1: \documentclass{aastex}
2: \usepackage{emulateapj5}
3: \usepackage{onecolfloatx}
4:
5: % no hyphens ( in emalte apj mode)
6: \lefthyphenmin=1000
7: \sloppy
8:
9: %\hyphenpenalty10000
10: %\exhyphenpenalty10000 \pretolerance10000
11: %\interdisplaylinepenalty10000
12:
13: \def\et{{\it et al.}}
14:
15: \shorttitle{New Limits on CMB Polarization}
16: \shortauthors{Hedman \et}
17:
18: \begin{document}
19:
20: \twocolumn[ %comment out if not using emulateapj
21:
22: \title{New Limits on the Polarized Anisotropy of the \\
23: Cosmic Microwave Background at Subdegree Angular Scales}
24:
25: \author{M.~M.~Hedman\altaffilmark{1,4}, D.~Barkats\altaffilmark{1},
26: J~O.~Gundersen\altaffilmark{2},
27: J.~J.~McMahon\altaffilmark{1},\\ S.~T.~Staggs\altaffilmark{1},
28: B.~Winstein\altaffilmark{3}}
29:
30: \affil{$^1$ Department of Physics, Princeton University, Princeton, NJ, 08544}
31: \affil{$^2$ Department of Physics, University of Miami, Miami, FL, 33146}
32: \affil{$^3$ Center for Cosmological Physics and Enrico Fermi Institute
33: and Department of Physics, University of Chicago, Chicago, IL, 60637}
34:
35:
36: \begin{abstract}
37:
38: We update the limit from the 90~GHz PIQUE ground-based polarimeter on the
39: magnitude of any polarized anisotropy of the cosmic microwave
40: radiation. With a second year of data, we have now limited both Q and
41: U on a ring of $1^\circ$ radius. The window functions are broad: for
42: E-mode polarization, the effective $\ell$ is $\left<\ell_E\right> =
43: 191_{-132}^{+143}$. We find that the E-mode signal can be no greater
44: than $8.4~\mu$K (95\% CL), assuming no B-mode polarization. Limits on
45: a possible B-mode signal are also presented.
46:
47: \end{abstract}
48:
49: \keywords{cosmology: cosmic background radiation ---
50: cosmology: polarization --- cosmology: observations}
51: ] %comment out if not using emulateapj
52:
53: \altaffiltext{4}{Present address: Center for Cosmological Physics and
54: Enrico Fermi Institute, University of Chicago, Chicago, IL, 60637}
55:
56: \section{INTRODUCTION}
57: \nopagebreak
58:
59: The current limits on cosmic microwave background (CMB) polarization
60: restrict the amplitude of its fluctuations to less than $10~\mu$K
61: at 95\% CL. At large angular scales, Keating \et~(2001) limit the amplitude to
62: $8~\mu$K. At subdegree angular
63: scales, the constraint from Hedman \et~(2001, hereafter H01) is
64: $10~\mu$K, while at
65: arcminute scales, Subrahmanyan \et~(2000)
66: set a limit of $10~\mu$K. Present estimates of the peak
67: polarized fluctuation amplitude are $\sim 6~\mu$K at an angular scale
68: of $\sim 0\fdg2$ ($\ell\!\sim\!950$).
69: These estimates are based on parameters gleaned from CMB temperature anisotropy
70: measurements (e.g. Pryke \et~2002; Jaffe \et~2001; Wang, Tegmark, \&
71: Zaldarriaga 2001). While CMB
72: polarization has yet to be detected, its characterization will
73: complement CMB temperature anisotropy data and impact our
74: understanding of:
75: gravitational waves from the inflationary epoch (e.g. Turner
76: 1997, Caldwell, Kamionkowski, \& Wadley 1999); peculiar velocities at
77: the surface of last scattering \citep{zalhar95}; the nature of primordial
78: perturbations (e.g. Spergel \& Zaldarriaga 1997); primordial magnetic
79: fields \citep{kosloe96}; and cosmological parity violation
80: (Lue, Wang, \& Kamionkowski 1999). Here, we report improved limits
81: derived from new data from the 2001 observing season of the Princeton IQU
82: Experiment
83: %\footnote{PIQUE measured three Stokes parameters: I, Q, and U.}
84: (PIQUE) at 90~GHz. We combine the
85: new data with data from the first observing season, and also present a
86: reanalysis of those earlier data. These data pass extensive checks for
87: systematic contamination. Future publications will report
88: results from a 40~GHz polarimeter also deployed during the 2001
89: observing season, and give details of the instrument.
