CDM, we estimate an overall
amplitude for the cross-spectrum as
ˆAT
= NT
LmaxX
L=Lmin
(2L + 1)CT ,fid.
L
ˆCT
L /(CTT
L NL ). (46)
The overall normalization NT
is determined from Monte-Carlo
simulations. For our processing of the data, we find that it is well
approximated (at the 5% level) by the analytical approximation
NT
⇡
2
6666664
LmaxX
L=Lmin
(2L + 1)
⇣
CT ,fid.
L
⌘2
/(CTT
L NL )
3
7777775
1
. (47)
Curvature of space
Expansionrate
Amountofdarkenergy
Joint analysis
38. Science with galaxy catalogs
(1) Bias model at the field level
(2) Cosmological parameter analysis
(3) Accounting for skewness
(4) Getting initial from final conditions
13
Figure 1. 2-d slices of the overdensity h(x) of simulated 1010.8
1011.8
h 1
M halos (top), compared with the cubic bias
model (center), and the linear Standard Eulerian bias model (bottom). Each panel is 500 h 1
Mpc wide and 110 h 1
Mpc high,
and each density is smoothed with a R = 2 h 1
Mpc Gaussian, WR(k) = exp[ (kR)2
/2]. The colorbar indicates the values of
this smoothed overdensity h(x).
Figure 2. Same as Fig. 1 but for more massive and less abundant 1011.8
1012.8
h 1
M halos.
simulations. The variance, skewness and kurtosis of the densities shown in the histograms are listed in Table II for
the full simulated and modeled densities, and in Table III for the model error.
The linear Standard Eulerian bias model tends to underpredict troughs and overpredict peaks of the halo density,
as shown in Fig. 3. The model error is not Gaussian for any of the shown smoothing scales; in particular its kurtosis
is larger than 1 for all smoothing scales.
The cubic model provides a more accurate description of the halo density pdf, as shown in Fig. 4. This emphasizes
the importance of using nonlinear bias terms even on rather large scales. Still, the cubic model tends to underpredict
the peaks of the true halo density, especially on small scales. This agrees with Fig. 2 where the model also underpredicts
39. Science with galaxy catalogs
(1) Bias model at the field level
(2) Cosmological parameter analysis
(3) Accounting for skewness
(4) Getting initial from final conditions
13
Figure 1. 2-d slices of the overdensity h(x) of simulated 1010.8
1011.8
h 1
M halos (top), compared with the cubic bias
model (center), and the linear Standard Eulerian bias model (bottom). Each panel is 500 h 1
Mpc wide and 110 h 1
Mpc high,
and each density is smoothed with a R = 2 h 1
Mpc Gaussian, WR(k) = exp[ (kR)2
/2]. The colorbar indicates the values of
this smoothed overdensity h(x).
Figure 2. Same as Fig. 1 but for more massive and less abundant 1011.8
1012.8
h 1
M halos.
simulations. The variance, skewness and kurtosis of the densities shown in the histograms are listed in Table II for
the full simulated and modeled densities, and in Table III for the model error.
The linear Standard Eulerian bias model tends to underpredict troughs and overpredict peaks of the halo density,
as shown in Fig. 3. The model error is not Gaussian for any of the shown smoothing scales; in particular its kurtosis
is larger than 1 for all smoothing scales.
The cubic model provides a more accurate description of the halo density pdf, as shown in Fig. 4. This emphasizes
the importance of using nonlinear bias terms even on rather large scales. Still, the cubic model tends to underpredict
the peaks of the true halo density, especially on small scales. This agrees with Fig. 2 where the model also underpredicts
40. Overview
We calculate halo density field in PT and compare to simulations
1. How well does perturbative bias expansion work?
2. How correlated is the halo density field with the initial conditions?
3. What are the properties of the noise?
Joint work with Marko Simonović, Valentin Assassi & Matias Zaldarriaga
41. Overview
Most of the analyses use n-point functions. Disadvantages:
These questions have been extensively explored in the past
Desjacques, Jeong, Schmidt: Large-Scale Galaxy Bias
— Cosmic variance, compromise on resolution/size of the box
— At high k hard to disentangle different sources of nonlinearities
— Overfitting (smooth curves, many parameters)
— Only a few lowest n-point functions explored in practice
— Difficult to isolate and study the noise
42. Overview
Advantages:
Use fields rather than summary statistics
— No cosmic variance, small boxes with high resolution are sufficient
— High S/N at low k, no need to go to the very nonlinear regime
— No overfitting, each Fourier mode (amplitude and phase) is fitted
— “All” n-point functions measured simultaneously
— Easier to isolate and study the noise
Roth & Porciani (2011)
Baldauf, Schaan, Zaldarriaga (2015)
Lazeyras, Schmit (2017)
Abidi, Baldauf (2018)
McQuinn, D’Aloisio (2018)
Taruya, Nishimichi, Jeong (2018)
43. Overview
Same initial conditions
Sim
ulation
model<latexit sha1_base64="fJ3rOzfgxg04Png+ZlhcqA4B/Z4=">AAACCnicbVC7TsMwFHXKq5RXaEcWiwoJMVRJGehGJRgYi0QfUhNFjuO0Vu0ksh1EFfUP+AZWYGVDrHwAKyN/gtN2oC1HutK559yre3X8hFGpLOvbKKytb2xuFbdLO7t7+wfmYbkj41Rg0sYxi0XPR5IwGpG2ooqRXiII4j4jXX90lfvdeyIkjaM7NU6Iy9EgoiHFSGnJM8tOQJhCXuYIDnmsm4lnVq2aNQVcJfacVJuVxvVX5fWs5Zk/ThDjlJNIYYak7NtWotwMCUUxI5OSk0qSIDxCA9LXNEKcSDeb/j6BJ1oJYBgLXZGCU/XvRoa4lGPu60mO1FAue7n4n9dPVdhwMxolqSIRnh0KUwZVDPMgYEAFwYqNNUFYUP0rxEMkEFY6roUrPs8zsZcTWCWdes0+r9VvdTiXYIYiOALH4BTY4AI0wQ1ogTbA4AE8gWfwYjwab8a78TEbLRjznQpYgPH5C8f9naA=</latexit>
truth<latexit sha1_base64="IB7CHTpqOeRx7pTZPfheU1Oiklg=">AAACCnicbVC7TgJBFJ3FF+JrhdJmIjExFmQXC+kk0cISE3kksCGzwwATZnY3M3eNZMMf+A22amtnbP0AW0v/xFmgEPAkNzk5596cm+NHgmtwnG8rs7a+sbmV3c7t7O7tH9iH+YYOY0VZnYYiVC2faCZ4wOrAQbBWpBiRvmBNf3SV+s17pjQPgzsYR8yTZBDwPqcEjNS1850eE0C6SUdJDCqG4aRrF52SMwVeJe6cFKuFyvVX4fWs1rV/Or2QxpIFQAXRuu06EXgJUcCpYJNcJ9YsInREBqxtaEAk014y/X2CT4zSw/1QmQkAT9W/FwmRWo+lbzYlgaFe9lLxP68dQ7/iJTyIYmABnQX1Y4EhxGkRuMcVoyDGhhCquPkV0yFRhIKpayHFl2kn7nIDq6RRLrnnpfKtKecSzZBFR+gYnSIXXaAqukE1VEcUPaAn9IxerEfrzXq3PmarGWt+U0ALsD5/AQRancY=</latexit>
PT
m
odel
44. First predict dark matter (purple), then halos (yellow)
Modeling halo number density
45. Simple peak model
Whenever dark matter density > threshold, a halo forms
halo formation in peaks
Gaussian fluctuations on various scales
(described by the power spectrum)
Peak-Background Split
• Schematic Picture:
3
2
1
0
x
δc
Large Scale "Background"
Enhanced
"Peaks"
→→first sites of halo formation
W. Hu
first sites of snowfall
→
→
Halo forms
46. Simple peak model
Whenever dark matter density > threshold, a halo forms
halo formation in peaks
Gaussian fluctuations on various scales
(described by the power spectrum)
Peak-Background Split
• Schematic Picture:
3
2
1
0
x
δc
Large Scale "Background"
Enhanced
"Peaks"
→→first sites of halo formation
W. Hu
first sites of snowfall
→
→
Halo forms
halo formation in peaks
Gaussian fluctuations on various scales
(described by the power spectrum)
Peak-Background Split
• Schematic Picture:
3
2
1
0
x
δc
Large Scale "Background"
Enhanced
"Peaks"
→→first sites of halo formation
W. Hu
first sites of snowfall
→
→
Snow falls
47. Simple peak model
Whenever dark matter density > threshold, a halo forms
halo formation in peaks
Gaussian fluctuations on various scales
(described by the power spectrum)
Peak-Background Split
• Schematic Picture:
3
2
1
0
x
δc
Large Scale "Background"
Enhanced
"Peaks"
→→first sites of halo formation
W. Hu
first sites of snowfall
→
→
Halo forms
halo formation in peaks
Gaussian fluctuations on various scales
(described by the power spectrum)
Peak-Background Split
• Schematic Picture:
3
2
1
0
x
δc
Large Scale "Background"
Enhanced
"Peaks"
→→first sites of halo formation
W. Hu
first sites of snowfall
→
→
Snow falls
If we zoom out, halos trace the large scale background that
modulates the small-scale fluctuations and increases P(>threshold).
