1: % PS.tex
2: %
3: % Ian O'Dwyer
4: %
5: % Power Spectrum paper for BEAST - DRAFT
6: %
7:
8: \documentclass[preprint]{aastex}
9: \begin{document}
10:
11:
12: \title{The CMB Anisotropy Power Spectrum from the {\em Background Emission Anisotropy Scanning Telescope} Experiment}
13:
14:
15: \author{Ian J. O'Dwyer\altaffilmark{1},
16: Marco Bersanelli\altaffilmark{2},
17: Jeffrey Childers\altaffilmark{3,4},
18: Newton Figueiredo\altaffilmark{5},
19: Doron Halevi\altaffilmark{3,4},
20: Gregory G. Huey\altaffilmark{1,6},
21: Philip M. Lubin\altaffilmark{3,4,7},
22: Davide Maino\altaffilmark{2},
23: Nazzareno Mandolesi\altaffilmark{8},
24: Joshua Marvil\altaffilmark{3,4},
25: Peter R. Meinhold \altaffilmark{3,4,7},
26: Jorge Mej\'{\i}a\altaffilmark{9},
27: Paolo Natoli\altaffilmark{10},
28: Hugh O'Neill\altaffilmark{3,4},
29: Agenor Pina\altaffilmark{5},
30: Michael D. Seiffert\altaffilmark{11},
31: Nathan C. Stebor\altaffilmark{3,4,7},
32: Camilo Tello\altaffilmark{9},
33: Thyrso Villela\altaffilmark{9},
34: Benjamin D. Wandelt\altaffilmark{1,6},
35: Brian Williams\altaffilmark{3,7},
36: Carlos Alexandre Wuensche\altaffilmark{9}
37: }
38:
39: \altaffiltext{1}{Astronomy Department, University of Illinois at Urbana-Champaign, IL 61801-3074}
40: \altaffiltext{2}{Physics Department, University of Milano, via Celoria 16, 20133 Milano, Italy}
41: \altaffiltext{3}{Physics Department, University of California, Santa Barbara, CA 93106}
42: \altaffiltext{4}{UC Santa Barbara Center for High Altitude Astrophyics at White Mountain}
43: \altaffiltext{5}{Universidade Federal de Itajub\'a, Departamento de F\'{\i}sica e Qu\'{\i}mica, Caixa Postal 50 37500-903, Itajub\'a, MG Brazil}
44: \altaffiltext{6}{Department of Physics, University of Illinois at
45: Urbana-Champaign, IL 61801-3080}
46: \altaffiltext{7}{University of California, White Mountain Research Station, CA 93514}
47: \altaffiltext{8}{IASF-CNR sezione di Bologna, via P.Gobetti, 101, 40129 Bologna, Italy}
48: \altaffiltext{9}{Instituto Nacional de Pesquisas Espaciais, Divis\~ao de Astrof\'{\i}sica, Caixa Postal 515, 12245-970 - S\~ao Jos\'e dos Campos, SP Brazil}
49: \altaffiltext{10}{Dipartimento di Fisica e sezione INFN, Universit\`a di Roma "Tor Vergata", Rome, Italy}
50: \altaffiltext{11}{Jet Propulsion Laboratory, California Institute of Technology, Oak Grove Drive, Pasadena, CA 91109}
51:
52:
53:
54:
55: \begin{abstract}
56: The Background Emission Anisotropy Scanning Telescope (BEAST) is a
57: 2.2m off-axis telescope with an 8 element mixed Q (38-45GHz) and Ka
58: (26-36GHz) band focal plane, designed for balloon borne and ground based studies of the Cosmic Microwave Background.
59: Here we present the Cosmic Microwave Background (CMB) angular
60: power spectrum calculated from 682 hours of data observed with the
61: BEAST instrument. We use a binned
62: pseudo-$C_{\ell}$ estimator (the MASTER method). We find results that are
63: consistent with other determinations of the CMB anisotropy for angular
64: wavenumbers $\ell$ between 100 and 600. We also perform cosmological parameter
65: estimation. The BEAST
66: data alone produces a good constraint on $\Omega_k\equiv 1-\Omega_{tot}=-0.074 \pm 0.070$, consistent with a flat Universe. A joint parameter estimation analysis
67: with a number of previous CMB experiments produces results consistent with previous determinations.
68:
69: %\noindent
70: \end{abstract}
71:
72: \keywords{}
73:
74: \section{Introduction}
75: Understanding the mechanisms of structure formation in the early
76: universe ($10<z<1000$) is one of the most important and active areas
77: in Cosmology today and measurements of the Cosmic Microwave Background
78: (CMB) anisotropy play a pivotal role in this field.