90:
91:
92: \section{INSTRUMENT, OBSERVATIONS AND CALIBRATION}
93: \nopagebreak
94:
95: PIQUE has been described previously (H01). The results reported here
96: are from PIQUE's broadband 90~GHz
97: correlation polarimeter, which
98: underilluminates a 1.2~m off-axis parabola \citep{wol97}, resulting in
99: a beamsize of $0\fdg 235$. The 84-100~GHz bandpass is divided into three
100: subbands called S0, S1, and S2 (H01). Observations are made of
101: a ring of radius $1^\circ$ around the NCP; the telescope site is
102: Princeton, NJ.
103:
104: The polarimeter observed the sky from
105: 2000 January 19 to 2000 April 2 and from 2000 December 19
106: to 2001 February 28. These two observing seasons yielded 810 hrs and
107: 660 hrs of raw data, respectively.
108:
109: The scanning strategy is designed to permit null tests for checking
110: sensitivity to systematics.
111: During both observing seasons, the telescope alternated between
112: two azimuth positions at fixed elevation. Data from the two
113: positions are differenced to remove sensitivity to DC offsets.
114: For the first season
115: these azimuth positions were $\pm 0\fdg 93$, and the elevation
116: was $41\fdg 0$. The telescope therefore
117: measured $\mp Q$ (as defined by the IAU) for two regions
118: separated by six hours in right ascension (RA) on the
119: ring of declination $89^\circ$. For the second season the azimuth
120: positions were $\pm 1\fdg 31$, with elevation $40\fdg 3$, so
121: the telescope measured $+U$ for two regions
122: separated by twelve hours in RA on the same
123: ring. The azimuth chop period was 13 seconds until 2001 January 23,
124: at which point it was doubled.
125:
126: The polarimetry channels are calibrated to 10\% using a nutating aluminum flat
127: (H01, Staggs \et~2002). Constant elevation scans of Jupiter
128: are used to
129: determine pointing accuracy and map the beams. For the second observing
130: season the measured beam FWHM are $0\fdg 235(7)$ in co-elevation and
131: $0\fdg 233(7)$
132: in elevation, in agreement with measurements from the first season.
133: The absolute pointing offsets in elevation are smaller than
134: $0\fdg 03$. However, early in the second observing season, the
135: encoder suffered a misalignment so that the azimuth offset increased
136: from $+0\fdg 03$ to $+0\fdg 08$. (No concomitant change in
137: beamshape was observed.) Note that we are able to neglect this in the
138: analysis because the overlap between the ideal and misaligned beams is
139: still 85\%. In fact, simulations
140: indicate the misalignment has less than a $2\%$ effect on our derived limit.
141:
142: \begin{figure}[ht]
143: %\epsscale{0.9}
144: \plotone{f1.eps}
145: \caption{\small
146: Binned data are shown in thermodynamic units for each observation
147: season. The three frequency channels have been coadded and the one
148: sigma errors include small correlations among the channels. For
149: clarity only 24 approximately beam-sized bins are shown: the
150: actual analysis uses 144 independent bins.
151: Due to PIQUE's differencing strategies, a 1~Jy point source with
152: 20\% polarization located at
153: $\alpha=12^h$, $\delta=89^{\circ}$ generates
154: the point source responses shown in the
155: two panels. No such point source is known, and
156: Toffolatti \et~(1998) estimate there are fewer than 200 point sources
157: brighter than 1~Jy in the entire sky at 90~GHz.}
158: \label{fig:datalike}
159: \end{figure}
160:
161:
162: \section {DATA REDUCTION}
163: \nopagebreak
164:
165: The 1470 hours of data from the two observing seasons include: 383 hours
166: of data taken while the telescope was slewing between the desired scan
167: positions, 339 hours of data corrupted by known electromechanical failures,
168: and 59 hours of data in isolated fragments less than 12 hours in length.