48. using fractional deviations from the mean
and proportionality constant (related to how likely a halo forms
when changing the threshold density)
Simple peak model
Tracing the large scale background means halo number density is
proportional to matter density, if both smoothed on large scales:
Write this as g(x) = b1 m(x)<latexit sha1_base64="XaHcr7zAchcoHCuF4bu18V4pL3I=">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</latexit>
g(x) ⌘
ng(x)
hngi
1
<latexit sha1_base64="zPWmnTIRWzX9q862OsLLcLaQHNU=">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</latexit>
m(x) ⌘
⇢m(x)
h⇢mi
1
<latexit sha1_base64="tVCTXJMrljgnXrmyqQ9zwu1iRlI=">AAACVXicbZBLTxsxFIWdgQJNaQl02Y1FVAkWjWZopXbRBVI3LEFqACkeRXc8dxILP6a2JxCN5tfxI6ouu2TbrivhJLMojytZOj7H19f+slIK5+P4VydaW3+xsbn1svtq+/Wbnd7u3rkzleU45EYae5mBQyk0Dr3wEi9Li6AyiRfZ1bdFfjFD64TR3/28xFTBRItCcPDBGvdSpvGaG6VA5zWb3TQ1U+CnWVHfNE2X5Sg9jNVBSA4pwx+VmFFWWOA1s1PTBqFHgp5IpCuT2eWu+ZCMe/14EC+LPhVJK/qkrdNx7zfLDa8Uas8lODdK4tKnNVgveLixyyqHJfArmOAoSA0KXVovMTT0fXByWhgblvZ06f7fUYNybq6ycHLxR/c4W5jPZaPKF1/SWuiy8qj5alBRSeoNXTClubDIvZwHAdyK8FbKpxAo+UD+wZRMNYFJ8pjAU3F+NEg+Do7OPvWPv7Z0tsg7sk8OSEI+k2NyQk7JkHByS+7IH/K387PzL1qPNlZHo07b85Y8qGjnHvnPuAU=</latexit>
ng(x) / ⇢m(x)<latexit sha1_base64="99dFa20NV5ubQOIpaYS4NMUSMrw=">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</latexit>
b1<latexit sha1_base64="vFL/SH1AeDkXA67O23grfojqbUQ=">AAAB+XicbVA9SwNBEJ31M8avqKXNYhCswl0UtLAI2FhGNB+QHGFvs5cs2d07dveEcOQn2GptJ7b+Gkv/iZvkCpP4YODx3gwz88JEcGM97xutrW9sbm0Xdoq7e/sHh6Wj46aJU01Zg8Yi1u2QGCa4Yg3LrWDtRDMiQ8Fa4ehu6reemTY8Vk92nLBAkoHiEafEOukx7Pm9UtmreDPgVeLnpAw56r3ST7cf01QyZakgxnR8L7FBRrTlVLBJsZsalhA6IgPWcVQRyUyQzU6d4HOn9HEUa1fK4pn6dyIj0pixDF2nJHZolr2p+J/XSW10E2RcJallis4XRanANsbTv3Gfa0atGDtCqObuVkyHRBNqXToLW0I5cZn4ywmskma14l9Wqg9X5dptnk4BTuEMLsCHa6jBPdShARQG8AKv8IYy9I4+0Oe8dQ3lMyewAPT1C5XxlB4=</latexit>
49. using fractional deviations from the mean
and proportionality constant (related to how likely a halo forms
when changing the threshold density)
Simple peak model
Tracing the large scale background means halo number density is
proportional to matter density, if both smoothed on large scales:
Write this as g(x) = b1 m(x)<latexit sha1_base64="XaHcr7zAchcoHCuF4bu18V4pL3I=">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</latexit>
g(x) ⌘
ng(x)
hngi
1
<latexit sha1_base64="zPWmnTIRWzX9q862OsLLcLaQHNU=">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</latexit>
m(x) ⌘
⇢m(x)
h⇢mi
1
<latexit sha1_base64="tVCTXJMrljgnXrmyqQ9zwu1iRlI=">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</latexit>
ng(x) / ⇢m(x)<latexit sha1_base64="99dFa20NV5ubQOIpaYS4NMUSMrw=">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</latexit>
b1<latexit sha1_base64="vFL/SH1AeDkXA67O23grfojqbUQ=">AAAB+XicbVA9SwNBEJ31M8avqKXNYhCswl0UtLAI2FhGNB+QHGFvs5cs2d07dveEcOQn2GptJ7b+Gkv/iZvkCpP4YODx3gwz88JEcGM97xutrW9sbm0Xdoq7e/sHh6Wj46aJU01Zg8Yi1u2QGCa4Yg3LrWDtRDMiQ8Fa4ehu6reemTY8Vk92nLBAkoHiEafEOukx7Pm9UtmreDPgVeLnpAw56r3ST7cf01QyZakgxnR8L7FBRrTlVLBJsZsalhA6IgPWcVQRyUyQzU6d4HOn9HEUa1fK4pn6dyIj0pixDF2nJHZolr2p+J/XSW10E2RcJallis4XRanANsbTv3Gfa0atGDtCqObuVkyHRBNqXToLW0I5cZn4ywmskma14l9Wqg9X5dptnk4BTuEMLsCHa6jBPdShARQG8AKv8IYy9I4+0Oe8dQ3lMyewAPT1C5XxlB4=</latexit>
linear regression
50. Simulations
log M[h 1
M ] ¯n [(h 1
Mpc) 3
] ¯n is comparable to
10.8 11.8 4.3 ⇥ 10 2
LSST [80, 81], Billion Object Apparatus [82]
11.8 12.8 5.7 ⇥ 10 3
SPHEREx [83, 84]
12.8 13.8 5.6 ⇥ 10 4
BOSS CMASS [85], DESI [86, 87], Euclid [88–90]
13.8 15.2 2.6 ⇥ 10 5
Cluster catalogs
Table I. Simulated halo populations at z = 0.6.
