BICEP1 three-year data release

2013 October 03

This page contains data corresponding to the results presented in Barkats et al. (2013), Degree-Scale CMB Polarization Measurements from Three Years of BICEP1 Data, submitted to ApJ, arXiv:1310:1422. If you have questions about the usage of these results, please contact Denis Barkats (

1. History

2. Data products

Text file containing frequency-combined (100+150 GHz) bandpowers and Fisher errors, corresponding to the filled circles in Figure 5 of Barkats et al..
Tarball containing the bandpower window function for each of the 9 ℓ bins
Tarball containing data necessary for constructing bandpower likelihoods following the method of Hamimeche & Lewis (2008). See Section 9.1 of Barkats et al. . File contains:
Text file containing 95% confidence level upper limits derived from the bandpower likelihood functions. See Section 9.1 of Barkats et al.
Text file containing the tabulated likelihood for the tensor-to-scalar ratio, r, computed using the “direct likelihood calculation” described in Section 9.3.1 of Barkats et al.
Text file containing fractional uncertainties on bandpowers due to systematic error in absolute gain calibration and beam width calibration.
Tarball containing theoretical power spectra for the E-no-B and B-no-E models, described in Section 5.3 of Barkats et al. Includes both the unbinned spectra calculated by CAMB as well as the binned expectation values of the bandpowers. File contains:
Gzipped tar archive containing python code and the relevant above data products to calculate the Hamimeche&Lewis likelihood. The python code also contains a specific wrapper to generate the HL likelihood versus r using the BB spectrum. et al..

3. Figures

Integration time, T, Q, U maps Maps of integration time, temperature, and Stokes parameters Q and U for BICEP1 at 100 and 150 GHz. Integration time is in units of seconds per square-degree. T, Q, and U maps have been lightly smoothed with a half-beam-size Gaussian, 0.465° full width at half-maximum for 100 GHz and 0.3° for 150 GHz. The E-mode polarization signal can be seen as vertically and horizontally oriented modes in the Q maps and modes oriented at ±45° in the U maps.

Single panel maps:
eps, png
T, E, B maps Figure 1 from Barkats et al.
Maps of CMB temperature and E/B-mode polarization generated from three years of BICEP1 observations. The 100 and 150 GHz maps are each smoothed to a common 1° (full width at half-maximum) beam size before taking the frequency sum (left column) and difference (right column). The E and B polarization maps have been additionally filtered to remove power outside the range 30<ℓ<200, in order to emphasize the angular scales of interest for BICEP1. As in Chiang et al., BICEP1 detects E-mode polarization with high significance while the B-mode signal is consistent with noise. The E- and B-mode frequency-difference maps are consistent with noise, indicating that they do not suffer contamination from polarized foregrounds.