169: The remaining 685 hours of data contain 186 hours of data corrupted by
170: meteorological phenomena (clouds), which are identified using the
171: selection criteria described below.
172:
173: As discussed in H01, the correlation polarimeter suffers a small
174: sensitivity ($\la 0.5$\%) to total power because signals reflected
175: from the input of the amplifier in one arm can couple into the other
176: arm through the orthomode transducer. Therefore, the distributions of
177: correlation coefficients
178: \begin{equation}
179: {\cal C}_N=\frac{<S_i S_j>}{<S_i^2><S_j^2>}
180: \end{equation}
181: between pairs of
182: polarimetry channels $(i,j)$ display large
183: positive tails due to periods of rapid atmospheric fluctuations.
184: Data corrupted by clouds are removed by requiring the coefficients to
185: be less than certain thresholds. Such selection criteria are
186: determined based upon a data set designed to be insensitive to real
187: astronomical
188: polarized signals: the quadrature data (for
189: which data from each scan position are split into halves and
190: differenced).
191: The ${\cal C}_N$ are generated as averages over $N$ chops; varying $N$
192: varies the time scale probed. For purely Gaussian noise, the shapes of the
193: resulting distributions of ${\cal C}_N$ for the whole data set depend
194: only on $N$. In order to avoid using additional
195: cutting measures (such as the 6-hr null test used in H01), the
196: ${\cal C}_N$-selection technique has been refined from that
197: used for H01. Here, two time scales are used rather than one.
198: First we calculate
199: the average correlation coefficients for segments of the time series
200: 40-70 minutes long (specifically, $N=200$). Segments with
201: coefficients larger than a threshold of 0.20 are removed.
202: Next, coefficients are
203: calculated for $N=50$, for which the cut threshold is
204: 0.32. These thresholds are at $2.5\sigma$ and $2\sigma$,
205: and are chosen so that either cut alone removes $\sim 20$\% of the
206: data. The combined cuts remove $ 27$\% of the data. Null test
207: results are not sensitive to the exact values for the thresholds.
208:
209: The 307(192) hours of data surviving the cuts detailed above
210: from the first(second)
211: observing season are parsed into 144 bins based on the Local Sidereal Time
212: (LST) when the data were taken, following the same procedures outlined in
213: H01. These data are plotted in Figure~\ref{fig:datalike}. Offsets on
214: the order of a few hundred $\mu$K are
215: removed from each
216: polarimetry channel for each ``deployment" (a period of $> 12$ hours
217: bracketed by periods when the instrument was tarped). The results are
218: not sensitive to the exact number of offsets removed.
219: Table~\ref{ta:chisq} presents
220: the results of the null tests described in H01, using the new
221: selection criteria. The $\chi^2$ distribution of these null tests is
222: consistent with noise, demonstrating that
223: the data do not suffer from residual atmospheric contamination.