. The shifted squared density ˜2 and shifted tidal field ˜G2 are computed similarly, using 2
1(
e mass.
Ran 5 MP-Gadget1 DM-only N-body sims with 15363 DM particles,
30723 mesh for PM forces, L=500 Mpc/h,
~4000 time steps to evolve z=99 to z=0.6
4 FoF halo mass bins
1Feng et al. https://github.com/bluetides-project/MP-Gadget
[derived from P-Gadget]
mptcle = 2.9 ⇥ 109
M /h<latexit sha1_base64="5gOaYc51Lo066Ov3uoAnov8D0Bc=">AAACH3icbVDLSgMxFM3UV62vqks3wSK6kDpTBS0oFNy4ESrYB3TqkEnTNnQyGZI7Qhn6CX6E3+BW1+7EbZf+ieljYVsPBA7n3JuTHD8KuAbbHlqppeWV1bX0emZjc2t7J7u7V9UyVpRVqAykqvtEs4CHrAIcAlaPFCPCD1jN792O/NozU5rL8BH6EWsK0gl5m1MCRvKyx8JLXCVwBDRgg5tCvugCF0xjx34quqf3nitbEs66XjZn5+0x8CJxpiSHpih72R+3JWksWGhuJlo3HDuCZkIU8FFSxo01iwjtkQ5rGBoSE9pMxh8a4COjtHBbKnNCwGP170ZChNZ94ZtJQaCr572R+J/XiKF91Ux4GMXAQjoJascBBolH7eAWV4xC0DeEUMXNWzHtEkUomA5nUnwxMJ048w0skmoh75znCw8XudL1tJ00OkCH6AQ56BKV0B0qowqi6AW9oXf0Yb1an9aX9T0ZTVnTnX00A2v4CxZtokU=</latexit>
51. Test of linear model
Reasonable prediction on large scales
Missing some structure on small scales
MS, Simonović et al. (2019)
d slices of the overdensity h(x) of simulated 1010.8
1011.8
h 1
M halos (top), compared with the cubic b
), and the linear Standard Eulerian bias model (bottom). Each panel is 500 h 1
Mpc wide and 110 h 1
Mpc hi
sity is smoothed with a R = 2 h 1
Mpc Gaussian, WR(k) = exp[ (kR)2
/2]. The colorbar indicates the value
d slices of the overdensity h(x) of simulated 1010.8
1011.8
h 1
M halos (top), compared with the cubic b
), and the linear Standard Eulerian bias model (bottom). Each panel is 500 h 1
Mpc wide and 110 h 1
Mpc hi
sity is smoothed with a R = 2 h 1
Mpc Gaussian, WR(k) = exp[ (kR)2
/2]. The colorbar indicates the value
d overdensity h(x).
Simulation
(= truth)
Model b1 m(x)<latexit sha1_base64="qPIVubm7zi7B0APirS0IQLfxYzc=">AAACJXicbVBNS8NAEN34WetX1aOXYBH0YEmqoAcPBS8eK1gtNCFsNhNd3N2E3U1tCfkV/gh/g1c9exPBk/hP3LY5WPXBwOO9GWbmhSmjSjvOhzUzOze/sFhZqi6vrK6t1zY2r1SSSQIdkrBEdkOsgFEBHU01g24qAfOQwXV4dzbyr/sgFU3EpR6m4HN8I2hMCdZGCmoHnoB7knCORZR7/UGRexzr2zDOB0VRDQPXi4BpHPA9Y+4HtbrTcMaw/xK3JHVUoh3UvrwoIRkHoQnDSvVcJ9V+jqWmhEFR9TIFKSZ3+AZ6hgrMQfn5+K3C3jVKZMeJNCW0PVZ/TuSYKzXkoekc3ax+eyPxP6+X6fjEz6lIMw2CTBbFGbN1Yo8ysiMqgWg2NAQTSc2tNrnFEhNtkpzaEvLCZOL+TuAvuWo23MNG8+Ko3jot06mgbbSD9pCLjlELnaM26iCCHtATekYv1qP1ar1Z75PWGauc2UJTsD6/AT9/psc=</latexit>
g<latexit sha1_base64="wmzm750Tk9wCdwjQ/kl+uGkG6Vo=">AAACHXicbVC7TsMwFHV4lvIqMLJEVJWYqqQgwcBQiYWxSPQhNVXlODetVduJbKe0ivIFfATfwAozG2JFjPwJ7mOgLUeydHTOffn4MaNKO863tba+sbm1ndvJ7+7tHxwWjo4bKkokgTqJWCRbPlbAqIC6pppBK5aAuc+g6Q9uJ35zCFLRSDzocQwdjnuChpRgbaRuoeQJeCQR51gEqTccZanHse77YTrKsrwXANO42+sWik7ZmcJeJe6cFNEctW7hxwsiknAQmjCsVNt1Yt1JsdSUMDCDEwUxJgPcg7ahAnNQnXT6ncwuGSWww0iaJ7Q9Vf92pJgrNea+qZzcqpa9ifif1050eN1JqYgTDYLMFoUJs3VkT7KxAyqBaDY2BBNJza026WOJiTYJLmzxeWYycZcTWCWNStm9KFfuL4vVm3k6OXSKztA5ctEVqqI7VEN1RNATekGv6M16tt6tD+tzVrpmzXtO0AKsr1+F/aPk</latexit>
52. Nonlinear model
So far used linear model
g(x) = b1 m(x)<latexit sha1_base64="XaHcr7zAchcoHCuF4bu18V4pL3I=">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</latexit>
53. Nonlinear model
So far used linear model
Include all nonlinear terms allowed by symmetries (EFT)
Fit parameters by minimizing mean-squared error (least-squares
‘polynomial’ regression)
bi<latexit sha1_base64="stVaTDZz5GHd9HUTbI9z2gtgqYY=">AAACK3icbVDLSgMxFM34tr7aunQTLIKrMqOCIi4ENy4rWhU6pSSZTBvMY0gytSXMJ7jVr/BrXClu/Q/T2oWtHggczr03956DM86MDcP3YG5+YXFpeWW1tLa+sblVrlRvjco1oU2iuNL3GBnKmaRNyyyn95mmSGBO7/DDxah+16faMCVv7DCjbYG6kqWMIOula9xhnXItrIdjwL8kmpAamKDRqQTVOFEkF1RawpExrSjMbNshbRnhtCjFuaEZIg+oS1ueSiSoabvxrQXc80oCU6X9kxaO1d8TDgljhgL7ToFsz8zWRuJ/tVZu05O2YzLLLZXkZ1Gac2gVHBmHCdOUWD70BBHN/K2Q9JBGxPp4SrGkj0QJgWTi4v6gcC4ercCpGxTFlCWHhf8cCnMKseLJ2IYp+RSj2cz+ktuDenRYP7g6qp2fTfJcATtgF+yDCByDc3AJGqAJCOiCJ/AMXoLX4C34CD5/WueCycw2mELw9Q0dSKcW</latexit>
MS, Simonović et al. (2019)
g(x) = b1 m(x) + b2
2
m(x) + tidal term + b3
3
m(x) + · · ·<latexit sha1_base64="hIhHT6q70TfpTrAHD6PG9mgELPM=">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</latexit>
54. Bulk flows
5
s x of a halo at the initial position q are given by x = q + (q). The overdensity
1 + h(x) =
Z
d3
q (1 + h(q)) D(x q (q)) , (12)
Fourier transform of this field in Eulerian space is
d3
x (1 + h(x)) e ik·x
=
Z
d3
q (1 + h(q)) e ik·(q+ (q))
. (13)
d in the rest of the paper we restrict the range of momenta to k 6= 0, so that the zero
er our formulas. The nonlinear displacement from Lagrangian to Eulerian position
ve series = 1 + 2 + · · · . At first order, we have the well-known Zel’dovich
1(q) =
Z
k
eik·q ik
k2 1(k) . (14)
n be written as
2(q) =
3
14
Z
k
eik·q ik
k2
G2(k) . (15)
of the nonlinear displacement field and expanding the exponent e ik· (q)
in Eq. (13)
Standard Eulerian bias expansion. This procedure also fixes the relation between
linear displacement is large
5
rdinates x of a halo at the initial position q are given by x = q + (q). The overdensity
en by
1 + h(x) =
Z
d3
q (1 + h(q)) D(x q (q)) , (12)
. The Fourier transform of this field in Eulerian space is
k) ⌘
Z
d3
x (1 + h(x)) e ik·x
=
Z
d3
q (1 + h(q)) e ik·(q+ (q))
. (13)
ion and in the rest of the paper we restrict the range of momenta to k 6= 0, so that the zero
not enter our formulas. The nonlinear displacement from Lagrangian to Eulerian position
urbative series = 1 + 2 + · · · . At first order, we have the well-known Zel’dovich
1(q) =
Z
k
eik·q ik
k2 1(k) . (14)
ment can be written as
2(q) =
3
14
Z
k
eik·q ik
k2
G2(k) . (15)
ription of the nonlinear displacement field and expanding the exponent e ik· (q)
in Eq. (13)
usual Standard Eulerian bias expansion. This procedure also fixes the relation between
s and their Standard Eulerian counterparts. Of course, this is not a surprise, as we expect
ee order by order in perturbation theory.
