hi#intensity#mapping# with## green#banktelescope#gcosmo.bao.ac.cn/sites/default/files/131113_lunch...

HI Intensity Mapping with

Green Bank Telescope

Yi-‐Chao LI (NAOC) Ph. D.Candidate

•  Team members: ASIAA

Tzu-‐Ching Chang Yu-‐Wei (Victor) Liao

Carnegie Mellon Kevin Bandura (now at McGill) Aravind Natarajan Jeff Peterson Tabitha Voytek

UW-‐Madison

Chris Anderson Peter Timbie Le Zhang

CITA Nidhi Banavar L.-‐M. Calin Kiyo Masui (now at UBC) Ue-‐Li Pen Richard Shaw Eric Switzer (now at NASA)

NAOC

Jaswant Kumar Yi-‐Chao Li Xuelei Chen

Outline

•  IntroducUon •  Our GBT observaUon •  Map making •  Foreground clean •  Power spectrum esUmaUon •  Result and future plan •  Summary

IntroducUon

•  HI 21cm emission

At low redshiA, instead of Galaxy counFng, HI intensity mapping can be used to trace the dark maIer distribuFon.

Our GBT ObservaUon

•  Why use GBT: •  angular resoluUon good for intensity

mapping ~ 0.25°è10 h-‐1Mpc at z ~ 0.8

•  clean beam, unblocked aperture •  quiet receiver for 700 – 900 MHz band

(TSystem ~ 25 K) •  RFI OK for 700 – 900 MHz receiver (z ~ 1 -‐

0.58) •  Available

•  other opUons: Parkes (64m), Effelsberg (100m)

Our GBT ObservaUon

•  ObservaUon field overlapped with WiggleZ survey. ‘15 hr deep’ ‘1 hr shallow’

coordinates 14h31m28.5s RA 2°0’ Dec 0h52m0s RA 2°9’ Dec

area 4.5 deg x 2.4 deg = 10.8 deg2

7.0 deg x 4.3 deg = 30.1 deg2

integraUon Ume 105 hrs 84 hrs

angular resoluUon 0.314° – 0.25° FWHM transverse spaUal resoluUon

9.6 h-‐1 Mpc at 800 MHz

RF bandwidth 700 – 900 MHz (z = 1 to 0.58) in 4096 channels radial resoluUon rebinned to 3.8 h-‐1Mpc

Our GBT ObservaUon

•  RFI removal –  RFI flagging using high resoluUon spectrum. –  Variance: remove frequencies with excess variance in a scan –  Linear polarizaUon: remove frequencies with strong correlaUon

between V and H polarizaUon channels

Our GBT ObservaUon

•  CalibraUon –  1.Noise diode (ideally fully linear polarized with well-‐known

polarizaUon angle) –  2. Quasar data (which is considered to be less polarized) as reference

observaUon –  3. Pulsar data (which is considered to be highly linear polarized) with

different parallacUc angles. CalibraFon is performed by first dividing by the noise diode power (averaged over a scan) in each channel, and then converFng to flux using calibraFon observaFons of 3C286 and 3C48.

Map Making

•  OpUmized Map Making Dt = Ptimi + ntχ 2 = (dt −Ptimi )

T N −1(dt −Ptimi )

(PtiT N −1Pti )

m = PtiT N −1d

The Astrophysical Journal Letters, 763:L20 (5pp), 2013 January 20 Masui et al.

1

1.5

2

2.5

3

215.5 216 216.5 217 217.5 218 218.5 219 219.5 220

Dec

RA

GBT 15hr field (800.4 MHz, z = 0.775)

-1000

-500

0

500

1000

Tem

pera

ture

(m

K)

1

1.5

2

2.5

3

215.5 216 216.5 217 217.5 218 218.5 219 219.5 220

Dec

RA

GBT 15hr field, cleaned, beam convolved (800.4 MHz, z = 0.775)

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

Tem

pera

ture

(m

K)

Figure 1. Maps of the GBT 15 hr field at approximately the band-center. The purple circle is the FWHM of the GBT beam, and the color range saturates in someplaces in each map. Left: the raw map as produced by the map-maker. It is dominated by synchrotron emission from both extragalactic point sources and smootheremission from the galaxy. Right: the raw map with 20 foreground modes removed per line of sight relative to 256 spectral bins, as described in Section 3.2. The mapedges have visibly higher noise or missing data due to the sparsity of scanning coverage. The cleaned map is dominated by thermal noise, and we have convolved byGBT’s beam shape to bring out the noise on relevant scales.(A color version of this figure is available in the online journal.)

redshift-space power spectrum using the empirical-non-linear(NL) model described by Blake et al. (2011).