79: In the framework of the standard cosmological model, the CMB radiation is
80: interpreted as the blackbody radiation associated with a
81: hot dense phase of the Universe, when matter and radiation were in
82: thermal equilibrium \citep{peebles}. On large angular scales the
83: CMB radiation traces the primordial power spectrum set by physical processes
84: during the first instants after the Big Bang. On smaller angular
85: scales, CMB anisotropies are influenced by factors that control the
86: expansion rate of the Universe and formation of large-scale structure,
87: such as the cosmological constant, the matter density and the
88: existence and nature of dark matter \citep{KT94}. By
89: measuring the angular power spectrum of CMB fluctuations, one can
90: discriminate among various competing theories that predict the
91: primordial mass distribution (e.g., inflation, cosmic strings and
92: textures, primordial isocurvature baryonic perturbations) and
93: understand the gravitational collapse that ultimately brought about
94: the formation of galaxies. Since the fluctuation amplitudes at angular
95: scales of a few degrees and smaller are also sensitive to the free
96: electron distribution, CMB measurements can also be used to determine
97: the ionization history of the universe.
98:
99: After the release of the WMAP full-sky data \citep{wmap}, sub-orbital CMB
100: anisotropy experiments are
101: still of high scientific interest as they can improve angular resolution
102: and sensitivity over limited sky regions. The Background Emission
103: Anisotropy Scanning Telescope (BEAST) is the only project currently
104: on-going which is probing a frequency range overlapping with that of WMAP, with
105: improved angular resolution (up to 0.38 degrees at $\sim 40$ GHz) and potentially better
106: sensitivity over approximately 5\% of the sky. The experiment is installed
107: in a conventionally accessible, high altitude site and it has so far
108: accomplished three observing campaigns, on which this paper is based.
109: In this paper we discuss the constraints BEAST
110: places on the power spectrum of CMB
111: anisotropies and its consistency with data taken from a subset of previous experiments (MAXIMA1 \citep{maxima}, TOCO \citep{toco98}, BOOMERANG02 \citep{boomerang02}, DASI01 \citep{dasi}, VSA1 \citep{vsa}, ACBAR1 \citep{acbar}, CBI \citep{cbi}, WMAP \citep{wmap}).
112:
113:
114: We present a brief overview of the
115: experiment in \S 2 and an overview of the estimator in \S 3. \S
116: 4 details our implementation of the estimator for the BEAST data and
117: \S 5 presents the power spectrum and the parameter estimation. We
118: summarize the results in \S 6.
119:
120: \section{The BEAST Experiment}
121:
122: BEAST is a 2.2 meter off axis telescope, currently configured with an
123: 8 element mixed Q (38-45 GHz) and Ka (26-36 GHz) focal plane, and a
124: modulating flat mirror. BEAST was designed as a high altitude balloon
125: system and had two flights: May 20-21, 2000 and October 16, 2000.
126: Subsequent to the second flight BEAST was reconfigured to take
127: advantage of the UC White Mountain Research Station, Barcroft Station at
128: an altitude of 3.8 km in the Eastern Sierra of California.
129: The instrument was fully installed and
130: operational at Barcroft in July, 2001, and took data nearly
131: continuously until December 2001 (except for weather and several
132: equipment failures due to power surges and lightning). Two more weeks
133: of data were obtained in February 2002. A second data taking campaign
134: proceeded in August and September of 2002. The data used for
135: determining the power spectrum presented in this paper are taken from
136: all three of these campaigns.
137:
138: The data presented in this paper were gathered using the BEAST
139: telescope in a fixed elevation mode. The telescope is kept at a fixed
140: elevation near 90 degrees and the rotation of the Earth provides the
141: map scanning. This strategy results in a sky coverage which forms an annulus
142: centered on the NCP. The annulus is 9 degrees wide and is located between
143: 33 and 42 degrees in declination.
144:
145: Other aspects of the BEAST experiment are described in the following papers:
146: The instrument is described in \citet{childers} and a more detailed discussion of
147: the optics can be found in \citet{optics}. The map-making procedure is
148: described in \citet{map_paper} and constraints on galactic foregrounds
149: in \citet{foreground}.
150:
151:
152: \section{The MASTER Method}
153:
154: We extract the CMB power spectrum from the BEAST data using the
155: MASTER method, a binned
156: pseudo-$C_\ell$ estimator \citep{whg,Hivon1}. We chose this estimator for its ease of implementation and
157: the flexibility it offers, which allows testing the analysis with
158: a number of cuts and filtering schemes designed to remove galactic,
159: terrestrial and instrumental foregrounds.
160:
161: The MASTER method is a de-biasing
162: scheme calibrated against Monte
163: Carlo simulations. Pseudo-$C_\ell$ are calculated on the noisy maps
164: over the observed region on the sky with no corrections made for the
165: effect of this cut in terms of the
166: couplings introduced between spherical harmonic modes. The
167: expectation values of these Pseudo-$C_\ell$ are modeled in terms of an
168: ansatz which involves, as parameters, an instrumental transfer function
169: $F_\ell$ and a noise bias
170: term $N_\ell$. These terms are estimated from Monte Carlo
171: simulations of CMB signal and of experimental noise.