5
f a halo at the initial position q are given by x = q + (q). The overdensity
h(x) =
Z
d3
q (1 + h(q)) D(x q (q)) , (12)
er transform of this field in Eulerian space is
x (1 + h(x)) e ik·x
=
Z
d3
q (1 + h(q)) e ik·(q+ (q))
. (13)
he rest of the paper we restrict the range of momenta to k 6= 0, so that the zero
ur formulas. The nonlinear displacement from Lagrangian to Eulerian position
ries = 1 + 2 + · · · . At first order, we have the well-known Zel’dovich
1(q) =
Z
k
eik·q ik
k2 1(k) . (14)
written as
2(q) =
3
14
Z
k
eik·q ik
k2
G2(k) . (15)
e nonlinear displacement field and expanding the exponent e ik· (q)
in Eq. (13)
Lagrangian space
Eulerian space
halo
5
t the initial position q are given by x = q + (q). The overdensity
d3
q (1 + h(q)) D(x q (q)) , (12)
rm of this field in Eulerian space is
)) e ik·x
=
Z
d3
q (1 + h(q)) e ik·(q+ (q))
. (13)
the paper we restrict the range of momenta to k 6= 0, so that the zero
s. The nonlinear displacement from Lagrangian to Eulerian position
1 + 2 + · · · . At first order, we have the well-known Zel’dovich
q) =
Z
k
eik·q ik
k2 1(k) . (14)
Slide credit: Marko Simonović
55. Large bulk flows lead to large model error
because fields are incoherent
Need a model that takes into account large bulk flows
Bulk flows
Good model of
x ! x +<latexit sha1_base64="+3MORj4q/XkrwqAejYnO59kK720=">AAACAHicbVBNS8NAEJ34WetX1IMHL4tFEISSVEGhBwtePFawH9CEstlu2qWbTdjdaEvpxb/ixYMiXv0Z3vw3btsctPXBwOO9GWbmBQlnSjvOt7W0vLK6tp7byG9ube/s2nv7dRWnktAaiXksmwFWlDNBa5ppTpuJpDgKOG0E/ZuJ33igUrFY3OthQv0IdwULGcHaSG37cOCVPcm6PY2ljB+9MhqceVXF2nbBKTpToEXiZqQAGapt+8vrxCSNqNCEY6VarpNof4SlZoTTcd5LFU0w6eMubRkqcESVP5o+MEYnRumgMJamhEZT9ffECEdKDaPAdEZY99S8NxH/81qpDq/8ERNJqqkgs0VhypGO0SQN1GGSEs2HhmAimbkVkR6WmGiTWd6E4M6/vEjqpaJ7XizdXRQq11kcOTiCYzgFFy6hArdQhRoQGMMzvMKb9WS9WO/Wx6x1ycpmDuAPrM8fPGuWKA==</latexit>
Displace by a shifttruth<latexit sha1_base64="dorPueCGyAIyn+LUjl5+vXhBTBQ=">AAAB+3icbVBNS8NAEN34WetXrEcvi0XwVJIq6M2CF48V7Ac0IWw2m3bpZhN2J2IJ/StePCji1T/izX/jts1BWx8MPN6bYWZemAmuwXG+rbX1jc2t7cpOdXdv/+DQPqp1dZoryjo0Fanqh0QzwSXrAAfB+pliJAkF64Xj25nfe2RK81Q+wCRjfkKGksecEjBSYNe8iAkgQeGpBIPKYTQN7LrTcObAq8QtSR2VaAf2lxelNE+YBCqI1gPXycAviAJOBZtWvVyzjNAxGbKBoZIkTPvF/PYpPjNKhONUmZKA5+rviYIkWk+S0HQmBEZ62ZuJ/3mDHOJrv+Ayy4FJulgU5wJDimdB4IgrRkFMDCFUcXMrpiOiCAUTV9WE4C6/vEq6zYZ70WjeX9ZbN2UcFXSCTtE5ctEVaqE71EYdRNETekav6M2aWi/Wu/WxaF2zyplj9AfW5w9uWJSv</latexit>
truth<latexit sha1_base64="dorPueCGyAIyn+LUjl5+vXhBTBQ=">AAAB+3icbVBNS8NAEN34WetXrEcvi0XwVJIq6M2CF48V7Ac0IWw2m3bpZhN2J2IJ/StePCji1T/izX/jts1BWx8MPN6bYWZemAmuwXG+rbX1jc2t7cpOdXdv/+DQPqp1dZoryjo0Fanqh0QzwSXrAAfB+pliJAkF64Xj25nfe2RK81Q+wCRjfkKGksecEjBSYNe8iAkgQeGpBIPKYTQN7LrTcObAq8QtSR2VaAf2lxelNE+YBCqI1gPXycAviAJOBZtWvVyzjNAxGbKBoZIkTPvF/PYpPjNKhONUmZKA5+rviYIkWk+S0HQmBEZ62ZuJ/3mDHOJrv+Ayy4FJulgU5wJDimdB4IgrRkFMDCFUcXMrpiOiCAUTV9WE4C6/vEq6zYZ70WjeX9ZbN2UcFXSCTtE5ctEVaqE71EYdRNETekav6M2aWi/Wu/WxaF2zyplj9AfW5w9uWJSv</latexit>
<latexit sha1_base64="0kQQ4bQnE6qcjjlOoHqoSjdsqzk=">AAAB63icbVBNSwMxEJ34WetX1aOXYBE8ld0q6M2CF48V7Ae0S8mm2TY0yS5JVihL/4IXD4p49Q9589+YbfegrQ8GHu/NMDMvTAQ31vO+0dr6xubWdmmnvLu3f3BYOTpumzjVlLVoLGLdDYlhgivWstwK1k00IzIUrBNO7nK/88S04bF6tNOEBZKMFI84JTaX+k3DB5WqV/PmwKvEL0gVCjQHla/+MKapZMpSQYzp+V5ig4xoy6lgs3I/NSwhdEJGrOeoIpKZIJvfOsPnThniKNaulMVz9fdERqQxUxm6Tkns2Cx7ufif10ttdBNkXCWpZYouFkWpwDbG+eN4yDWjVkwdIVRzdyumY6IJtS6esgvBX355lbTrNf+yVn+4qjZuizhKcApncAE+XEMD7qEJLaAwhmd4hTck0Qt6Rx+L1jVUzJzAH6DPH/PXjis=</latexit>
✏(x) ⌘ truth model<latexit sha1_base64="j8JWHHDkGfzWSFy/jB9sc9v3ags=">AAACOXicbVDLSgNBEJz1Gd9Rj14Gg6AHw64KevAQ8eJRwaiQDWF20jGD81hnesWw5Hv8CL/Bq+LRm3j1B5yNOWi0YKC6qpvuqSSVwmEYvgZj4xOTU9Olmdm5+YXFpfLyyoUzmeVQ50Yae5UwB1JoqKNACVepBaYSCZfJzXHhX96BdcLoc+yl0FTsWouO4Ay91CofxZA6IY3evN+iMdxm4o7SuA0SWStWDLtW5Wgz7Pbp9qiujK/7rXIlrIYD0L8kGpIKGeK0VX6L24ZnCjRyyZxrRGGKzZxZFFxCfzbOHKSM37BraHiqmQLXzAdf7dMNr7Rpx1j/NNKB+nMiZ8q5nkp8Z3GlG/UK8T+vkWHnoJkLnWYImn8v6mSSoqFFbrQtLHCUPU8Yt8LfSnmXWcbRp/trS6KKTKLRBP6Si51qtFvdOdur1A6H6ZTIGlknmyQi+6RGTsgpqRNOHsgTeSYvwWPwFrwHH9+tY8FwZpX8QvD5BQAFrsc=</latexit>
56. Including bulk flows in the model
licity, in this equation and in the rest of the paper we restrict the range of momenta to k 6= 0, so that the ze
r mean density do not enter our formulas. The nonlinear displacement from Lagrangian to Eulerian positi
xpanded in a perturbative series = 1 + 2 + · · · . At first order, we have the well-known Zel’dovi