3.2. From Maps to Power Spectra

The approach to 21 cm foreground subtraction in literaturehas been dominated by the notion of fitting and subtractingsmooth, orthogonal polynomials along each line of sight. Thisis motivated by the eigenvectors of smooth synchrotron fore-grounds (Liu & Tegmark 2011, 2012). In practice, instrumentalfactors such as the spectral calibration (and its stability) andpolarization response translate into foregrounds that have morecomplex structure. One way to quantify this structure is to usethe map itself to build the foreground model. To do this, wefind the frequency–frequency covariance across the sample ofangular pixels in the map, using a noise inverse weight. We thenfind the principal components along the frequency direction, or-der these by their singular value, and subtract a fixed number ofmodes of the largest covariance from each line of sight. Becausethe foregrounds dominate the real map, they also dominate thelargest modes of the covariance.

There is an optimum in the number of foreground modes toremove. For too few modes, the errors are large due to residualforeground variance. For too many modes, 21 cm signal is lost,and so after compensating based on simulated signal loss (seebelow), the errors increase modestly. We find that removing20 modes in both the 15 hr and 1 hr field maximizes the signal.Figure 1 shows the foreground-cleaned 15 hr field map.

We estimate the cross-power spectrum using the inverse noisevariance of the maps and the WiggleZ selection function asthe weight for the radio and optical survey data, respectively.The variance is estimated in the mapping step and representsnoise and survey coverage. The foreground cleaning processalso removes some 21 cm signal. We compensate for signalloss using a transfer function based on 300 simulations wherewe add signal simulations to the observed maps (which aredominated by foregrounds), clean the combination, and find thecross-power with the input simulation. Because the foregroundsubtraction is anisotropic in k! and k", we estimate and applythis transfer function in two dimensions (2D). The GBT beamacts strictly in k!, and again we develop a 2D beam transferfunction using signal simulations with the beam.

0.001

0.01

0.1

1

0.1

!(k)

2 (m

K)

k (h Mpc-1)

15 hr1 hr"HI bHI r = 0.43 10-3

Figure 2. Cross-power between the 15 hr and 1 hr GBT fields and WiggleZ.Negative points are shown with reversed sign and a thin line. The solid line isthe mean of simulations based on the empirical-NL model of Blake et al. (2011)processed by the same pipeline.(A color version of this figure is available in the online journal.)

The foreground filter is built from the real map which has alimited number of independent angular elements. This causesthe transfer function to have components in both the angularand frequency direction (Nityananda 2010), with the angularpart dominating. This is accounted for in our transfer function.Subtleties of the cleaning method will be described in a futuremethods paper.

We estimate the errors and their covariance in our cross-power spectrum by calculating the cross-power of the cleanedGBT maps with 100 random catalogs drawn from the WiggleZselection function (Blake et al. 2010). The mean of thesecross powers is consistent with zero, as expected. The varianceaccounts for shot noise in the galaxy catalog and variance inthe radio map either from real signal (sample variance), residualforegrounds or noise. Estimating the errors in this way requiresmany independent modes to enter each spectral cross-power bin.This fails at the lowest k values and so these scales are discarded.In going from the 2D power to the 1D powers presented here,we weight each 2D k-cell by the inverse variance of the 2Dcross-power across the set of mock galaxy catalogs. The 2D–1Dbinning weight is multiplied by the square of the beam andforeground cleaning transfer functions. Figure 2 shows theresulting galaxy–H i cross-power spectra.

3

K. W. Masui et. al. 2012

Foreground clean

•  We use SVD method for foreground cleanning – Find frequency frequency covariance of the foreground

– Find the line-‐of-‐sight (LoS) modes, by SVD – Subtract N nodes from each LoS Note: these modes are NOT smooth polynomials.

Foreground clean

•  Spliong the season to get lower noise. Splilng%the%season%

Full% C%A% B% D%

split%

A% ×% A%Signal%power%plus%noise%power.%

A% ×% Thermal%noise%and%some%RFI%is%uncorrelated%across%Zme.%B%

0V25%hrs% 25V50%hrs% 50V75%hrs% 75V100%hrs%100%hrs%

Form%6%unique%pairs;%rudimentary%errors%

Foreground clean

•  Convolve to common beam The Astrophysical Journal Letters, 763:L20 (5pp), 2013 January 20 Masui et al.