Single panel maps:
  • T (frequency-sum): eps, png / T (frequency-difference): eps, png
  • E (frequency-sum): eps, png / E (frequency-difference): eps, png
  • B (frequency-sum): eps, png / B (frequency-difference): eps, png
eps, png
Suppression factor, impact of filtering on signal-to-noiseFigure 2 from Barkats et al.
Shown in the top panel is the suppression factor, F150E×150E, including contributions from beam smoothing and our filtering choices. At large angular scales, power suppression is dominated by our conservative choice of polynomial and azimuth-fixed filters, shown both individually (plus sign and downward triangles) and combined (squares). At small angular scales, beam smoothing (dotted line; calculated analytically) dominates. The total suppression factor (circles) also includes small contributions from relative gain deprojection and the pixel window function. The suppression factor should not be mistaken for a measure of the low ℓ performance of the experiment (see text). The bottom two panels show the actual impact of our filtering choices on the signal-to-noise of EE and BB bandpowers. The total sensitivity loss from polynomial plus azimuth-fixed filtering is shown by squares, with the gray shaded region indicating the 1σ uncertainty on this calculation. The plus signs and downward triangles indicate the loss of sensitivity from polynomial or azimuth-fixed filtering individually. While the two filters supress similar amounts of power in the maps (top panel), only the azimuth-fixed filter has a significant effect on signal-to-noise (bottom panels).
eps, png
Bandpower window functionsFigure 3 from Barkats et al.
BICEP1 bandpower window functions for 150 GHz BB autospectrum. The gray dashed line shows the mask window function for the first ℓ bin. Bandpower window functions for other spectra are visually similar.
eps, png
Individual frequency power spectraFigure 4 from Barkats et al.
BICEP1 individual-frequency CMB power spectra. The horizontal axis is multipole ℓ, and the vertical axis is ℓ(ℓ+1)C/2π in units of μK2CMB. Black points show the full set of BICEP1 power spectra up to ℓ=350 with statistical error bars (including sample variance) only. The spectra agree well with a ΛCDM model (black lines) derived from WMAP five-year data and r =0. The blue points correspond to the boresight-angle jackknife. The red open circles show the TT, TE, and TB spectra calculated using one half of the BICEP1 boresight-angle jackknife maps as the temperature map, as described in §6.2 of Barkats et al.
eps, png
Combined frequency power spectraFigure 5 from Barkats et al.
BICEP1 frequency-combined power spectra (black points) are in excellent agreement with a ΛCDM model (black lines) derived from WMAP five-year data. The χ2 (for nine degrees of freedom) and PTE values from a comparison of the data with the model are listed in the plots. Gray crosses denote bandpower expectation values for the model. Power spectrum results from Chiang et al. (2010) are shown by open circles and are offset in ℓ for clarity. In both cases, the error bars are the square root of diagonal elements in the frequency-combined bandpower covariance matrix described in §5.5 of Barkats et al., and do not include systematic uncertainties.
eps, png
Jackknife PTE histogramFigure 6 from Barkats et al.
Probabilities-to-exceed from χ2 tests of 50 polarization-only (EE, BB, and EB) jackknives are consistent with a uniform distribution between zero and one (dashed line).
eps, png
Statistical and systematic uncertaintyFigure 7 from Barkats et al.
Summary of statistical and systematic uncertainties for the BICEP1 three-year result. Total random (solid gray line) indicates the overall statistical uncertainty due to both instrumental noise (black dashed) and CMB sample variance (black dotted). Systematic uncertainty contributions from absolute calibration and beam width (red line), relative gain mismatch (green line), and differential pointing (blue line) are also included. Sample variance dominates the TT and TE spectra. The EE (and TB) spectra are dominated by sample variance at low ℓ and noise at high ℓ. Noise dominates the BB (and EB) uncertainty at all angular scales.
eps, png
Foregrounds estimateFigure 8 from Barkats et al.
The estimated Galactic dust and synchrotron emission in the BICEP1 field is well below current BB upper limits. These foreground emission estimates come from processing the Planck Sky Model foreground maps (Delabrouille et al. 2013) through the BICEP1 pipeline. The BB upper limits are derived from bandpower likelihoods calculated in §9.1 of Barkats et al.
eps, png
Combined frequency EE and BB spectraFigure 9 from Barkats et al.
Close-up of the EE and BB spectra from Figure 5. BICEP1 measures EE polarization (open circles) with high signal-to-noise at degree angular scales (§9.2 of Barkats et al.). The BB spectrum (black points) is consistent with zero. Theoretical BB and EE spectra with r =0.1 are shown in solid and dashed gray lines respectively. The gray crosses are bandpower expectation values for the EE spectrum. They diverge from the ΛCDM curve (especially for the first ℓ-bin) because of the detailed shape of the bandpower window functions. The inset shows the low-ℓ region in more detail.
eps, png
Likelihood for rFigure 10 from Barkats et al.
The likelihood for r calculated from the BICEP1 BB spectrum is shown in the left panel. The red curve comes from a direct likelihood calculation described in §9.3.1 of Barkats et al. The blue curve comes from an alternate calculation based on the bandpower likelihood approximation (§9.3.2). The maximum likelihood value and 1σ interval, r =0.03-0.23+0.27, are shown as the blue solid and dashed lines. A histogram of maximum likelihood values derived from 499 signal-plus-noise simulations (with r =0 input) is shown in the central panel. In the right panel, we derive 95% confidence upper limits on r from simulated likelihoods (gray histogram) and real data likelihood. BICEP1 obtains an upper limit of r <0.70 (red line), which lies within the simulated distribution. The gray dashed line shows the median (r <0.65) of the upper limits derived from simulations.
eps, png
Probability density function for ρFigure 11 from Barkats et al.
Bayesian 95% upper limit (dashed line) and Feldman-Cousins 95% confidence interval (dotted lines) on r for the BICEP1 three-year result, as a function of the value of ρ. The shaded image shows the probability density of ρ as a function of r, derived from simulations; each horizontal slice of the image yields a normalized PDF for ρ given a particular theory. Vertical slices correspond to likelihood functions for r. The solid vertical line indicates the value of ρ measured by BICEP1.
eps, png
Change in upper limit on r from two to three yearsFigure 12 from Barkats et al.
Histogram of the shift in the 95% confidence upper limit on r from simulations upon including the additional 52% data of the full BICEP1 observations. A negative value indicates that the three-year upper limit is tighter than the two-year limit, but 7% of realization show a positive value. The dashed black line indicates the median of the distribution (-0.27). The solid black line indicates the value of this shift for the real data (-0.10).
eps, png
Summary of EE and BB measurementsFigure 13 from Barkats et al.
BICEP1's EE and BB power spectra complement existing data from other CMB polarization experiments (Leitch et al. 2005; Montroy et al. 2006; Sievers et al. 2007; Bischoff et al. 2008; Brown et al. 2009; QUIET Collaboration et al. 2011; QUIET Collaboration et al. 2012; Bennett et al. 2013). For visual clarity, we only display the experiments where at least one of the EE bandpowers has a center value that is greater than twice the distance between the center value and the lower end of the 68% confidence interval. Theoretical spectra from a ΛCDM model with r =0.1 are shown for comparison; the BB curve is the sum of inflationary and gravitational lensing components. At degree angular scales, BICEP1's constraints on BB are the most powerful to date.
eps, png