224:
225: \begin{deluxetable}{llrrr}
226: \tablecaption{Results of $\chi^2$ consistency tests.\label{ta:chisq}}
227: \tablecolumns{5}
228: \tabletypesize{\small}
229: \tablewidth{3truein}
230: \tablehead{ \colhead{Year}
231: & \colhead{Test} & \colhead{\phantom{0000}S0\tablenotemark{a}}
232: & \colhead{\phantom{0000}S1\tablenotemark{a}}
233: & \colhead{\phantom{0000}S2\tablenotemark{a}} \\
234: }
235: \startdata
236: &Quad\tablenotemark{b} & 0.96 & 0.25 & 0.19 \\
237: &H1-H2\tablenotemark{c} & 0.18 & 0.14 & 0.33 \\
238: \raisebox{1.5ex}[0pt]{2000} &Pattern\tablenotemark{d} & 0.46 & 0.26 & 0.14 \\
239: % & \colhead{S0-S1} &\colhead{S0-S2} & \colhead{S1-S2} \\
240: %$S_i-S_j$
241: &Si-Sj\tablenotemark{e} & 0.62 & 0.83 & 0.83 \\ \hline
242:
243: &Quad\tablenotemark{b} & 0.38 & 0.61 & 0.28 \\
244: &H1-H2\tablenotemark{c} & 0.17 & 0.61 & 0.58 \\
245: \raisebox{1.5ex}[0pt]{2001}&Pattern\tablenotemark{d} & 0.80 & 0.12 & 0.38 \\
246: % & \colhead{S0-S1} &\colhead{S0-S2} & \colhead{S1-S2} \\
247: %$S_i-S_j$
248: &Si-Sj\tablenotemark{e}& 0.47 & 0.66 & 0.04\\
249: \enddata
250: \tablenotetext{a}{\footnotesize Each numerical entry gives the
251: probability of exceeding the $\chi^2$ for the given frequency
252: channel.}
253: \tablenotetext{b}{\footnotesize
254: The quadrature test uses data from each scan position (east and west) split into two halves and differenced to yield the quantity
255: $(E_1-E_2)/2+(W_1-W_2)/2$.}
256: \tablenotetext{c}{\footnotesize Data from the second half of each
257: season are subtracted from the first half.}
258: \tablenotetext{d}{\footnotesize Pattern nulls are generalizations of
259: the 6-hour null test
260: from H01, and are data sets constructed from the various
261: combinations of the data that should be zero given the differencing
262: scan strategy. If $d_t$ is the measured signal at Local Sidereal Time
263: $t$ in hours,
264: then for 2000 these combinations are $d_t-d_{t+6}+d_{t+12}-d_{t+18}$,
265: while for 2001 these combinations are $d_t+d_{t+12}$.}
266: \tablenotetext{e}{\footnotesize Data differenced between two
267: channels. The column entries are S0-S1, S0-S2 and S1-S2.}
268:
269: \end{deluxetable}
270:
271: \begin{figure}
272: \plotone{f2.eps}
273: \caption{\small
274: Zero-lag window functions for E- and B-modes (solid and dashed lines)
275: are shown in the middle panel for the 2000 observation season and in
276: the lower panel for the 2001 season. The top panel shows
277: the limit on $T_E$ assuming $T_B\equiv 0$ for the combined W-band
278: observations.
279: For comparison we also plot (dashed line) E-mode predictions from the best-fit
280: model in Pryke \et~(2002), as well as B-mode predictions (shown in the
281: top panel multiplied by ten) assuming the same model with $T/S=1$.}
282: \label{fig:jogmega}
283: \end{figure}
284:
285:
286: \section{DATA ANALYSIS}
287: \nopagebreak
288:
289: The likelihood of a model given a data vector ${\bf x}$ is
290: ${\cal
291: L}\propto\exp(-{\bf{x}}^T{\bf{C}}^{-1}{\bf{x}}/2)/|{\bf{C}}|^{1/2}$,
292: where
293: the covariance matrix ${\bf C}$ sums both theoretical correlations
294: from the model (signal) and correlations specific to the experiment
295: (noise): ${\bf C} = {\bf S} + {\bf N}$.
296: The analysis of the combined 2000-2001 data uses ${\bf x}$ with
297: 864 elements
298: comprising 144 spatial pixels measured in three different
299: frequency bands over the course of two observing seasons.
300: The noise matrix ${\bf N}$ encodes the variances for each
301: pixel for each channel and also accounts for interchannel
302: correlations from both atmospheric fluctuations weakly coupled into the
303: polarimeter channels and correlated gain fluctuations in the
304: cryogenic amplifiers.
305: The interchannel correlation coefficients are $\le 8$\%
306: on average, and smaller for the two most sensitive channels. (S2
307: has just $20$\% of the total weight.)
308: The noise matrix does not include pixel-pixel correlations, since
309: no such correlations are observed in the time series data.
310: An independent analysis using a 288-element data vector, for which
311: data from the three
312: frequency channels are combined for each spatial pixel, with errors
313: calculated to account for the interchannel correlations, yields
314: consistent results.
315: % the likelihood figure was here and STS moved it to later.
316: %this table used to be later in the text. STS.