1

1.5

2

2.5

3

215.5 216 216.5 217 217.5 218 218.5 219 219.5 220

Dec

RA

GBT 15hr field (800.4 MHz, z = 0.775)

-1000

-500

0

500

1000

Tem

pera

ture

(m

K)

1

1.5

2

2.5

3

215.5 216 216.5 217 217.5 218 218.5 219 219.5 220

Dec

RA


-0.6

-0.4

-0.2

0

0.2

0.4

0.6

Tem

pera

ture

(m

K)







0.001

0.01

0.1

1

0.1

!(k)

2 (m

K)

k (h Mpc-1)

15 hr1 hr"HI bHI r = 0.43 10-3




3

K. W. Masui et. al. 2012 20 modes subtracted

Power spectrum esUmaUon

•  Cross power spectrum CrossVpairs:%WiggleZ%crosspower%

(% *%

(%

)%

)%*%×%

Map%A%

SelecZon%funcZon%

Galaxy%catalog%

Weight%A%


•  Auto power spectrum CrossVpairs:%GBT%power%

(% *%

(%

)%

)%*%×%

Map%A%

Weight%B%

Map%B%

Weight%A%


•  Power spectrum compensaUon – Due to the non-‐smooth foreground modes, subtracUons may kill power signal.

– Subtract the signal only simulaUon map to get the signal loss transfer funcUon

21 cm auto-power L3

however, can convert these into new degrees of freedom. An im-perfect and time-dependent passband calibration will cause intrin-sically spectrally smooth foregrounds to occupy multiple modes inour maps with non-trivial spectral structure. We control this usinga pulsed electronic calibrator, averaged for each scan.

We believe that the most challenging spectral structure from fore-grounds is caused by leakage of polarization into intensity. Here,each Mueller matrix element has a characteristic beam on the sky,dependent on offset from the boresight and frequency. The spectralstructure converts spectrally smooth polarization into new degreesof freedom. Faraday rotation of the polarization introduces furtherspectral degrees of freedom.

The leakage beam is optical in origin, mixes !10 per cent ofpolarization to intensity, is antisymmetric about the boresight to agood approximation and is slightly broader than the primary beam.In addition, the frequency dependence of the pure Stokes I beammixes spatial into spectral structure. We mitigate both of these termsby convolving to a common resolution corresponding to 1.4 timesthe beam size at 700 MHz (the largest beam). This convolution isbased on a frequency-dependent beam model from source scans.Such a convolution is viable because GBT has roughly twice theresolution needed to map LSS in the linear regime. However, thisconvolution reduces the number of independent resolution elementsin the map by a factor of 2, increasing the challenges discussed inSection 2.1.

The present results are limited largely by the area of the regionsand our understanding of the instrument. With a factor of roughly10 more than the present area, the resolution could be degraded atless expense to the signal. This requires significant telescope timebecause the area must also be covered to roughly the same depthas our present fields. It would however provide a significant boostin overall sensitivity for scientific goals such as measurement ofthe redshift-space distortions. In addition, we are investigating mapmaking that would unmix polarization using the Mueller matrix ofbeams, as determined from source scans.

2.3 Power spectrum estimation

Our starting point for power spectral estimation is the optimalquadratic estimator described in Liu & Tegmark (2011). To avoidthe thermal noise bias, we only consider cross-powers between foursubseason maps (Tristram et al. 2005), labelled here as {A, B, C,D}. Thermal noise is uncorrelated between these sections, whichwe have chosen to have similar integration depth and coverage.The foreground modes are determined separately for each side ofthe pair using the SVD of Section 2.1. Up to a normalization, theresulting estimator for the pair of submaps A and B is P (ki)A"B #(wA!AmA)TQiwB!B mB . Here, we have unwrapped the map ma-trix MA into a one-dimensional map vector mA and written theforeground cleaning projection (1/WA) $ (1 % UASUT

A)WA $ MA

as !AmA. The weighted mean of each frequency slice of the mapis also subtracted. The map weight wA is the matrix WA used inthe SVD, but unwrapped, and along the diagonal. Procedurally, theestimator amounts to weighting both foreground-cleaned maps, tak-ing the Fourier transform, and then summing the three-dimensionalcross-pairs to find power in annuli in two-dimensional k-space,ki = {k&,i , k',i}. The Fourier transform and binning are performedby Qi here. We calculate six such crossed pairs from the four-waysubseason split of the data, and let the average over these be theestimated power P (ki).

We calculate transfer functions to describe signal lost in theforeground cleaning and through the finite instrumental resolution.