172:
173: The signal and noise are simulated by taking separate random realizations of
174: pure CMB signal and realistic simulations of experimental
175: noise and subjecting them separately to exactly the same data processing (such
176: as beam smoothing, scanning, cuts in the time-ordered data, filtering,
177: template removal and map-making) as the real data.
178:
179: The power spectra of the resulting signal and noise maps are averaged
180: over the Monte
181: Carlo runs to produce expectation values of the signal-only and
182: noise-only power spectra. These are used to compute the transfer
183: function and noise bias terms in the pseudo-$C_\ell$ estimator.
184:
185: To the extent to which the MASTER ansatz models the expectation values
186: of the pseudo-$C_\ell$ and to which our Monte Carlo procedure mimics
187: the acquisition of the real data, we are guaranteed an unbiased power
188: spectrum result.
189:
190: The experimental data is now passed through the data processing pipeline and
191: the pseudo-$C_\ell$ are calculated. Since the experiment covers
192: only a fraction of the sky, a coupling is introduced when performing
193: the spherical harmonic transforms to calculate the power spectra.
194: By calculating the mode-mode coupling kernel for the observed unmasked
195: region on the sky, it is possible to correct for this effect.
196:
197: Lastly, a binning scheme is chosen in $\ell$ for the final power
198: spectrum and a number of Monte Carlo simulations containing both signal and noise
199: are performed. The covariance matrix of the estimates is calculated
200: by computing the pseudo-$C_\ell$ estimator on these
201: simulations. The diagonal elements of the binned covariance matrix
202: are the variances of the binned power spectrum.
203:
204: \section{Implementation of MASTER for BEAST}
205:
206: In order to produce an accurate CMB power spectrum from the BEAST
207: data, a detailed knowledge of the experimental beam shape and pointing
208: is required. A residual $\chi^2$ fit of a smoothed delta function to
209: maps of Cygnus A and a best fit to the flux from Cygnus A lead us to
210: characterize the beam as circularly symmetric, with an effective FWHM of
211: $23^{\prime}\pm1^{\prime}$. We use the pointing information reconstructed from a
212: pointing model, which is included in the raw data files, to project
213: our simulations onto the sky in the same manner that the real data is
214: scanned.
215:
216: A total of 682 individual hours of experimental data are used
217: for the analysis. The data are naturally divided into 55 minute
218: sections by our hourly calibration cycles. These hourly sections are
219: a useful size for several reasons. In addition to the natural delineation
220: by calibrations, 55 minutes is a very manageable size for manipulation in
221: the \emph{IDL} software package on a desktop computer. Also, sky rotation over
222: one hour at our observing angle provides redundant scanning over
223: a nearly symmetric sky patch. The most important effect of this choice of time
224: slices is on our 'template removal' described below.
225: We tested the sensitivity of our results to varying the timescale of our
226: template removal from the fiducial hour down to a minimum
227: (set by sky rotation) of 600 seconds, and observed no significant changes.
228:
229: The data has been inspected and spurious signal events, e.g. due to aircraft,
230: have been removed. The data includes both the signal measured by the
231: experiment and the experimental pointing at that instant. This information
232: is used to construct a sky map of the observed signal. For all the maps
233: created in the data analysis we use the HEALPix
234: \footnote[1]{http://www.eso.org/science/healpix/} \citep{healpix} pixelization scheme with an
235: \emph{nside} parameter of 512. This results in a map containing 3,145,728
236: pixels. Given the size of the experimental beam and the high sampling frequency which is possible with a ground based instrument (450Hz for BEAST), the effects of pixel smoothing are negligible and are ignored here. For the experimental data we create a HEALPix map and
237: calculate the CMB power spectrum using the HEALPix \emph{anafast} package.
238: Further details of the map making process
239: can be found in \citet{map_paper}. Fig. \ref{fig:flow} shows an overview of the steps in the BEAST simulation and analysis pipeline.
240:
241: The WMAP \citep{wmap} best-fit theoretical power spectrum is used to create
242: random realizations of the pure CMB sky.
243: We tested the BEAST pipeline with the power spectra from two fiducial
244: cosmological models and found the final power spectrum to be unchanged by
245: this choice. The first model was a set of
246: reasonable current estimates for cosmological parameters prior to the WMAP
247: data release and the second was the best-fit power spectrum published by the
248: WMAP team.
249:
250: We scan these signal maps using our experimental pointing strategy read
251: from the time-ordered data (TOD) files. We expect the
252: time-averaged atmospheric contributions
253: to the data to vary with elevation. To remove this
254: foreground we fit a function of elevation angle to the TOD for each hour and subtract it from the TOD samples. Subsequently a
255: 10Hz high pass filter is used. The simulation has now been subjected to
256: exactly the same scanning and filtering as the real BEAST data and we project
257: this simulated data back onto a sky map.
