Abstract
The coming flood of cosmic microwave background (CMB) polarization experiments, spurred by the recent detection of CMB polarization by the DASI and WMAP instruments, will be confronted by many new analysis tasks specific to polarization. For the analysis of CMB polarization data sets, the devil is truly in the details. With this in mind, we present details of the data analysis for the POLAR experiment, which recently led to the tightest upper limits on the polarization of the cosmic microwave background radiation at large angular scales. We discuss the data selection process, map-making algorithms, offset removal, and likelihood analysis which were used to find upper limits on the polarization. Stated using the modern convention for reporting CMB Stokes parameters, these limits are 5.0 μK on both E- and B-type polarization at 95% confidence. Finally, we discuss simulations used to test our analysis techniques and to probe the fundamental limitations of the experiment.
Introduction
While the Cosmic Microwave Background (CMB) temperature fluctuations have been mapped extensively, its polarization is far more elusive—orders of magnitude weaker and fraught with systematic errors. Detecting polarization at large angular scales (greater than a degree) is a "holy grail" for cosmologists because it contains the signature of reionization and potentially primordial gravitational waves from inflation.
The challenge is that ground-based instruments like POLAR must contend with atmospheric interference, ground pickup, and instrumental offsets that can easily mimic or drown out the cosmological signal. This paper describes the end-to-end analysis framework used to produce reliable upper limits from the POLAR data set.
Methodology and Data Selection
The POLAR instrument was a HEMT-based correlation polarimeter operating at 26–36 GHz, using a drift-scan strategy to observe the sky as the Earth rotated. Approximately 750 hours of raw data were collected near Madison, Wisconsin. However, ground-based observations in humid environments are notoriously difficult, leading to a high rejection rate; only about 10.5% of the data survived the final analysis.
Key to this pruning was a "battery of conditions" to identify contaminated data:
- The $1\phi$ Cut: In a rotating polarimeter, a true celestial signal appears at twice the rotation frequency ($2\phi$). Signals at the rotation frequency itself ($1\phi$) were identified as purely instrumental or atmospheric. By monitoring the $1\phi$ power, the team could flag periods of "bad weather" that would otherwise corrupt the $2\phi$ polarization signal.
- The $\zeta$ Cut: This used autocorrelation properties to flag low-frequency drifts.
Mapmaking and Signal Analysis
To handle the massive data sets, the researchers employed a circulant matrix approximation. This allowed them to invert noise covariance matrices in $O(n \ln n)$ time rather than the standard $O(n^3)$, making "Minimum Variance Mapmaking" computationally feasible.
A critical challenge was the discovery of slowly-varying, non-zero offsets in the Stokes $Q$ and $U$ parameters. Since the physical cause was unknown, the authors used a mathematical technique called marginalization to "ignore" the mean value of each data submap. This effectively filtered out the offsets while preserving the underlying fluctuations.
Key Results and Insights
The success of the marginalization technique proved that physical understanding of a systematic offset (like groundscreen interference) isn't always necessary to remove its bias from cosmological constraints. However, the paper also highlights a fundamental geometric limitation: the 1D drift scan resulted in significant E-B leakage. In a one-dimensional scan, $E$ and $B$ modes become nearly indistinguishable, meaning the experiment could only effectively constrain the sum or average of the two.
Final results established 95% confidence upper limits of 5.0 μK for both $E$-type and $B$-type polarization. When assuming $B$-modes are negligible (per standard models), the limit for $E$-modes was refined to $T_E < 7.7\ \mu\text{K}$.
Conclusion
POLAR set some of the most stringent limits of its time, but the high data rejection rate underscores the difficulty of ground-based CMB observations. Future experiments require 2D sky coverage to break the $E$-$B$ degeneracy and significantly longer integration times to detect the actual signal, predicted to be below 1 μK.
Summary updated