mation [69]
1(q) =
Z
k
eik·q ik
k2 1(k) . (1
nd-order displacement can be written as
2(q) =
3
14
Z
k
eik·q ik
k2
G2(k) . (1
e perturbative description of the nonlinear displacement field and expanding the exponent e ik· (q)
in Eq. (1
sible to recover the usual Standard Eulerian bias expansion. This procedure also fixes the relation betwe
an bias parameters and their Standard Eulerian counterparts. Of course, this is not a surprise, as we expe
descriptions to agree order by order in perturbation theory.
e other hand we do not want to expand the full nonlinear displacement. We are going to keep the large
q) exponentiated and expand only the higher-order terms.3
In this way the largest part of the problema
acements is not expanded in perturbation theory. With this in mind, we can rewrite Eq. (13) in the followi
h(k) =
Z
d3
q
⇣
1 + bL
1 1(q) + bL
2 ( 2
1(q) 2
1) + bL
G2
G2(q) + · · ·
ik · 2(q) + · · ·
⌘
e ik·(q+ 1(q))
, (1
e new contributions come from expanding the second (and higher) order displacement field in the exponen
ortant to stress that at leading order this new term can be expressed through the second order operator
(15)). Therefore, at second order in perturbation theory, expanding the nonlinear terms in the displaceme
) only shifts some of the standard Lagrangian bias parameters by a calculable constant. We will give mo
Usual approximation in (C)LPT for example: Vlah, Castorina, White (2016)
art with the description of biased tracers in Lagrangian space. The displacement field is then
t linear contribution and smaller higher order corrections. The nonlinear corrections to are
while the linear piece is kept in the exponent. In this way, the dominant part of the large
treated exactly, and the resulting operators once written in Eulerian space are automatically
est of this section we give the details of this construction.
ity field at Lagrangian position q is modeled using a bias expansion in the linear Lagrangian-
L
h(q) = bL
1 1(q) + bL
2 ( 2
1(q) 2
1) + bL
G2
G2(q) + · · · , (8)
re Lagrangian bias parameters, 2
1 is the r.m.s. fluctuation of the linear density field
2
1 =
⌦ 2
1(q)
↵
=
Z 1
0
dk
2⇡2
k2
P11(k) , (9)
) is defined as2
G2(q) ⌘
@i@j
@2 1(q)
2
2
1(q) . (10)
this operator in momentum space is given by
G2(k) =
Z
p
(p · (k p))2
p2|k p|2
1 1(p) 1(k p) . (11)
ng notation in which
R
p
⌘
R
d3
p/(2⇡)3
. In the bias expansion (8) we kept only terms up to
bation theory. We will continue to work at this order throughout this section, because it is
ng notation and motivating the bias model that we are going to use to make comparisons with
er order or higher derivative operators needed for the consistent one-loop calculation can be
ded. We will come back to this in Section VIII.
n Eq. (8) is in Lagrangian space. In order to go to Eulerian space, let us start from Eq. (8) and
split into the dominant linear contribution and smaller higher order corrections. The nonlinear corrections
treated perturbatively, while the linear piece is kept in the exponent. In this way, the dominant part of
displacements can be treated exactly, and the resulting operators once written in Eulerian space are auto
IR-resummed. In the rest of this section we give the details of this construction.
The proto-halo density field at Lagrangian position q is modeled using a bias expansion in the linear La
space density 1(q):
L
h(q) = bL
1 1(q) + bL
2 ( 2
1(q) 2
1) + bL
G2
G2(q) + · · · ,
where bL
1, bL
2, bL
G2
, . . . are Lagrangian bias parameters, 2
1 is the r.m.s. fluctuation of the linear density field
2
1 =
⌦ 2
1(q)
↵
=
Z 1
0
dk
2⇡2
k2
P11(k) ,
and the operator G2(q) is defined as2
G2(q) ⌘
@i@j
@2 1(q)
2
2
1(q) .
The representation of this operator in momentum space is given by
G2(k) =
Z
p
(p · (k p))2
p2|k p|2
1 1(p) 1(k p) .
Notice that we are using notation in which
R
p
⌘
R
d3
p/(2⇡)3
. In the bias expansion (8) we kept only te
second order in perturbation theory. We will continue to work at this order throughout this section, bec
su cient for introducing notation and motivating the bias model that we are going to use to make compar
simulations. The higher order or higher derivative operators needed for the consistent one-loop calculati
straightforwardly included. We will come back to this in Section VIII.