These are functions of k& and k'. The beam transfer function isestimated using Gaussian 21 cm signal simulations that have beenconvolved to the effective, frequency-independent beam describedin Section 2.2. The foreground cleaning transfer function can beefficiently estimated through Monte Carlo simulations as

T (ki) =!

[wA!A+s(mA + ms) % wA!AmA]TQi ms

(wAms)TQi ms

"2

, (1)

where the A + s subscript denotes the fact that the foreground clean-ing modes have been estimated from a ! % ! ( covariance that hasadded 21 cm simulation signal, ms. This quantity is squared becausecleaning is applied to both sides of the quadratic estimator of thepower spectrum. The limited number of angular resolution elements(Section 2.1) results in an anticorrelation of the cleaned foregroundswith the signal itself, represented by the term (wA!A+s mA)TQi ms.To reduce the noise of the simulation cross-power, note that wesubtract wA!AmA in the numerator. Finally, we find the weightedaverage of these across the four-way split of maps. We find that 300signal simulations are sufficient to estimate the transfer function.

After compensating for lost signal using transfer functions forthe beam and foreground cleaning, we bin the two-dimensionalpowers on to one-dimensional band-powers. We weight binsby their two-dimensional Gaussian inverse noise variance #N (ki)T (ki)2/Pauto(ki)2, where Pauto(ki) is the average of {PA"A,PB"B, PC"C, PD"D} (pairs which contain the thermal noise bias)and N (ki) is the number of three-dimensional k modes that enter atwo-dimensional bin ki . In addition to the Gaussian noise weights,we impose two additional cuts in the two-dimensional k-power. Fork' < 0.035 h Mpc%1, k& < 0.08 h Mpc%1 for the deep field and k& <

0.04 h Mpc%1 for the wide field, there are few harmonics in thevolume, resulting in strips in the two-dimensional power spectrumwhere the errors are poorly estimated and strongly correlated. Fork& > 0.3 h Mpc%1, the instrumental resolution produces significantsignal loss, so this is also truncated.

Foregrounds in the input maps and the 21 cm signal itself arenon-Gaussian, but after cleaning, the thermal noise dominates bothcontributions in an individual map, and Gaussian errors (see e.g.Das et al. 2011) provide a reasonable approximation. These takeas input the auto-power measurement itself (for sample variance)and PA"A terms that represent the thermal noise. Sample varianceis significant only in the deep field in the lower 1/3 of the reportedwavenumbers. Gaussian errors agree with the standard deviation ofthe six crossed pairs that enter the spectral estimation in the regimewhere sample variance is negligible.

The finite survey size and weights result in correlations betweenadjacent k-bins. We apodize in the frequency direction using aBlackman window and in the angular direction using the mapweight itself (which falls off at the edges due to scan coverage).The bin–bin correlations are estimated using 3000 signal plus ther-mal noise simulations assuming Tsys = 25 K. To construct a fullcovariance model, these are then recalibrated by the outer productof the Gaussian error amplitudes for the data relative to the thermalnoise simulation errors.

The Bayesian method developed in the next section assumes thatadjacent bins are uncorrelated. To achieve this, we take the matrixsquare root of the inverse of our covariance model matrix and nor-malize its rows to sum to one. This provides a set of functions whichdecorrelates (Hamilton & Tegmark 2000) the pre-whitened powerspectrum and boosts the errors. At large scales (k = 0.1 h Mpc%1)where these effects are relevant, decorrelation and sample varianceincrease the errors by a factor of 1.5 in the wide field and 4 in thedeep field.

at National A

stronomical O

bservatory on June 21, 2013http://m

nrasl.oxfordjournals.org/D

ownloaded from

E. R. Switzer et. al. 2013 -‐-‐-‐ eq (1)

Result and future plan

•  Result

The Astrophysical Journal Letters, 763:L20 (5pp), 2013 January 20 Masui et al.

1

1.5

2

2.5

3

215.5 216 216.5 217 217.5 218 218.5 219 219.5 220

Dec

RA

GBT 15hr field (800.4 MHz, z = 0.775)

-1000

-500

0

500

1000

Tem

pera

ture

(mK

)

1

1.5

2

2.5

3

215.5 216 216.5 217 217.5 218 218.5 219 219.5 220

Dec

RA


-0.6

-0.4

-0.2

0

0.2

0.4

0.6

Tem

pera

ture

(mK

)







0.001

0.01

0.1

1

0.1

!(k)

2 (mK

)

k (h Mpc-1)

15 hr1 hr"HI bHI r = 0.43 10-3




3

L4 E. R. Switzer et al.