258:
259: A foreground mask is applied to remove the Galaxy and point
260: source contamination from known sources. We remove from the analysis all
261: pixels with latitude $b\le17.5^\circ$. We tested the analysis pipeline with a range of
262: galactic latitude cuts and found that below $b=17.5^\circ$ there was significant
263: galactic foreground contamination. In addition to this, a separate analysis of the Galactic foregrounds for the BEAST experiment \citep{foreground} showed that $b \le 17.5^\circ$ gives an optimal compromise between maximizing the sky fraction observed by the experiment and minimizing the amount of foreground contamination. In this work it was also found that
264: residual Galactic foregrounds outside the mask are small and they are ignored here.
265:
266: Finally a power spectrum is generated from each signal map and
267: these power spectra are averaged to produce an average signal-only
268: power spectrum.
269:
270: To construct noise-only maps we subtract our signal estimate for the
271: map from each sample in the experimental TOD and assume that each
272: hourly segment of experimental data is now noise-dominated.
273: We further assume the noise to be piecewise stationary
274: over one hour sections of data and that each one hour noise chunk
275: is independent. We estimate noise power spectra using a windowed FFT
276: on each hourly segment \citep{NR}. We are then able to generate synthetic
277: noise simulations which have the same power spectrum as the actual
278: noise from the experiment. We filter the simulated noise TOD in the
279: same manner as for the data and signal simulations and project the noise
280: onto a sky map, then calculate the average noise power spectrum.
281: Comparisons of the data map and the maps created in the simulation
282: pipeline are shown in Fig.\ref{fig:4_maps}.
283:
284: Since we have all of the pointing information, we can also create the
285: experimental window function on the sky. This is a simple geometrical
286: construction which is 1 for any HEALPix pixel which the experiment
287: observes and 0 elsewhere. We use this window function to calculate
288: the mode-mode coupling kernel, $M_{\ell\ell '}$, which depends only on the
289: geometry of the observed region of sky. We use the ansatz for the expected
290: pseudo-$C_\ell$ which was proposed in \citet{Hivon1}. From the
291: signal-only simulations we can calibrate the transfer function
292:
293: $$F_\ell=M_{\ell\ell'}^{-1}\langle{C_\ell}_s\rangle \langle{C_\ell}\rangle^{-1} (B_\ell^2)^{-1},$$
294: where $\langle{C_\ell}_s\rangle$ are the signal-only pseudo $C_\ell$ and
295: $ \langle{C_\ell}\rangle$ are the best-fit theory $C_\ell$ from the WMAP
296: experiment. $B_\ell$ is the experimental beam, a Gaussian with FWHM
297: of 23 arcmin in this case. Since the coupling kernel is ill conditioned we
298: use an iterative approach for computing
299: $M_{\ell\ell'}^{-1}\langle{C_\ell}_s\rangle$. The transfer
300: function for the BEAST experiment is shown in Fig. \ref{fig:tf}.
301:
302: Now our $C_\ell$ estimate is given by
303: $$\hat{C_\ell} = {{M_{\ell\ell'}^{-1} \tilde{C_\ell} - \langle\tilde{N_\ell}\rangle} \over {F_\ell B_\ell^2}}$$
304: where $\langle\tilde{N_\ell}\rangle$ are the pseudo-$C_\ell$ from the noise Monte Carlo simulations and $\tilde{C_\ell}$ are the pseudo-$c_\ell$ from the data.
305:
306: In practice we use the binned
307: version of the above equation as given in \citet{Hivon1}. The binned
308: mode-mode coupling kernel is shown in Fig. \ref{fig:mm}.
309:
310: By averaging the power spectrum over bins in $\ell$ we effectively reduce correlations
311: between the $C_\ell$ bins which were introduced by the sky cut and
312: we also reduce the errors on the resulting power spectrum estimator. We have
313: tried different
314: binning schemes and choose a bin width of $\Delta\ell=55$.
315:
316: Finally, we create sky simulations by adding the signal and the
317: noise maps, produced as described above. The covariance
318: matrix $C_{bb'}$ of the binned power spectrum is calculated from
319: these simulations and the diagonal elements give us the error bars
320: on the binned power spectrum estimator. The power spectrum obtained from
321: this process is discussed in the next section.
322:
323: The code for the BEAST analysis pipeline was written and executed
324: on an IBM SP RS/6000 (\emph{Seaborg}) at the National Energy Research
325: Scientific Computing Center. The code was
326: parallelized using MPI and ran on 640 processors. In order to
327: obtain a stable PS estimate and to estimate our error bars to $\sim$20\%
328: accuracy we required 40 Monte Carlo runs. The operation count for our
329: analysis pipeline scales approximately as $N_{tod}log(N_{tod})$ with a
330: large prefactor, where $N_{tod}$ is the number of samples in the TOD.