The bias expansion in Eq. (8) is in Lagrangian space. In order to go to Eulerian space, let us start from E
include the gravitational evolution. The gravitational evolution is encoded in the nonlinear displacement fi
5
Eulerian coordinates x of a halo at the initial position q are given by x = q + (q). The overdensity
is way is given by
1 + h(x) =
Z
d3
q (1 + h(q)) D(x q (q)) , (12)
Dirac delta. The Fourier transform of this field in Eulerian space is
h(k) ⌘
Z
d3
x (1 + h(x)) e ik·x
=
Z
d3
q (1 + h(q)) e ik·(q+ (q))
. (13)
n this equation and in the rest of the paper we restrict the range of momenta to k 6= 0, so that the zero
density do not enter our formulas. The nonlinear displacement from Lagrangian to Eulerian position
ed in a perturbative series = 1 + 2 + · · · . At first order, we have the well-known Zel’dovich
[69]
1(q) =
Z
k
eik·q ik
k2 1(k) . (14)
er displacement can be written as
2(q) =
3
14
Z
k
eik·q ik
k2
G2(k) . (15)
ik· (q)
57. Shifted operators
Z(k) = ˜1(k) +
1
2
˜G2(k)
1
3
˜G3(k) + · · · , (20)
or analogous to ˜G2 (see Appendix D). In other words, Z(k) can be absorbed in the bias
ging the bias parameters. Of course, this is just a choice, and there is nothing wrong in
he formulas. As we are going to see later, di↵erent choices may be more appropriate for
us point out that in the formula (20) the displacements 1(q) are treated exactly. In other
k· 1(q)
is never expanded in 1(q). The only expansion parameter is the derivative of the
1(q), which is a small quantity.5
This is consistent with the way the shifted operators
asis of shifted operators we can write the bias expansion of the halo density field in Eulerian
order in perturbation theory, in the following way
h(k) = b1
˜1(k) + b2
˜2(k) + bG2
˜G2(k) + · · · . (21)
this section. Notice that the new bias parameters bi di↵er from the original Lagrangian
his di↵erence comes from expanding the nonlinear part of the displacement (Eq. (16)) and
sity field in terms of shifted operators (Eq. (20)). We give the explicit relation of bi and bL
i
(21) has a similar structure as the usual Standard Eulerian bias expansion
h(k) = bE
1 (k) + bE
2 2(k) + bE
G2
G2(k) + · · · , (22)
ce that all fields in this equation are nonlinear. Apart from the IR resummation of the large
ant di↵erence compared to the expansion in terms of ˜O is that in Eq. (21) all operators are
inear field 1. As we are going to see, for the purposes of describing the biased tracers on
her important virtue of the expansion (21).
+ noise
ik · 2(q) + · · · e , (16)
me from expanding the second (and higher) order displacement field in the exponent.
t leading order this new term can be expressed through the second order operator G2
cond order in perturbation theory, expanding the nonlinear terms in the displacement
he standard Lagrangian bias parameters by a calculable constant. We will give more
s in Section VIII.
tivates us to write down the bias expansion in Eulerian space in terms of shifted
he following way
˜O(k) ⌘
Z
d3
q O(q) e ik·(q+ 1(q))
, (17)
2
1), G2, . . .}.4
We would like to stress again a few important advantages that this
d operators are written in Eulerian space and therefore allow for easy comparisons
tion of their importance. (b) The large displacement terms 1(q) are kept resummed,
s with simulations on the level of realizations. Notice that this also implies that in
ass filter, compared to the wavelength of a Fourier mode 1(k). For a given wavenumber k, the
nto the long-wavelength and short-wavelength part: 1 = L
1 + S
1 , where L
1 = W(k) 1 and
L
1 on the short modes is fixed by the Equivalence Principle. Therefore, strictly speaking, only L
1
in any perturbative calculation S
1 has to be expanded order by order in perturbation theory. The
y keeping the full 1 in the exponent is always higher order in S
1 than terms we calculate. Also,
all scales. In order to keep the formulas simple, we decide not to do the long-short splitting in our
not just given by a translation of the position argument because they implicitly include the inverse
n @xi/@qj due to the coordinate transformation. This is similar to the Zel’dovich density, which is
gian space shifted by 1(q).
Motivates bias expansion in “shifted” operators (incl. bulk flows)
58. Shifted operators
Z(k) = ˜1(k) +
1
2
˜G2(k)
1
3
˜G3(k) + · · · , (20)
or analogous to ˜G2 (see Appendix D). In other words, Z(k) can be absorbed in the bias
ging the bias parameters. Of course, this is just a choice, and there is nothing wrong in
he formulas. As we are going to see later, di↵erent choices may be more appropriate for
us point out that in the formula (20) the displacements 1(q) are treated exactly. In other
k· 1(q)
is never expanded in 1(q). The only expansion parameter is the derivative of the
1(q), which is a small quantity.5
This is consistent with the way the shifted operators
asis of shifted operators we can write the bias expansion of the halo density field in Eulerian
order in perturbation theory, in the following way
h(k) = b1
˜1(k) + b2
˜2(k) + bG2
˜G2(k) + · · · . (21)
this section. Notice that the new bias parameters bi di↵er from the original Lagrangian
his di↵erence comes from expanding the nonlinear part of the displacement (Eq. (16)) and
sity field in terms of shifted operators (Eq. (20)). We give the explicit relation of bi and bL
i
(21) has a similar structure as the usual Standard Eulerian bias expansion
h(k) = bE
1 (k) + bE
2 2(k) + bE
G2
G2(k) + · · · , (22)
ce that all fields in this equation are nonlinear. Apart from the IR resummation of the large
ant di↵erence compared to the expansion in terms of ˜O is that in Eq. (21) all operators are
inear field 1. As we are going to see, for the purposes of describing the biased tracers on
her important virtue of the expansion (21).
+ noise
ik · 2(q) + · · · e , (16)
me from expanding the second (and higher) order displacement field in the exponent.
t leading order this new term can be expressed through the second order operator G2
cond order in perturbation theory, expanding the nonlinear terms in the displacement
he standard Lagrangian bias parameters by a calculable constant. We will give more
s in Section VIII.
tivates us to write down the bias expansion in Eulerian space in terms of shifted
he following way
˜O(k) ⌘
Z
d3
q O(q) e ik·(q+ 1(q))
, (17)
2
1), G2, . . .}.4
We would like to stress again a few important advantages that this
d operators are written in Eulerian space and therefore allow for easy comparisons
tion of their importance. (b) The large displacement terms 1(q) are kept resummed,
s with simulations on the level of realizations. Notice that this also implies that in
ass filter, compared to the wavelength of a Fourier mode 1(k). For a given wavenumber k, the
nto the long-wavelength and short-wavelength part: 1 = L
1 + S
1 , where L
1 = W(k) 1 and
L
1 on the short modes is fixed by the Equivalence Principle. Therefore, strictly speaking, only L
1
in any perturbative calculation S
1 has to be expanded order by order in perturbation theory. The
y keeping the full 1 in the exponent is always higher order in S
1 than terms we calculate. Also,
all scales. In order to keep the formulas simple, we decide not to do the long-short splitting in our
not just given by a translation of the position argument because they implicitly include the inverse
n @xi/@qj due to the coordinate transformation. This is similar to the Zel’dovich density, which is
gian space shifted by 1(q).