3 R ESULTS

The auto-power spectra presented in Fig. 1 will be biased by anunknown positive amplitude from residual foreground contami-nation. These data can then be interpreted as an upper boundon the neutral hydrogen fluctuation amplitude, !H IbH I. In addi-tion, we have also measured the cross-correlation with the Wig-gleZ galaxy survey (Masui et al. 2013). This finds !H IbH Ir =[0.43 ± 0.07(stat.) ± 0.04(sys.)] ! 10"3, where r is the WiggleZgalaxy–neutral hydrogen cross-correlation coefficient (taken hereto be independent of scale). Since |r| < 1 by definition and is mea-sured to be positive, the cross-correlation can be interpreted as alower bound on !H IbH I. In this section, we will develop a pos-terior distribution for the 21 cm signal auto-power between thesetwo bounds, as a function of k. We will then combine these into aposterior distribution on !H IbH I.

The probability of our measurements given the 21 cm signal auto-power and foreground model parameters is

p(dk|! k) = p(dc|sk, r)p!d

deepk |sk, f

deepk

"p

#dwide

k |sk, fwidek

$. (2)

Here, dk = {dc, ddeepk , dwide

k } contains our cross-power anddeep and wide field auto-power measurements, while ! k ={sk, r, f

deepk , f wide

k } contains the 21 cm signal auto-power, cross-correlation coefficient, and deep and wide field foreground con-tamination powers, respectively. The cross-power variable dc

represents the constraint on !H IbH Ir from both fields and the rangeof wavenumbers used in Masui et al. (2013). The band-powers d

deepk

and dwidek are independently distributed following decorrelation of

Figure 1. Temperature scales in our 21 cm intensity mapping survey. Thetop curve is the power spectrum of the input deep field with no cleaningapplied (the wide field is similar). Throughout, the deep field results aregreen and the wide field results are blue. The dotted and dash–dotted linesshow thermal noise in the maps. The power spectra avoid noise bias bycrossing two maps made with separate data sets. The points below show thepower spectrum of the deep and wide fields after the foreground cleaningdescribed in Section 2.1. Individual modes in the map are dominated bythermal noise rather than residual foregrounds or signal. Errors are thethermal noise power divided by the number of modes in the k-bin, plussample variance. The negative values are shown with thin lines and hollowmarkers. The red dashed line shows the 21 cm signal expected from theamplitude of the cross-power with the WiggleZ survey (for r = 1) and basedon simulations processed by the same pipeline.

finite-survey effects. We assume that the foregrounds are uncorre-lated between k-bins and fields, also. This is conservative becauseknowledge of foreground correlations would yield a tighter con-straint. We take p(dc|sk, r) to be normally distributed with meanproportional to r

#sk , and p(ddeep

k |sk, fdeepk ) to be normally dis-

tributed with mean sk + fdeepk and errors determined in Section 2.3

(and analogously for the wide field). Only the statistical uncer-tainty is included in the width of the distributions, as the systematiccalibration uncertainty is perfectly correlated between cross- andauto-power measurements and can be applied at the end of theanalysis.

We apply Bayes’ theorem to obtain the posterior distribution forthe parameters, p(! k|dk) $ p(dk|! k)p(sk)p(r)p(f deep

k )p(f widek ).

For the nuisance parameters, we adopt conservative priors. p(f deepk )

and p(f widek ) are taken to be flat over the range 0 < fk < %. Like-

wise, we take p(r) to be constant over the range 0 < r < 1, whichis conservative given the theoretical bias towards r & 1. Our goalis to marginalize over these nuisance parameters to determine sk.We choose the prior on sk to be flat, which translates into a priorp(!H IbH I) $ !H IbH I. The signal posterior is

p(sk|dk) =%

p!sk, r, f

deepk , f wide

k |dk

"dr df

deepk df wide

k . (3)

This involves integrals of the form& 1

0 p(dc|s, r)p(r) dr which,given the flat priors that we have adopted, can generally be writtenin terms of the cumulative distribution function of p(dc|s, r). Fig. 2shows the allowed signal in each spectral k-bin.

Taking the analysis further, we combine band-powers into a sin-gle constraint on !H IbH I. Following Masui et al. (2013), we con-sider a conservative k-range where errors are better estimated (k >

0.12h Mpc"1 to avoid edge effects in the decorrelation operation)and before uncertainties in non-linear structure formation becomesignificant (k < 0.3 h Mpc"1). Fig. 3 shows the resulting posteriordistribution.