331:
332: In order to minimize the computational time, we modified the Healpix
333: routines \emph{synfast} (which makes a sky map from a power
334: spectrum) and \emph{anafast} (which calculates the power spectrum from a
335: sky map) so that they only performed analysis on the portion of the
336: sky where BEAST scans. Since the data set read in for the BEAST simulations
337: is $\sim$ 80GB and the output maps for 40 MC runs are $\sim$1.7TB,
338: we also implemented compression algorithms for storing the output maps on disk.
339:
340:
341:
342:
343:
344:
345: \section{Power Spectrum and Parameter Estimation}
346:
347: The CMB power spectrum extracted from the BEAST data is shown in
348: Fig. \ref{fig:pow_spec}. The values of the power spectrum are shown in
349: Table 1 . The 1-$\sigma$ error bars shown in the figure should be
350: interpreted
351: with some caution. 40 Monte Carlo simulations allow us to calculate these
352: error
353: bars to within 20\%, which is sufficient for our purposes here, but more
354: simulations
355: would lead to more accurate error bars. In addition, we use the Monte Carlo
356: simulations to calculate the transfer function ($F_{\ell}$), which is then
357: used to produce
358: the $C_{\ell}$ estimates and we use these same simulations to calculate the
359: error bar on these estimates. Therefore, our estimate of the error bars on
360: the power spectrum is not unbiased and we underestimate the size of these
361: error bars.
362: In calculating our 8 binned $C_{\ell}$ estimates, we effectively compute a binned transfer function $T_b$ and a binned noise estimate $N_b$ for each bin. We use 40 Monte Carlo simulations of noise to estimate $N_b$ and 40 signal simulations to estimate $T_b$. Based on the number of degrees of freedom used to produce these 8 binned $N_b$ and $T_b$, we estimate the bias in the error bar to be approximately 15\%,
363: of the same order as our Monte Carlo
364: uncertainty in the errors. However, since this latter effect is a systematic
365: bias, the comparison of the BEAST power spectrum estimates and the resulting
366: parameter estimates to WMAP should be taken as "worst-case" consistency
367: checks.
368:
369: A $\chi^2$ comparison of the BEAST data and the WMAP data was
370: performed. For this comparison the WMAP data was assumed to
371: have zero error. We find a $\chi^2$ parameter of 15.02. With 9 degrees of
372: freedom this means a larger value of $\chi^2$ would occur approximately 10\%
373: of the time, so the BEAST power spectrum is marginally consistent with the
374: WMAP
375: result. We show the BEAST power spectrum overplotted with the power spectra
376: from several recent experiments in Fig. \ref{fig:ps_comp}.
377:
378: After the mean power spectrum was determined, its likelihood was sampled
379: 40 times, producing 40 sample binned power spectra. The likelihood around
380: the power spectrum is not, in general, Gaussian distributed, but through
381: a change of variables - to the log-offset-normal variables of
382: Bond, Jaffe and Knox (BJK parameterization \citep{BJK}) - the distribution can
383: be
384: mapped into one that is much more nearly Gaussian. However, it was found
385: that 40 samples of the power spectrum distribution was too few for a
386: reliable determination of the BJK parameters, and thus it was decided
387: that the power spectrum likelihood would be approximated as
388: Gaussian-distributed. We then calculate the Likelihood $L$ of a theoretical
389: power
390: spectrum, $D^{th}_i$, as follows:
391:
392: $ \chi^2 = \sum_{ij} (D^{th}_i - D^{ob}_i) M_{ij} (D^{th}_j - D^{ob}_j) $
393:
394: $ L = \exp(- \chi^2 / 2) $
395:
396: $ D^{ob}_i \equiv C^{ob}_i l(l+1)/2\pi $
397:
398: where $C^{ob}_i$ is the observed band-power of the i-th bin, and $M_{ij}$ is
399: the covariance matrix.
400:
401: We determined the best-fit (maximally likely) points in parameter space for:
402:
403: 1) BEAST data + WMAP + MAXIMA1,MAT98,BOOMERANG02,DASI01,VSA1,ACBAR1,CBI +
404: Hubble Key Project + Big Bang Nucleosynthesis relation between $\Omega_B h^2$ and $He^4$
405: mass-fraction ($Y_p$) \citep{hcw} over the parameter space:
406: $\Omega_m$, $\Omega_{\Lambda}$, $h$, $n_s$, $\Omega_B h^2$, $Y_p$,
407: $\tau$,
408: $n_t$, $r$
409:
410: 2) BEAST data alone.
411:
412: For 1) BEAST data + other recent cosmological data, we found the parameter values
413: and errors via a Markov chain approach. Starting from a 30000 point Markov chain
414: previously run with the experiments: WMAP + MAXIMA1,MAT98,BOOMERANG02,DASI01,VSA1,ACBAR1,CBI +
415: Hubble Key Project + BBN $\Omega_B h^2$-$He^4$ relation, the Markov chain was
416: thinned by discarding 99 out of every 100 points. Each point was then weighted
417: by the Beast likelihood. From this weighted point distribution the parameter means
418: and variance matrix were determined. The parameter estimates were taken to be the means,
419: and the parameter errors were taken as square roots of the diagonal elements of
420: the variance matrix.