Motivates bias expansion in “shifted” operators (incl. bulk flows)
Result is in Eulerian space so easy to compare against simulations
Has IR resummation, giving correct halo positions and BAO spread
Model is perturbative, only linear fields used in the construction
Power spectrum agrees with resummed 1-loop PT
59. Model on the grid
Distribute 15363 particles on regular grid
Assign artificial particle masses
Displace by linear displacement
Interpolate to Eulerian grid using CIC weighted by particle masses
[Very similar to generating N-body initial conds./Zeldovich density]
mi = O(qi)<latexit sha1_base64="OIMXXFzMNCtfp4mc7I6Q6k7Bu8E=">AAACF3icbZC7TsMwFIadcivlFmBkwKJCKkuVFCRYEJVY2CgSvUhtFTmu01q1k2A7SFWUkYfgGVhhZkOsjIy8CU6agbb8kqVP/zlH5/h3Q0alsqxvo7C0vLK6VlwvbWxube+Yu3stGUQCkyYOWCA6LpKEUZ80FVWMdEJBEHcZabvj67TefiRC0sC/V5OQ9Dka+tSjGCltOeYhdyi8hD2O1AgjFt8mlYxdL35IHHrimGWramWCi2DnUAa5Go750xsEOOLEV5ghKbu2Fap+jISimJGk1IskCREeoyHpavQRJ7IfZx9J4LF2BtALhH6+gpn7dyJGXMoJd3VneqScr6Xmf7VupLyLfkz9MFLEx9NFXsSgCmCaChxQQbBiEw0IC6pvhXiEBMJKZzezxeWJzsSeT2ARWrWqfVqt3Z2V61d5OkVwAI5ABdjgHNTBDWiAJsDgCbyAV/BmPBvvxofxOW0tGPnMPpiR8fULYv6f9w==</latexit>
q<latexit sha1_base64="Ft5o3g7n3eKWwngUx/MdJMYDuiE=">AAACAHicbVDLSgMxFL3js9ZX1aWbYBFclZkq6M6CG5cV7APboWTSTBuaZMYkI5ShG7/Bra7diVv/xKV/YqadhW09EDiccy/35AQxZ9q47rezsrq2vrFZ2Cpu7+zu7ZcODps6ShShDRLxSLUDrClnkjYMM5y2Y0WxCDhtBaObzG89UaVZJO/NOKa+wAPJQkawsdJDV2AzDML0cdIrld2KOwVaJl5OypCj3iv9dPsRSQSVhnCsdcdzY+OnWBlGOJ0Uu4mmMSYjPKAdSyUWVPvpNPEEnVqlj8JI2ScNmqp/N1IstB6LwE5mCfWil4n/eZ3EhFd+ymScGCrJ7FCYcGQilH0f9ZmixPCxJZgoZrMiMsQKE2NLmrsSiKwTb7GBZdKsVrzzSvXuoly7ztspwDGcwBl4cAk1uIU6NICAhBd4hTfn2Xl3PpzP2eiKk+8cwRycr1/QxZes</latexit>
q ! q + 1(q)<latexit sha1_base64="WM5LTA6NvvqFh353BeqMK/OW1Ew=">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</latexit>
1(q)<latexit sha1_base64="TJaAHFoYRyspIjBVZgsQyBThdy0=">AAACKXicbVDLSsNAFJ34rPVVdekmWIS6KUkVdGfBjcsK9gFNKJPppB06j3RmUikh3+FH+A1ude1O3Yk/4rSNYFsPDBzOuZdz5wQRJUo7zoe1srq2vrGZ28pv7+zu7RcODhtKxBLhOhJUyFYAFaaE47ommuJWJDFkAcXNYHAz8ZsjLBUR/F6PI+wz2OMkJAhqI3UKrsfxAxKMQd5NvNEwTTwGdT8Ik2Ga5n+5V1Mk7bglM3DWKRSdsjOFvUzcjBRBhlqn8OV1BYoZ5hpRqFTbdSLtJ1Bqgig2IbHCEUQD2MNtQzlkWPnJ9GupfWqUrh0KaR7X9lT9u5FAptSYBWZycqta9Cbif1471uGVnxAexRpzNAsKY2prYU96srtEYqTp2BCIJDG32qgPJUTatDmXErDUdOIuNrBMGpWye16u3F0Uq9dZOzlwDE5ACbjgElTBLaiBOkDgETyDF/BqPVlv1rv1ORtdsbKdIzAH6/sHurmolw==</latexit>
O(q)<latexit sha1_base64="fNNlinCcTxszFIVZ7rBkeZk8/5Y=">AAACJXicbVDLSgMxFM3UV62vqks3g0WoC8tMFXRnwY07K9gHtEPJZDJtaB7TJFMpw3yFH+E3uNW1OxFciX9i+ljY1gOBwzn3cm6OH1GitON8WZmV1bX1jexmbmt7Z3cvv39QVyKWCNeQoEI2fagwJRzXNNEUNyOJIfMpbvj9m7HfGGKpiOAPehRhj8EuJyFBUBupkz9rc/yIBGOQB0l7OEiTNoO654fJIE1zE44gTe7SojFPO/mCU3ImsJeJOyMFMEO1k/9pBwLFDHONKFSq5TqR9hIoNUEUm4BY4QiiPuzilqEcMqy8ZPKt1D4xSmCHQprHtT1R/24kkCk1Yr6ZHN+pFr2x+J/XinV45SWER7HGHE2DwpjaWtjjjuyASIw0HRkCkSTmVhv1oIRImybnUnyWmk7cxQaWSb1ccs9L5fuLQuV61k4WHIFjUAQuuAQVcAuqoAYQeAIv4BW8Wc/Wu/VhfU5HM9Zs5xDMwfr+Ba4ppw4=</latexit>
˜O(x)<latexit sha1_base64="jp0htYvlhK/ioIFAQhH9fbs5T6c=">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</latexit>
60. d slices of the overdensity h(x) of simulated 1010.8
1011.8
h 1
M halos (top), compared with the cubic b
), and the linear Standard Eulerian bias model (bottom). Each panel is 500 h 1
Mpc wide and 110 h 1
Mpc hi
sity is smoothed with a R = 2 h 1
Mpc Gaussian, WR(k) = exp[ (kR)2
/2]. The colorbar indicates the value
d overdensity h(x).