317: \begin{deluxetable}{clrrr}
318: \tablecaption{$95\%$ Confidence Limits \label{ta:likes}}
319: \tablecolumns{5}
320: \tabletypesize{\footnotesize}
321: \tablewidth{3truein}
322: \tablehead{\colhead{Year}&\colhead{{Data}\tablenotemark{a}} &
323: \colhead{$\widetilde{T}_E$\tablenotemark{b}}&
324: \colhead{\phm{00}$\widehat{T}_E$\tablenotemark{c}} &
325: \colhead{\phm{00}$\widehat{T}_B$\tablenotemark{c}} \\
326: \colhead{} & \colhead{} &\colhead{$(\mu$K)} & \colhead{\phm{00}$(\mu$K)} &
327: \colhead{\phm{00}$(\mu$K)} \\
328: }
329:
330: \startdata
331: &CMB & 12.7 & 15.8 &14.7 \\
332: &Quad & 10.3 & 13.9 & 12.9 \\
333: \raisebox{1.5ex}[0pt]{2000}
334: &(H1-H2)/2 & 14.9 &19.7 &17.4 \\ \hline
335:
336: &CMB & 10.4 &15.9 &17.8 \\
337: &Quad & 10.4 & 16.2 &18.0 \\
338: \raisebox{1.5ex}[0pt]{2001}
339: &(H1-H2)/2 & 11.2 &17.2 &19.4 \\ \hline
340:
341: 2000 &CMB & 8.4 & 11.2 &11.5 \\
342: $+$ &Quad & 6.1 & 8.3 & 8.4 \\
343: 2001&(H1-H2)/2 & 8.5 &11.5 & 11.6 \\ \hline
344: \enddata
345:
346:
347: \tablenotetext{a}{\footnotesize The two null tests are
348: described in Table~\ref{ta:chisq}.}
349: \tablenotetext{b}{\footnotesize The limit $\widetilde{T}_E$ is found
350: assuming $T_B\equiv0$.}
351: \tablenotetext{c}{\footnotesize The limits $\widehat{T}_E$ and
352: $\widehat{T}_B$ are determined simultaneously
353: by finding the contour of constant likelihood enclosing
354: 95\% of the volume. }
355:
356: \end{deluxetable}
357:
358:
359: Since PIQUE measured $Q$ in 2000 and $U$ in 2001,
360: the signal covariance matrix takes the form:
361: \begin{equation}
362: {\bf{S}}= \pmatrix{\langle\tilde{Q}\tilde{Q}\rangle &
363: \langle\tilde{Q}\tilde{U}\rangle \cr
364: \langle\tilde{Q}\tilde{U}\rangle &
365: \langle\tilde{U}\tilde{U}\rangle}
366: \end{equation}
367: where
368: $\langle\tilde{Q}_i\tilde{Q}_j\rangle$
369: (or $\langle\tilde{U}_i\tilde{U}_j\rangle$) represents the
370: theoretical correlation between two spatial pixels from the 2000
371: (or 2001) data
372: set, and $\langle\tilde{Q}_i\tilde{U}_j\rangle$ encodes correlations
373: between pixels from different years.
374: Note that the $\tilde{Q}$ are sums of Q separated by six hours in
375: RA and the $\tilde{U}$ are differences of U separated
376: by 12 hours in RA.
377: The expression for
378: $ \langle\tilde{Q}_i\tilde{Q}_j\rangle$ is given in
379: H01. Following Zaldarriaga (1998), the expression for
380: $\langle\tilde{Q}_i\tilde{U}_j\rangle$ is given in terms
381: of the E- and B-mode angular power spectra $C^E_\ell$ and $C^B_\ell$ by
382:
383: \begin{equation}
384: \langle\tilde{Q}_i\tilde{U}_j\rangle =
385: \sum_{\ell m}{ \frac{(2\ell +1)}{4\pi}}[C_\ell^E
386: +C_\ell^B]W_{12,\ell}(\phi_{ij}),
387: \end{equation}
388: where
389: $\phi_{ij}=\cos^{-1}(\hat{n}_i\cdot \hat{n}_j)$ is the lag.