421:
422: For 2) BEAST data alone,
423: cosmic parameter space was searched for the maximally likely point by first
424: trying several candidate points, and then applying the Numerical Recipes
425: Amoeba algorithm \citep{NR} to minimize the trial $\chi^2$. The Amoeba
426: algorithm has no inherent minimum scale (similar to adaptive mesh refinement,
427: the resolution increases as necessary, with the precision limited only by
428: the machine floating point arithmetic), and makes no assumptions about the
429: shape of likelihood function.
430:
431: Once the BEAST-alone best-fit cosmic parameters have been found, we determined the errors
432: in these values. Ideally a method that, again, does not depend on the
433: parameter likelihood function having a particular shape (ie: Gaussian,
434: for example), such as a Markov Chain algorithm, would be used. In this case
435: however, a less computationally costly method can be employed. We determined
436: the errors in the best-fit parameter values by fitting the likelihood
437: function around that point to a multivariate Gaussian. The resulting estimate
438: of the errors is crude, but sufficient to give an overall measure of the
439: dispersion.
440: To the extent to which the likelihoods are approximately Gaussian in the
441: narrowly constrained case (1) we expect these errors to be more accurate.
442: The results of the
443: joint parameter estimation for BEAST plus other experiments are shown in
444: Table 2. We found the BEAST alone parameters to be consistent with these
445: values, although much less well constrained. For example we found
446: $\Omega_k\equiv 1-\Omega_{tot}=-0.074 \pm 0.070$ for BEAST alone compared
447: with $-0.014 \pm 0.012$ for the joint estimate.
448:
449: In order to examine possible future directions for the BEAST
450: experiment we analyzed the effect of increased quantities of data on
451: the power spectrum error bars. A two- and four-fold reduction in the
452: simulated noise were considered, equating to four and sixteen times
453: more data respectively, assuming no improvement in radiometer sensitivity. We found that over the first peak in the
454: power spectrum there was not a significant improvement in the error
455: bars with more data (see Fig \ref{fig:error_bars}). This is expected, since in this region we are
456: sample variance limited by the relatively small patch of sky we observe. However, at
457: larger $\ell$ we do see a significant improvement in the power spectrum error
458: bars, up to the point where the experimental beam cuts off around an
459: $\ell$ of 600, when the error bars become large regardless of the
460: amount of data.
461:
462:
463: \section{Conclusions}
464:
465: We have presented the angular power spectrum of the cosmic microwave
466: background as measured by the BEAST experiment. We have demonstrated
467: that it is possible to extract cosmological signal from
468: an easily accessible, ground based CMB experiment which is dominated
469: by correlated noise and that the resulting power spectrum and parameter
470: estimation is consistent with previous results.
471:
472: The MASTER method was successfully implemented and although this method
473: is approximate, it proved to be flexible and robust and
474: produced a power spectrum with less than 1000 CPU hours of computational
475: time. We believe the BEAST CMB dataset to be one of the largest
476: TOD's analyzed to date and
477: this proved feasible within the MASTER framework. This suggests that the
478: analysis of future, larger CMB datasets (e.g. Planck) should be
479: computationally feasible.
480:
481: We also analyzed how additional observing time would improve the power spectrum errors
482: and found that significant improvements could be made
483: between $250 \le l \le 600$ with additional time. We note that the atmospheric conditions
484: at White Mountain allow for a better than 50\% 'good observing' fraction over the year and that the 26 days of data presented here were limited by funding constraints.
485:
486: \acknowledgements{}
487:
488: This work was partially supported by the University of Illinois at
489: Champaign-Urbana. This work has been partially supported by the National
490: Computational Science Alliance under grant number AST020003N. This work was funded by NASA grants NAG5-4078, NAG5-9073 ,
491: and NAG5-4185 , and by NSF grants
492: 9813920 and 0118297. In addition we were supported by the
493: White Mountain Research Station, the California Space
494: Institute (CalSpace), and the UCSB Office of Research. This research
495: used resources of the National Energy Research Scientific Computing Center, which
496: is supported by the Office of Science of the U.S. Department of Energy under
497: Contract No. DE-AC03-76SF00098.
498: The research described in this paper was carried out in part at the Jet
499: Propulsion Laboratory, California Institute of Technology, under a contract
500: with the National Aeronautics and Space Administration. J.M is
501: supported by FAPESP grants 01/13235-9 and 02/08471-1. T.V. and
502: C.A.W were partially supported by FAPESP grant 00/06770-2. T.V was
503: partially supported by CNPq grants 466184/00-0 and 302266/88-7-FA. CAW
504: was partially supported by CNPq grant 300409/97-4-FA and FAPESP grant
505: 96/06501-4. We acknowledge the use of the Legacy Archive for Microwave Background
506: Data Analysis (LAMBDA). Support for LAMBDA is provided by the NASA
507: Office of Space Science. N.F and A.P. were partially supported by CNPq grant number 470531/2001-0. B.D.W. acknowledges the 2003/4 NCSA Faculty Fellowship.