g<latexit sha1_base64="wmzm750Tk9wCdwjQ/kl+uGkG6Vo=">AAACHXicbVC7TsMwFHV4lvIqMLJEVJWYqqQgwcBQiYWxSPQhNVXlODetVduJbKe0ivIFfATfwAozG2JFjPwJ7mOgLUeydHTOffn4MaNKO863tba+sbm1ndvJ7+7tHxwWjo4bKkokgTqJWCRbPlbAqIC6pppBK5aAuc+g6Q9uJ35zCFLRSDzocQwdjnuChpRgbaRuoeQJeCQR51gEqTccZanHse77YTrKsrwXANO42+sWik7ZmcJeJe6cFNEctW7hxwsiknAQmjCsVNt1Yt1JsdSUMDCDEwUxJgPcg7ahAnNQnXT6ncwuGSWww0iaJ7Q9Vf92pJgrNea+qZzcqpa9ifif1050eN1JqYgTDYLMFoUJs3VkT7KxAyqBaDY2BBNJza026WOJiTYJLmzxeWYycZcTWCWNStm9KFfuL4vVm3k6OXSKztA5ctEVqqI7VEN1RNATekGv6M16tt6tD+tzVrpmzXtO0AKsr1+F/aPk</latexit>
Test of nonlinear model
MS, Simonović et al. (2019)
b1 m(x)<latexit sha1_base64="qPIVubm7zi7B0APirS0IQLfxYzc=">AAACJXicbVBNS8NAEN34WetX1aOXYBH0YEmqoAcPBS8eK1gtNCFsNhNd3N2E3U1tCfkV/gh/g1c9exPBk/hP3LY5WPXBwOO9GWbmhSmjSjvOhzUzOze/sFhZqi6vrK6t1zY2r1SSSQIdkrBEdkOsgFEBHU01g24qAfOQwXV4dzbyr/sgFU3EpR6m4HN8I2hMCdZGCmoHnoB7knCORZR7/UGRexzr2zDOB0VRDQPXi4BpHPA9Y+4HtbrTcMaw/xK3JHVUoh3UvrwoIRkHoQnDSvVcJ9V+jqWmhEFR9TIFKSZ3+AZ6hgrMQfn5+K3C3jVKZMeJNCW0PVZ/TuSYKzXkoekc3ax+eyPxP6+X6fjEz6lIMw2CTBbFGbN1Yo8ysiMqgWg2NAQTSc2tNrnFEhNtkpzaEvLCZOL+TuAvuWo23MNG8+Ko3jot06mgbbSD9pCLjlELnaM26iCCHtATekYv1qP1ar1Z75PWGauc2UJTsD6/AT9/psc=</latexit>
sim
g (x)<latexit sha1_base64="dAEZhBUIThUpHuyxiN4tqPX1ous=">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</latexit>
d slices of the overdensity h(x) of simulated 1010.8
1011.8
h 1
M halos (top), compared with the cubic b
), and the linear Standard Eulerian bias model (bottom). Each panel is 500 h 1
Mpc wide and 110 h 1
Mpc hi
sity is smoothed with a R = 2 h 1
Mpc Gaussian, WR(k) = exp[ (kR)2
/2]. The colorbar indicates the value
Nonlinear
model
d slices of the overdensity h(x) of simulated 1010.8
1011.8
h 1
M halos (top), compared with the cubic b
), and the linear Standard Eulerian bias model (bottom). Each panel is 500 h 1
Mpc wide and 110 h 1
Mpc hi
sity is smoothed with a R = 2 h 1
Mpc Gaussian, WR(k) = exp[ (kR)2
/2]. The colorbar indicates the value
Much better agreement than previous model
Simulation
(= truth)
61. Measures of success
Error power spectrum (= MSE per wavenumber k)
Cross-correlation between model and truth
rcc(k) =
h truth(k) ⇤
model(k)i
p
Ptruth(k)Pmodel(k)<latexit sha1_base64="cRIq0DTnUaRL8LfD6bOtGkmKbM8=">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</latexit>
For best-fit model, , so focus on herePerr = Ptruth(1 r2
cc)<latexit sha1_base64="tRkpT8H4AD9a4kRuQlWo+CYRCgo=">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</latexit>
Perr(k) ⌘ h| truth(k) model(k)|2
i<latexit sha1_base64="okwU7UK76wFFi51s3hFUJl6za4k=">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</latexit>
Perr<latexit sha1_base64="HbsFOTY+FC07lz5Qau+blG/LzoY=">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</latexit>
62. Error power spectrum (= MSE)
Poisson noise 1/¯n
Perr = h| truth
h
model
h |2
i
Simulated halos (truth)
Quadr. bias (model)
isson noise 1/¯n
h model
h |2
i
s (model)
Poisson noise 1/¯n
Perr = h| truth
h
model
h |2
i
Simulated halos (truth)
Quadr. bias (model)
Fourier wavenumber k
Powerspectrum
Error is quite white, similar to Poisson shot noise (discrete sampling)
MS, Simonović et al. (2019)
63. Tried many other nonlinear bias operators
Give few times larger model error
Fourier wavenumber k
Figure 20. Left panel: Model error power spectrum for Standard Eulerian bias models, for
nonlinear dark matter NL from simulations as the input for the Standard Eulerian bias mo
large scales because it involves squaring NL, which is rather UV-sensitive. Alternatively,
density as the input to the bias model (dark orange) is treating large bulk flows perturbaMS, Simonović et al. (2019)
Errorpowerspectrum
64. Scale dependence of the error
±1% of Phh
Expanding Z
Quadratic bias
Cubic bias
Linear Std.
Eul. bias
Scale dependence of the noise important around the nonlinear scale
Potentially dangerous because can bias cosmological parameters
65. kmax up to which noise is constant
kmax when scale dependence of Perr is detectable with 1σ in V=10
h-3Gpc3 volume (or V=0.5 in brackets):
kmax [hMpc 1
]
log M[h 1
M ] ¯n [(h 1
Mpc) 3
] Lin. Std. Eul. Cubic
10.8 11.8 4.3 ⇥ 10 2
0.1 (0.14) 0.3 (0.37)
11.8 12.8 5.7 ⇥ 10 3
0.08 (0.1) 0.18 (0.24)
12.8 13.8 5.6 ⇥ 10 4
0.07 (0.1) 0.13 (0.18)
13.8 15.2 2.6 ⇥ 10 5
0.1 (0.14) 0.24 (0.32)
ximum wavenumber kmax when a scale dependence of the model error can be detected with 1 in a 10 h
a 0.5 h 3
Gpc3
volume, shown in brackets), for the linear Standard Eulerian and the cubic bias models. Th
model is typically 2 to 3 higher than that of linear Standard Eulerian bias. This can improve measurem
arameters that a↵ect the galaxy power spectrum at these scales, for example the sum of neutrino masse
significant uncertainty due to the noise in our measurement of the model error, as shown in Fig. 11.
of the survey volume. We also summarize this in Table IV for two survey volumes.
ear Standard Eulerian bias we find the following. For the lowest halo mass bin, the scale depen
ificant for kmax ' 0.1 hMpc 1
in a 10 h 3
Gpc3
volume or for kmax ' 0.14 hMpc 1
in a 0.5 h
Nonlinear bias has 2-3x higher kmax
8-30x more Fourier modes
4-5x smaller error bars (in principle; also have more params!)
66. Weighting halos by their mass
Halo number density
(how many halos per cell)
Halo mass density
(how much halo mass per cell)
used so far more similar to dark matter
smaller shot noise
Seljak, Hamaus & Desjacques (2009)
Hamaus, Seljak & Desjacques (2010, 2011, 2012)
Cai, Bernstein & Sheth (2011)
67. Weighting halos by their mass
Shot noise (squared
model error) 17x
lower for light halos,
2-7x lower for heavy
halos
With 60% halo mass
scatter (green), still
get factor few
How well can we do
observationally?
49
mass weighting on the mean-square model error divided by the Poisson expectation, h| obs
h (k)
obs obs truth ?
68. Mass weighting questions
How well can halo masses be measured (e.g. BOSS, DESI)?
What observable properties of galaxies can we use? What sims?
New ideas to get halo masses?
For shot noise limited applications, gain may be large
What if mass estimates are biased?
Use for BAO reconstruction? (Suffers from high shot noise)
69. Science with galaxy catalogs
(1) Bias model at the field level
(2) Cosmological parameter analysis
(3) Accounting for skewness
(4) Getting initial from final conditions
13
Figure 1. 2-d slices of the overdensity h(x) of simulated 1010.8
1011.8
h 1
M halos (top), compared with the cubic bias
model (center), and the linear Standard Eulerian bias model (bottom). Each panel is 500 h 1
Mpc wide and 110 h 1
Mpc high,
and each density is smoothed with a R = 2 h 1
Mpc Gaussian, WR(k) = exp[ (kR)2
/2]. The colorbar indicates the values of
this smoothed overdensity h(x).
Figure 2. Same as Fig. 1 but for more massive and less abundant 1011.8
1012.8
h 1
M halos.
simulations. The variance, skewness and kurtosis of the densities shown in the histograms are listed in Table II for
the full simulated and modeled densities, and in Table III for the model error.
The linear Standard Eulerian bias model tends to underpredict troughs and overpredict peaks of the halo density,
as shown in Fig. 3. The model error is not Gaussian for any of the shown smoothing scales; in particular its kurtosis
is larger than 1 for all smoothing scales.
The cubic model provides a more accurate description of the halo density pdf, as shown in Fig. 4. This emphasizes
the importance of using nonlinear bias terms even on rather large scales. Still, the cubic model tends to underpredict
the peaks of the true halo density, especially on small scales. This agrees with Fig. 2 where the model also underpredicts