Our analysis yields !H IbH I = [0.62+0.23"0.15] ! 10"3 at 68 per cent

confidence with 9 per cent systematic calibration uncertainty. Therange of allowed !H IbH I is bracketed by the cross- and auto-power

Figure 2. Comparison with the thermal noise limit. The dark and lightshaded regions are the 68 and 95 per cent confidence intervals of the mea-sured 21 cm fluctuation power from equation (3). The dashed line shows theexpected 21 cm signal implied by the WiggleZ cross-correlation if r = 1.The solid line represents the best upper 95 per cent confidence level that wecould achieve given our error bars in both fields, in the absence of fore-ground contamination. Note that the autocorrelation measurements, whichconstrain the signal from above, are uncorrelated between k-bins, while asingle global fit to the cross-power (in Masui et al. 2013) is used to constrainthe signal from below. Confidence intervals do not include the systematiccalibration uncertainty, which is 18 per cent in this space.

at National A

stronomical O



ownloaded from

K. W. Masui et. al. 2012

E. R. Switzer et. al. 2013


•  Result Combine 2 field auto-‐ and cross power spectrum, we get at 68% confidence E. R. Switzer et. al. 2013

21 cm auto-power L5

Figure 3. The posterior distribution for the parameter !H IbH I coming fromthe WiggleZ cross-power spectrum, deep field and wide field auto-powers aswell as the joint likelihood from all three data sets. The priors are describedin Section 3. The distributions do not include the systematic calibrationuncertainty of 9 per cent.

measurements, and is a robust statement. The peak of the posteriorbetween these bounds is sensitive to the prior choice, and so thequoted posterior should be interpreted in the context of our priorchoices here. Another reasonable signal prior is that P (!H IbH I) isflat, which shifts the central value by !10 per cent. Note that weare unable to calculate a goodness of fit to our model because eachmeasurement is associated with a free foreground parameter whichcan absorb any anomalies.

4 D I S C U S S I O N A N D C O N C L U S I O N S

Through the measurement of the auto-power, we extend our previ-ous cross-power measurement of !H IbH Ir (Masui et al. 2013) to adetermination of !H IbH I. This is the first constraint on the amplitudeof 21 cm fluctuations at z ! 0.8, and it circumvents the degeneracywith the cross-correlation r. The 21 cm auto-power yields a true up-per bound because it derives from the integral of the mass function.In the future, redshift distortions (Wyithe 2008; Masui, McDonald& Pen 2010) can be used to further break the degeneracy betweenbH I and !H I, and complement challenging Hubble Space Telescopemeasurements of !H I (Rao, Turnshek & Nestor 2006). Our presentsurvey is limited by area and sensitivity, but we have shown thatforegrounds can be suppressed sufficiently, to nearly the level ofthe 21 cm signal, using an empirical mode subtraction method. Fu-ture surveys exploiting the auto-power of 21 cm fluctuations mustdevelop statistics that are robust to the additive bias of residual fore-grounds and that control instrumental systematics such as polarizedbeam response and passband stability.

ACK NOWLEDG E ME NTS

We would like to thank John Ford, Anish Roshi and the rest ofthe GBT staff for their support, and Paul Demorest and Willemvan-Straten for help with pulsar instruments and calibration.

The National Radio Astronomy Observatory is a facility of theNational Science Foundation operated under cooperative agree-ment by Associated Universities, Inc. This research is supportedby NSERC Canada and CIFAR. JBP and TCV acknowledge NSFgrant AST-1009615. XLC acknowledges the Ministry of Scienceand Technology Project 863 (2012AA121701), The John Temple-ton Foundation and NAOC Beyond the Horizon programme and TheNSFC grant 11073024. AN acknowledges financial support fromthe McWilliams Center for Cosmology. Computations were per-formed on the GPC supercomputer at the SciNet HPC Consortium.SciNet is funded by the Canada Foundation for Innovation.