508: Some of the results in this paper have been derived using HEALPix \citep{healpix}.
509: We would like to thank Julian Borrill at NERSC for valuable discussions on the
510: computational aspects of this project. We also thank members of the Planck community
511: for stimulating discussions.
512:
513: \begin{figure}
514: %\figurenum{fig:flow}
515: %\epsscale{.5}
516: \plotone{f1.eps}
517: \caption{\small \sl Overview of the steps in the BEAST simulation and analysis pipeline. }
518:
519: \label{fig:flow}
520: \end{figure}
521:
522: \begin{figure}
523: %\figurenum{fig:4_maps}
524: %\epsscale{.5}
525: \plotone{f2.eps}
526: \caption{\small \sl Comparison of simulated and actual BEAST maps in units of Kelvin. The noise dominated nature of the BEAST data can be seen by comparing the noise map to the BEAST data map. }
527: \label{fig:4_maps}
528: \end{figure}
529:
530:
531: \begin{figure}
532: \plotone{f3.eps}
533: \caption{\small \sl Unbinned transfer function for BEAST. Monte Carlo noise is visible, which is smoothed by the binning process. The turnover at $\ell \sim$ 550 is caused by the ill conditioned mode-mode coupling kernel.}
534: \label{fig:tf}
535: \end{figure}
536:
537: \begin{figure}
538: \plotone{f4.eps}
539: \caption{\small \sl Mode-mode coupling kernel for the BEAST experiment. The z-axis is logarithmically scaled in order to show the off diagonal elements, which decrease rapidly. The width of the diagonal is approximately 25 in $\ell$ either side of the peak. In order to avoid correlations between the bins in our final power spectrum, we therefore choose a bin width of 55 in $\ell$. }
540: \label{fig:mm}
541: \end{figure}
542:
543: \begin{figure}[t]
544: \plotone{f5.eps}
545: \caption{\small \sl CMB anisotropy power spectrum for the BEAST
546: experiment. Error bars are 1$\sigma$}
547: \label{fig:pow_spec}
548: \end{figure}
549:
550: \begin{figure}[t]
551: \plotone{f6.eps}
552: \caption{\small \sl Comparison of the BEAST determination of the CMB power spectrum with 8 other recent experiments. The 4 BEAST points over the first peak are hashed circles. These points were not used in the parameter estimation since they overlap with WMAP, which is cosmic variance limited over this range. The remaining 4 BEAST data points are solid circles.}
553: \label{fig:ps_comp}
554: \end{figure}
555:
556: \begin{figure}[t]
557: \plotone{f7.eps}
558: \caption{\small \sl We examined the effect on the power spectrum error
559: bars of increasing the quantity of data. 4 and 16 times more data
560: were considered, effectively reducing the noise by factors of 2 and
561: 4 respectively. The original error bars are plotted, followed by
562: the half and quarter noise error bars. The original error bars are
563: centered on the $\ell$ bin, while the half and quarter noise are offset
564: from the original position for illustrative purposes. In
565: the analysis all of the error bars were calculated at the same $\ell$.}
566: \label{fig:error_bars}
567: \end{figure}
568:
569:
570:
571:
572:
573: \begin{deluxetable}{llcc}
574: \tablecaption{Beast Power Spectrum Estimates}
575: \tablewidth{0pt}
576: \scriptsize
577: \tablehead{
578: Bin & Bin & Estimate in $\mu K^2$ of & 1-$\sigma$ \\
579: $\ell_{min}$ & $\ell_{max}$ & $\ell(\ell+1)C_{\ell}/2\pi$ & error \\
580: }
581: \startdata
582: 139 & 193 & 3776 & $\pm$552 \\
583: 194 & 248 & 4744 & $\pm$781 \\
584: 249 & 303 & 3597 & $\pm$782 \\
585: 304 & 358 & 3374 & $\pm$625 \\
586: 359 & 413 & 1829 & $\pm$969 \\
587: 414 & 468 & 5040 & $\pm$1571 \\
588: 469 & 523 & 711 & $\pm$3319 \\
589: 524 & 678 & 4599 & $\pm$6136 \\
590: \enddata
591: \tablecomments{\small \sl The BEAST $C_{\ell}$ estimates obtained using the MASTER method. The starting and ending values of each $\ell$ bin are shown. The $C_{\ell}$ values in the table and those shown in Fig. \ref{fig:pow_spec} are averaged over these bins.}
592: \end{deluxetable}
593:
594:
595: \begin{deluxetable}{lc}
596: \tablecaption{Cosmological Parameter Estimates}
597: \tablewidth{0pt}
598: \scriptsize
599: \tablehead{
600: Parameter & BEAST+others \\
601: }
602: \startdata
603: $\Omega_k$ & -0.014$\pm$0.011 \\
604: $\Omega_{CDM}h^2 $ & 0.094$\pm$0.012 \\
605: $\Omega_bh^2$ & 0.024$\pm$0.002 \\
606: $h$ & 0.727$\pm$0.048 \\
607: $n_s$ & 1.002$\pm$0.052 \\
608: $\tau$ & 0.154$\pm$0.074 \\
609: $Y_p$ & 0.249$\pm$0.001 \\
610: \enddata
611: \tablecomments{\small \sl BEAST parameter estimates calculated using a joint
612: analysis with other CMB data and BBN and Hubble Key Project constraints.
613: $\Omega_k\equiv 1-\Omega_{tot}$. The parameter errors were obtained from the variance of a Markov chain in parameter space.}
614: \end{deluxetable}
615:
616: \begin{thebibliography}{}
617:
618: \bibitem[Bennett et al. (2003)]{wmap}
619: Bennett, C.~L.~et al. 2003, \apjs, 148, 1
620:
621: \bibitem[Childers et al. (2003)]{childers}
622: Childers, J. et al., 2003, in preparation
623:
624: \bibitem[Figueiredo et al. (2003)]{optics}
625: Figueiredo, N. et al, 2003, in preparation
626:
627: \bibitem[G{\' o}rski, Hivon \& Wandelt (1999)]{healpix}
628: G{\' o}rski, K.~M., Hivon, E., \& Wandelt, B.~D.\ 1999, Analysis Issues for Large CMB Data Sets. Proceedings: Evolution of Large Scale Structure, Garching.
629:
630:
631: \bibitem[Grainge et al. (2002)]{vsa}
632: Keith Grainge \emph{et al.} Mon. Not. R. Astron. Soc. 000, 15 (2002)
633:
634: \bibitem[Halverson et al. (2001)]{dasi}
635: Halverson, N. W. \emph{et al.}, 2001, astro-ph/0104489
636:
637: \bibitem[Hanany et al. (2000)]{maxima}
638: Hanany, S. \emph{et al.}\ 2000, \apjl, 545, L5
639:
640: \bibitem[Hivon et al. (2002)]{Hivon1}Hivon, E., G{\' o}rski, K.~M., Netterfield, C.~B., Crill, B.~P.,
641: Prunet, S., \& Hansen, F.\ 2002, \apj, 567, 2
642:
643: \bibitem[Huey, Cyburt \& Wandelt (2003)]{hcw}Huey, G, Cyburt, R. H., Wandelt, B.D., 2003, astro-ph/0307080, Phys. Rev. D, in press
644:
645: \bibitem[Knox et al. (1998)]{BJK} Knox, L., Bond, J.~R.,
646: Jaffe, A.~H., Segal, M., \& Charbonneau, D.\ 1998, \prd, 58, 83004
647:
648: \bibitem[e.g. Kolb and Turner (1994)]{KT94}
649: Kolb, E.W. and Turner, M.S., The Early Universe, 1994, Addison-Wesley
650:
651: \bibitem[Kuo et al.(2002)]{acbar} Kuo, C.~L.~et al.\ 2002,
652: Bulletin of the American Astronomical Society, 34, 649
653:
654: \bibitem[Mej\'{\i}a et al. (2003)]{foreground}
655: Mej\'{\i}a, J. et al., 2003, In Preparation
656:
657: \bibitem[Meinhold et al. (2003)]{map_paper}
658: Meinhold, P. M, et al, ApJ, submitted. (astro-ph/0302034)
659:
660: \bibitem[Miller et al. (1999)]{toco98}Miller, A.D. \emph{et al.} 1999 Astrophys.J. 524 (1999) L1-L4
661:
662: \bibitem[Padin et al. (2001)]{cbi} Padin, S. \emph{et. al.}, ApJ 549, L1, (2001)
663:
664: \bibitem[e.g. Peebles (1993)]{peebles} Peebles, P.J.E., Principles of Physical Cosmology, 1993, Princeton University Press
665:
666: \bibitem[Press et al. (1986)]{NR}
667: Press, W.H., Flannery, B.P., Teukolsky, S.A., \& Vetter, W.T., Numerical Recipes - The Art of Scientific Computing, Cambridge University Press, Cambridge 1986
668:
669: \bibitem[Ruhl et al. (2002)]{boomerang02}Ruhl, J.E. \emph{et al.} 2002, astro-ph/0212229
670:
671: \bibitem[Wandelt, Hivon \& G\'orski (2001)]{whg} Wandelt, B.D., Hivon, E.F, G\'orski, K.M. 2001, Phys. Rev. D64, 083003
672:
673: \end{thebibliography}
674:
675:
676: \end{document}
677:
678: