Eiichiro Komatsu
University of Texas at Austin
Non-Gaussianity From Inflation April 19, 2006
CMB High-z Cl
usters
LSS
Observational Constr
aints on Non-Gaussia
nity
Why Study NG? (Why Care?!)
• Who said that CMB should be Gaussian?
Don’t let people take it for granted!
It is remarkable that the observed CMB is (very close to bein g) Gaussian.
The WMAP map, when smoothed to 1 degree, is entirely do minated by the CMB signal.
If it were still noise dominated, no one would be surprised that the ma p is Gaussian.
The WMAP data are telling us that primordial fluctuations are very close to being Gaussian.
How common is it to have something so close to being Gaussian in as tronomy? E.g., Maxwellian velocity distribution, what else?
• It may not be so easy to explain that CMB is Gaussian, unless we have a compelling early universe model that predicts Gaussian primordial fluctuations: Inflation.
“Gaussianity” should be taken as seriously as “Flatness”.
Gaussianity vs Flatness
• People are generally happy that geometry of our Universe is flat.
1-total=-0.003 (+0.013, -0.017) (68% CL) (WMAP03+H
ST)
Geometry of our Universe is consistent with being flat t o ~3% accuracy at 95% CL.
• What do we know about Gaussianity?
For GfNLG2, -54<fNL<114 (95% CL) (WMAP03)
Primordial fluctuations are consistent with being Gaussi an to ~0.001% accuracy at 95% CL.
• In a way, inflation is supported more by Gaussiani
ty of primordial fluctuations than by flatness. (Just
kidding.)
Let’s Hunt Some NG!
• The existing CMB data already suggest that primordial fluc tuations are very close to being Gaussian; however, this d oes not imply, by any means, that they are perfectly G aussian.
In fact, we would be in a big trouble if fNL turned out to be too clos e to zero. Second-order GR perturbations in the standard cosmol ogical model must produce fNL~5 or so. (See Sabino Matarrese’s and Nicola Bartolo’s Talks, and Michelle Liguori’s Poster)
• Some inflationary models produce much larger f
NL. We ma y be able to distinguish candidate inflationary models by N G. (Not to mention that the simplest, single-field, slow-roll models produce tiny NG.)
• The data keep getting better.
• If we could find it, it would lead us to something huge.
How Do We Test Gaussianity
of CMB?
Finding NG?
•
Two approaches to•
I. Null (Blind) Tests / “Discovery” Mode This approach has been most widely used in the literature.
One may apply one’s favorite statistical tools (higher-order correlations, topolog y, isotropy, etc) to the data, and show that the data are (in)consistent with Gauss ianity at xx% CL.
PROS: This approach is model-independent.
CONS: We don’t know how to interpret the results.
“The data are consistent with Gaussianity” --- what physics do we learn from that? It is not clear what could be ruled out on the basis of this kind of test.
•
II. “Model-testing” Mode Somewhat more recent approaches.
Try to constrain “NG parameter(s)” (e.g., fNL)
PROS: We know what we are testing, we can quantify our constraints, and we c an compare different data sets.
CONS: Highly model-dependent. We may well be missing other important NG si gnatures.
Recent Tendency
• I. Null (Blind) Tests / “Discovery” Mode
This approach is being applied mostly to the “large-sc ale anomaly” of the WMAP data.
North-south asymmetry
Quadrupole-octopole alignment
Some pixels are too cold (Marcos Cruz’s Poster)
“Axis of Evil” (Joao Magueijo’s Talk)
Large-scale modulation
• II. “Model-testing” Mode
A few versions of fNL have been constrained using the bispectrum, Minkowski functionals and other statistic al methods.
App. II: What Do We Need?
• We need to know the predicted form of stati stical tools as a function of model paramet ers to fit the data.
• For
Gf
NL
G2, there are only three statistic al tools for which the analytical predictions are known:
The angular bispectrum
Komatsu & Spergel (2001); Babich & Zaldarriaga (2004)
The angular trispectrum
Okamoto & Hu (2002); Kogo & Komatsu (2006)
Minkowski functionals
Hikage, Komatsu & Matsubara (2006)
Simplified Model:
• Working assumption: f
NLis independent of scales
Clearly an oversimplification! (Note, however, that this form i s actually predicted from curvaton models and the non-linear Sachs-Wolfe effect in the large-scale limit.)
• Why use this ansatz? Current observations are not yet sensitive to scale-dependence of f
NL, but are only sens itive to the overall amplitude.
See Creminelli et al. (2005) for an alternative ansatz.
• Sensitivity Goal: f
NL~1
Why fNL~1? NG “floor”: the ubiquitous signal from the secon d-order GR produces something like fNL~5, which would set t he lower limit to which one may hope to detect the primordial non-Gaussianity.
It may be possible to achieve fNL~3 using the angular bispect rum from the Planck and CMBPol data. How do we go from t here to fNL~1?
x G
x fNLG2
xBartolo, Komatsu, Matarrese & Riotto (20
04)
R
5
3
How Do They Look?
x
G x f
NL
G2 x
Simulated temperature maps from
f
NL=0 f
NL=100
f
NL=1000 f
NL=5000
Is One-point PDF Useful?
Conclusion: 1-point PDF is not very u seful. (As far as CMB is concerned.)
A positive
f
NL yields nega tively skewed temperatur e anisotropy.One-point PDF from WMAP
• The one-point distribution of CMB temperature anisotropy looks pretty Gaussian.
Galaxy has been masked.
Left to right: Q (41GHz), V (61GHz), W (94GHz).
Spergel et al. (2006)
Angular Bispectrum, B lmn
• A simple statistic that captures all of the information contained in the third-order moment of CMB anisotropy.
• Theoretical predictions exist.
• Statistical properties well understood.
How Does It Look?
• Primordial
Inflation
Second-order PT
• Secondary
Gravitational lensing
Sunyaev-Zel’dovich effe
• Nuisance ct
Radio point sources
Our Galaxy
Komatsu & Spergel (2001)
?Angular Bispectrum, B lmn ?
•
Physical non-Gaussian signals should be generated in real space (via e.g., non-linear coupling), and thus should be mo re apparent in real space. The Central Limit Theorem makes alm coefficients more Gaussi an!
Non-Gaussianity localized in real space is obscured and spread ove r many l’s and m’s.
•
Challenges in the analysis Sky cut complicates the analysis in harmonic space in many ways.
Computationally expensive (but not impossible).
8 hours on 16 procs of an SGI Origin 300 for measuring all configur ations up to l=512.
Optimal “Cubic” Statistics
•
Motivation We know the shape of the angular bispectrum.
Find the “best statistic” in real space that is most sensitive to th e kind of non-Gaussianity we are looking for.
•
Results The statistics already combine all configurations of the bispectr um optimally.
Optimized just for fNL: Maximum adaptation of the approac h II.
1000 times faster for lmax=512 (30 sec vs 8 hours)
4000 times faster for lmax=1024 (4 minutes vs 11 days)
•
One can also find an optimized statistics just for point s ources.Komatsu, Spergel & Wandelt (2005)
Cubic Estimator = Skewness of Filtered Maps
•B(x) is a Wiener reconstructed primordial potential field.
•A(x) picks out relevant configurations of the bispectrum.
S
prim dVA x B
2 x f
NL
A r, ˆ n
l r b
lC
l a
lmY
lm n ˆ
lm
B r, ˆ n
l r b
lC
l a
lmY
lm n ˆ
lm
l r d
3kg
Tl k j
l kr
l r d
3kP k g
Tl k j
l kr
Wiener-reconstructed Primordial Curvature
Reconstruction can be made even better by including the polarization data.
(See Ben Wandelt’s Talk)
3 1 T
T
On the largest scale,
It Works Very Well.
•The statistics tested against simulations
UNBIASED UNCORRELATE
D UNBIASED
Komatsu et al. (2006)
Bispectrum Constraints
•
Still far, far away from fNL~1, but it could put so me interesting limits on parameters of curvaton models, ghost inflation, and DBI inflation model s.
58 fNL 134(95%)
Komatsu et al. (2003); Spergel et al. (2006)
54 fNL 114(95%)
(1yr) (3yr)
Angular Trispectrum, T lmpq (L)
• Why care?! Two reasons.
The trispectrum could be non-zero even when the bispectrum is exactly zero.
We may increase our sensitivity to primordial NG
by including the trispectrum in the analysis.
Not For WMAP, But Perhaps For Planck…
• Trispectrum (~ f
NL 2) is more sensitiv e than the bispect rum (~ f
NL) when f
N Lis large.
At Planck resolution, the trispectrum woul d be detected more s ignificantly than the b ispectrum, if fNL > 50.
Kogo & Komatsu (2006)
Minkowski Functionals
• Morphology
Area
Contour length
Euler characteristics (or “Genus”)
The number of hot spots minus cold spots.
• All quantities are evaluated as a function of the peak height relative to r.m.
s.
MFs from WMAP
f
NL 137(95%)
125 f
NL 139(95%)
(1yr)
Komatsu et al. (2003); Spergel et al. (2006)
(3yr)
Area Contour Length Genus
Analytical Calculations
•
The analytical formulae s hould be very useful: we do not need to run NG si mulations for doing MFs any more.•
The formulae indicate th at MFs are actually as se nsitive to fNL as the bispe ctrum; however, MFs do not contain any more info rmation than the bispectr um does.Hikage, Komatsu & Matsubara (2006)
Using Galaxies
• Not only CMB, but also the large-scale str ucture of the universe does contain inform ation about primordial fluctuations on larg e scales. (See Peter Coles’s Talk.)
• One example: Galaxy Bispectrum
Cosmic Inflation Probe (CIP), a galaxy survey measuring 10 mil lion galaxies at 3<z<6, would offer an opportunity to use this for mula to constrain fNL~5 (note that the scale measured by CIP is smaller than that measured by CMB by a factor of ~10!)
CIP
Using High-z Objects
• Massive objects forming at high-z are e xtremely rare: they form at high peaks o f (nearly) Gaussian random field.
• Even a slight distortion of a Gaussian tai
l can enhance (or reduce) the number of
high-z object dramatically. The higher th
e mass is, or the higher the redshift is, t
he bigger the effect becomes.
•
The current WMAP limits still permit large changes in the number of objects at high z.•
A golden object, like a few times 1014 solar masses at z=3, would be a smoking gun.Implications for high- z objects
Komatsu et al. (2003)
Summary
•
There have been a lot of development in this field, and there is still a lot more to do. It is timely to have a workshop that focus es on non-Gaussianity from Inflation! We need more accurate predictions for the form of observables, such a s the CMB bispectrum, trispectrum, Minkowski functionals, and others, from
Various models of inflation
Second-order PT, including second-order effects in Boltzmann equation
•
Not just CMB! Galaxies and high-z objects might give us some surprises.
•
Toward the sensitivity goal, fNL~1. What would be the best way to achieve this sensitivity?
Currently, the CMB bispectrum seems to be ahead of everything else.
The temperature plus polarization bispectrum would allow us to get down t o fNL~3. How do we break this barrier?