R E F E R E N C E S

Ansari R., Campagne J.-E., Colom P., Magneville C., Martin J.-M., MoniezM., Rich J., Yeche C., 2012a, C. R. Phys., 13, 46

Ansari R. et al., 2012b, A&A, 540, A129Battye R. A. et al., 2012, preprint (arXiv:1209.1041)Bebbington D. H. O., 1986, MNRAS, 218, 577Bowman J. D., Rogers A. E. E., 2010, Nat, 468, 796Chang T.-C., Pen U.-L., Peterson J. B., McDonald P., 2008, Phys. Rev. Lett.,

100, 091303Chang T.-C., Pen U.-L., Bandura K., Peterson J. B., 2010, Nat, 466, 463Chen X., 2012, Int. J. Mod. Phys., 12, 256Das S. et al., 2011, ApJ, 729, 62Drinkwater M. J. et al., 2010, MNRAS, 401, 1429Hamilton A. J. S., Tegmark M., 2000, MNRAS, 312, 285Komatsu E. et al., 2009, ApJS, 180, 330Lah P. et al., 2009, MNRAS, 399, 1447Liu A., Tegmark M., 2011, Phys. Rev. D, 83, 103006Liu A., Tegmark M., 2012, MNRAS, 419, 3491Loeb A., Wyithe J. S. B., 2008, Phys. Rev. Lett., 100, 161301Masui K. W., McDonald P., Pen U.-L., 2010, Phys. Rev. D, 81, 103527Masui K. W. et al., 2013, ApJ, 763, L20Nityananda R., 2010, Technical Reports, NCRAPaciga G. et al., 2013, preprint (arXiv:1301.5906)Pober J. C. et al., 2013a, preprint (arXiv:1301.7099)Pober J. C. et al., 2013b, AJ, 145, 65Rao S. M., Turnshek D. A., Nestor D. B., 2006, ApJ, 636, 610Seo H.-J., Dodelson S., Marriner J., Mcginnis D., Stebbins A., Stoughton

C., Vallinotto A., 2010, ApJ, 721, 164Shaw J. R., Sigurdson K., Pen U.-L., Stebbins A., Sitwell M., 2013, preprint

(arXiv:1302.0327)Subrahmanyan R., Anantharamaiah K. R., 1990, JA&A, 11, 221Tristram M., Macıas-Perez J. F., Renault C., Santos D., 2005, MNRAS, 358,

833Wieringa M. H., de Bruyn A. G., Katgert P., 1992, A&A, 256, 331Wyithe S., 2008, preprint (arXiv:0804.1624)

This paper has been typeset from a TEX/LATEX file prepared by the author.

at National A

stronomical O



ownloaded from

MNRASL (2013) doi:10.1093/mnrasl/slt074

Determination of z ! 0.8 neutral hydrogen fluctuations using the 21 cmintensity mapping autocorrelation

E. R. Switzer,1‹ K. W. Masui,1,2‹ K. Bandura,3 L.-M. Calin,1 T.-C. Chang,4

X.-L. Chen,5,6 Y.-C. Li,5 Y.-W. Liao,4 A. Natarajan,7 U.-L. Pen,1 J. B. Peterson,7

J. R. Shaw1 and T. C. Voytek7

1Canadian Institute for Theoretical Astrophysics, University of Toronto, 60 St George St, Toronto, Ontario M5S 3H8, Canada2Department of Physics, University of Toronto, 60 St George St, Toronto, Ontario M5S 1A7, Canada3Department of Physics, McGill University, 3600 Rue University, Montreal, Quebec H3A 2T8, Canada4Academia Sinica Institute of Astronomy and Astrophysics, PO Box 23-141, Taipei 10617, Taiwan5National Astronomical Observatories, Chinese Academy of Science, 20A Datun Road, Beijing 100012, China6Center of High Energy Physics, Peking University, Beijing 100871, China7McWilliams Center for Cosmology, Department of Physics, Carnegie Mellon University, 5000 Forbes Ave, Pittsburgh PA 15213, USA

Accepted 2013 May 30. Received 2013 May 22; in original form 2013 April 15

ABSTRACTThe large-scale distribution of neutral hydrogen in the Universe will be luminous through its21 cm emission. Here, for the first time, we use the auto-power spectrum of 21 cm intensityfluctuations to constrain neutral hydrogen fluctuations at z ! 0.8. Our data were acquired withthe Green Bank Telescope and span the redshift range 0.6 < z < 1 over two fields totalling"41 deg2 and 190 h of radio integration time. The dominant synchrotron foregrounds exceedthe signal by !103, but have fewer degrees of freedom and can be removed efficiently. Even inthe presence of residual foregrounds, the auto-power can still be interpreted as an upper boundon the 21 cm signal. Our previous measurements of the cross-correlation of 21 cm intensityand the WiggleZ galaxy survey provide a lower bound. Through a Bayesian treatment of signaland foregrounds, we can combine both fields in auto- and cross-power into a measurement of!H IbH I = [0.62+0.23

#0.15] $ 10#3 at 68 per cent confidence with 9 per cent systematic calibrationuncertainty, where !H I is the neutral hydrogen (H I) fraction and bH I is the H I bias parameter.We describe observational challenges with the present data set and plans to overcome them.

Key words: galaxies: evolution – large-scale structure of universe – radio lines: galaxies.

1 IN T RO D U C T I O N

There is substantial interest in the viability of cosmological struc-ture surveys that map the intensity of 21 cm emission from neutralhydrogen. Such surveys could be used to study large-scale structure(LSS) at intermediate redshifts, or to study the epoch of reionizationat high redshift. Surveys of 21 cm intensity have the potential to bevery efficient since the resolution of the instrument can be matchedto the large scales of cosmological interest (Chang et al. 2008; Loeb& Wyithe 2008; Seo et al. 2010; Ansari et al. 2012b). Several ex-periments, including BAOBAB (Pober et al. 2013b), BAORadio(Ansari et al. 2012a), BINGO (Battye et al. 2012), CHIME1 andTianLai (Chen 2012) propose to conduct redshift surveys from z !0.5 to 2.5 using this method.

" E-mail: [email protected] (ERS); [email protected] (KWM)1 http://chime.phas.ubc.ca/

The principal challenges for 21 cm experiments are astronomi-cal foregrounds and terrestrial radio frequency interference (RFI).Extragalactic sources and the Milky Way produce synchrotron emis-sion that is three orders of magnitude brighter than the 21 cm signal.However, the physical process of synchrotron emission is knownto produce spectrally smooth radiation, occupying few degrees offreedom along each line of sight. In the absence of instrumentaleffects, these degrees of freedom are thought to be separable fromthe signal (Liu & Tegmark 2011, 2012; Shaw et al. 2013). RFIcan be minimized through site location, sidelobe control and bandselection. In the Green Bank Telescope (GBT) data analysed here,RFI is not found to be a significant challenge or limiting factor.

Subtraction of synchrotron emission has proven to be challengingin practice. Instrumental effects such as passband calibration andpolarization leakage couple bright foregrounds into new degrees offreedom that need to be removed from each line of sight to reachthe level of the 21 cm signal. The spectral functions describingthese systematics cannot all be modelled in advance, so we take anempirical approach to foreground removal by estimating dominant

C% 2013 The AuthorsPublished by Oxford University Press on behalf of the Royal Astronomical Society

MNRASL Advance Access published June 19, 2013

at National A

stronomical O



ownloaded from


•  Future plan – Using Parkes mulUbeam speed up survey – Change GBT to mulUbeam – Tianlai & Chime


•  Parkes 1997PASA...14..111S

Staveley-‐Smith, L. (2013) Publica.ons of the Astronomical Society of Australia, 14(01), 111–116. doi:10.1071/AS97111

(a) HIPASS image

(b) HIPASS image mapped to our map shape

(c) Parkes image rescaled using point sources

Figure 1: Fig 1(a) shows the HIPASS map Fig 1(b) shows the HIPASS mapped into our map shape. Fig 1(b) shows theParkes image, made by the new data observed in October, 2012. It include five days observation and I ignore three beams(3,4,9), which are not working.

1

(a) HIPASS image

(b) HIPASS image mapped to our map shape

(c) Parkes image rescaled using point sources

Figure 1: Fig 1(a) shows the HIPASS map Fig 1(b) shows the HIPASS mapped into our map shape. Fig 1(b) shows theParkes image, made by the new data observed in October, 2012. It include five days observation and I ignore three beams(3,4,9), which are not working.

1

Parkes map

HIPASS


•  GBT new feed array ASIAA Tzu-‐Ching Chang Yuh-‐Jing Hwang Chin-‐Ting Ho Chi-‐Chang Lin

NRAO

Rich Bradley John Ford S. Srikanth Steve White

CMU

Jeff Peterson Tabitha Voytek

UW – Madison

Chris Anderson Peter Timbie


•  Tianlai & Chime – Quilter environment; – Fixed antenna, more stable; – Long observing Ume, large field of view.

•  Challenge – Mass data and Large field of view imaging – CalibraUon – Foreground clean

Summary

•  HI survey became more important in LSS •  ObservaUon of 2 WiggleZ fields using GBT •  OpUmized map making •  SVD foreground clean •  Power spectrum esUmaUon and constrains on \Omega_{HI}

•  Parkes data analysis •  Tianlai & Chime

Thanks J

hi#intensity#mapping# with## green#banktelescope#gcosmo.bao.ac.cn/sites/default/files/131113_lunch...

Documents