390: The window function $W_{12,\ell}$ has the form:
391: \begin{equation}
392: W_{12,\ell}=\sum_m (B_{lm}^{QU})^2F_{1,\ell m}
393: F_{2,\ell m}\cos(m\phi_{ij}),
394: \end{equation}
395: where the $F_{\{1,2\},lm}$ are given in terms
396: of associated Legendre polynomials evaluated at the ring radius
397: $\theta=1^{\circ}$.
398: Here, the beam function is:
399: \begin{equation}
400: (B_{\ell
401: m}^{QU})^2=4\sin\left(\frac{m\pi}{2}\right)\cos\left(\frac{m\pi}{4}\right)\frac{\sin^2(m\pi/144)}{(m\pi/144)^2}
402: e^{-\ell(\ell+1)\sigma^2},
403: \end{equation}
404: where $\sigma=0\fdg 10$ for the PIQUE beams.
405: Similarly, the expression for $\langle\tilde{U}_i\tilde{U}_j\rangle$
406: is:
407: \begin{equation}
408: \langle\tilde{U}_i\tilde{U}_j\rangle =
409: \sum_{\ell}\frac{(2\ell +1)}{4\pi}[
410: C_\ell^E W_{2,\ell} (\phi_{ij}) +
411: C_\ell^B W_{1,\ell}(\phi_{ij})],
412: \end{equation}
413: where the $W_{1,\ell}$ and $W_{2,\ell}$ are the associated window
414: functions given in H01. In this case, the beam function is given by
415: \begin{equation}
416: (B_{lm}^{UU})^2=(B_{lm}^{QU})^2\frac{\sin(m\pi/2)}{\cos(m\pi/4)}.
417: \end{equation}
418: This formalism allows for
419: combination of the two data sets; however, separate analyses of
420: the $\tilde{Q}$ data and the $\tilde{U}$ data are
421: also presented. We plot the zero-lag window functions in
422: Figure~\ref{fig:jogmega}.
423:
424: %The likelihood figure used to be earlier in the text. STS.
425: \begin{figure}
426: \plotone{f3.eps}
427: \caption{\small
428: Normalized likelihoods versus flat-bandpower $T_E$ under the
429: assumption $T_B=0$, as described in
430: the text, for each year of data, and for the combined data.
431: In each panel, likelihoods for the CMB data are plotted
432: along with likelihoods for two of the null data sets described in
433: Table~\ref{ta:chisq}.
434: }
435: \label{fig:likes}
436: \end{figure}
437:
438: \setcounter{footnote}{0}
439:
440: The likelihood analysis proceeds by considering flat
441: angular spectra, such that $\ell(\ell+1)C_\ell^X/2\pi = T_X^2$,
442: where $X=E, B$. Since the
443: amplitude of $C^B_\ell$ is predicted to be much smaller than that
444: of $C^E_\ell$, we first find the limit $\widetilde{T}_E$ on
445: $T_E$ under the assumption $T_B$ is identically zero\footnote{The
446: limit $\widetilde{T}_E$ is found by
447: integrating ${\cal L}(T_E,0)$; the result is $\approx 30$\% higher
448: if ${\cal L}(T_E^2,0)$ is integrated.}. This limit is compared to
449: predictions in Figure~\ref{fig:jogmega}. Next, joint
450: upper limits $(\widehat{T}_E, \widehat{T}_B)$ are determined by finding the
451: constant contour of ${\cal L}(T_E,T_B)$ enclosing 95\% of the
452: volume of ${\cal L}$. This is repeated for each year separately and for
453: the
454: combination of the two years. The 95\% confidence level (CL) upper limits
455: are
456: shown in Table~\ref{ta:likes} and the normalized likelihoods are shown in
457: Figure~\ref{fig:likes}.
458: The null data sets are treated in an identical manner; results are
459: tabulated in Table~\ref{ta:likes} and plotted in
460: Figure~\ref{fig:likes}. The limits in Table~\ref{ta:likes} do not include
461: calibration
462: errors. Note that for PIQUE's broad window functions,
463: the 5\% beam errors only add 2\% errors in quadrature with the 10\%
464: calibration errors.
465:
466: %The confidence limits table used to be here. STS.
467:
468:
469: \section{DISCUSSION}
470: \nopagebreak
471:
472: The main result here is a new constraint on the amount of
473: polarized anisotropy in the CMB at sub-degree angular scales.
474: The result derives from combining data on Q from our first
475: campaign with U from our second; in so doing, important
476: information on the Q-U cross-correlation is included.
477: We have summarized this result as a 95\% CL limit of $8.4~\mu$K on
478: E-modes. Given PIQUE's window functions and current theoretical
479: predictions (Figure~\ref{fig:likes}), we might expect a signal of a few
480: $\mu$K. The likelihood for our CMB data (Figure 3, bottom panel)
481: is consistent with this expectation. When we fit to an
482: offset lognormal distribution (Bond, Jaffe, \& Knox 2000), we find a central
483: value of $2.2~\mu\mbox{K}^2$, a variance of
484: $(16~\mu\mbox{K}^2)^2$ and a noise-related offset of
485: $18~\mu\mbox{K}^2$.
486:
487: The results presented here have been checked with two independent
488: likelihood analyses and supported by extensive simulations. Our new
489: selection criteria, as described above: are better able to deal with
490: instrumental effects on different time scales; allow us to discard
491: the 6-hr null test criterion we previously used; and work for both
492: data sets together. However, applying the new criteria just
493: to year 2000 data, we find a weaker 95\% CL limit on polarized CMB
494: anisotropy than in H01: 12.6~$\mu$K rather than 10.3~$\mu$K.
495:
496: We have performed a variety of simulations to address the probability of
497: such a change. Recall that our result for the year 2000 data was
498: essentially unchanged by relaxing the null test cut and allowing in 80
499: extra hours of data, for a total of 330 hours.
500: For these simulations, we start with roughly a
501: 330 hour data set which is pure noise, generated assuming the actual
502: weights for each period of the data set. We then investigate how cutting
503: the data can change the derived limit. From this we find
504: that although a) the probability of the limit not worsening when 330 hours is
505: reduced to 250 hours is only about 1\%,
506: b) a $10 ~\mu$K error from 330 hours of data is within one standard
507: deviation of what is expected from pure noise
508: (given our experimental weights); and
509: c) the expected change in the limit from removing chunks of the
510: data to get to 300 hours of data, (our final sample for year 2000), is about
511: $+1.1~\mu$K with a standard deviation of about $1.6~\mu$K. Thus our
512: observed change is again within one standard deviation of what is
513: expected.
514:
515: We thus conclude that fluctuations alone can account for the change in
516: the limit derived from the year 2000 data under different selection
517: criteria. This is
518: so even though the former cut on the null test, which we have shown is not
519: needed, was probably too restrictive.
520:
521:
522: The expected signal, even including foregrounds, is still smaller than our
523: new limit. Given the multipole range probed by PIQUE, this result
524: provides the tightest constraint yet on the polariation spectrum
525: predicted from primordial density fluctuations.
526:
527: \acknowledgements
528:
529: We thank Norman Jarosik, Lyman Page,
530: and David Wilkinson for helpful discussions and
531: Al Dietrich for mechanical contributions. We are grateful to Sami Amasha
532: and Liam Fitzpatrick for assistance with programming.
533: We also thank Marian Pospieszalski and the NRAO for supplying the
534: HEMT amplifers. Data (including the correlation matrix and
535: likelihood functions) will be made public upon publication of
536: this Letter.
537:
538: This work was supported by a NIST precision measurement grant \#NANB8D0061,
539: by NSF grants \#PHY96-00015, \#PHY99-84440, \#PHY-0114422, and by a Sloan
540: Fellowship (to STS).
541:
542: %\bibliographystyle{apj1c}
543: %\bibliography{paper2-b}
544:
545: \begin{thebibliography}{}
546:
547:
548: \bibitem[Bond, Jaffe, \& Knox 2000] {bjk00} Bond, J. R., Jaffe, A. H.,
549: \& Knox, L. 2000, ApJ, 533, 19
550:
551: \bibitem[{Caldwell}, {Kamionkowski}, \&{ Wadley} 1999]{caldwell99}
552: {Caldwell}, R.~R., {Kamionkowski}, M., \& {Wadley}, L. 1999, \prd, 59, 027101
553:
554: \bibitem[{Hedman} \et 2001]{PIQUE01}
555: {Hedman}, M.~M., {Barkats}, D., {Gundersen}, J.~O., {Staggs}, S.~T., \&
556: {Winstein}, B. 2001, \apjl, 548, L111
557:
558: \bibitem[{Jaffe} \et~2001]{jaffe01}
559: {Jaffe}, A.~H. \et~2001, \prl, 86, 3475
560:
561: \bibitem[{Keating} \et~2001]{polar01}
562: {Keating}, B.~G., {O'Dell}, C.~W., {de Oliveira-Costa}, A., {Klawikowski}, S.,
563: {Stebor}, N., {Piccirillo}, L., {Tegmark}, M., \& {Timbie},
564: P.~T. 2001, \apjl, 560, L1
565:
566: \bibitem[{Kosowsky} \& {Loeb} 1996]{kosloe96}
567: {Kosowsky}, A. \& {Loeb}, A. 1996, \apj, 469, 1
568:
569: \bibitem[{Lue}, {Wang}, \& {Kamionkowski} 1999]{lwk99}
570: {Lue}, A., {Wang}, L., \& {Kamionkowski}, M. 1999, \prl, 83, 1506
571:
572: \bibitem[{Pryke} \et~2002]{pryke02}
573: {Pryke}, C., {Halverson}, N.~W., {Leitch}, E.~M., {Kovac}, J., {Carlstrom},
574: J.~E., {Holzapfel}, W.~L., \& {Dragovan}, M. 2002, \apj, 568, 46
575:
576: \bibitem[{Spergel} \& {Zaldarriaga} 1997]{sperzal97}
577: {Spergel}, D.~N. \& {Zaldarriaga}, M. 1997, \prl, 79, 2180
578:
579:
580: \bibitem[{Staggs} \et~2002]{sta02}
581: {Staggs}, S.~T., {Barkats}, D., {Gundersen}, J.~O., {Hedman}, M.~M., {Herzog},
582: C.~P., {McMahon}, J.~J., \& {Winstein}, B. 2002, in { AIP Conf. Proc. 609:
583: Astrophysical Polarized Backgrounds}, 183
584:
585: \bibitem[{Subrahmanyan} \et 2000]{ATCA}
586: {Subrahmanyan}, R., {Kesteven}, M.~J., {Ekers}, R.~D., {Sinclair}, M., \&
587: {Silk}, J. 2000, \mnras, 315, 808
588:
589: \bibitem[{Toffolatti} \et 1998]{tof98}
590: {Toffolatt}i, L., {Argueso Gomez}, F., {de Zotti}, G., {Mazzei}, P.,
591: {Franceschini}, A., {Danese}, L., \& {Burigana}, C. 1998, \mnras, 297, 117
592:
593: \bibitem[Turner (1997)]{tenthings}
594: {Turner}, M.~S. 1997, in { NATO ASIC Proc. 503: Generation of Cosmological
595: Large-Scale Structure.}, 153
596:
597: \bibitem[{Wang},{Tegmark},\&{Zaldarriaga} (2001)]{wtz01}
598: {Wang}, X., {Tegmark}, M., \& {Zaldarriaga}, M. 2001.
599: \newblock astro-ph/0105091, accepted for publication in \prd
600:
601: \bibitem[{Wollack} \et~1997]{wol97}
602: {Wollack}, E.~J., {Devlin}, M.~J., {Jarosik}, N., {Netterfield}, C.~B., {Page},
603: L., \& {Wilkinson}, D. 1997, \apj, 476, 440
604:
605: \bibitem[{Zaldarriaga} (1998)]{zal98}
606: {Zaldarriaga}, M. 1998, \apj, 503, 1
607:
608: \bibitem[{Zaldarriaga} \& {Harari} 1995]{zalhar95}
609: {Zaldarriaga}, M.~. \& {Harari}, D.~D. 1995, \prd, 52, 3276
610:
611:
612: \end{thebibliography}
613: \end{document}
614: