Self-Calibration of BICEP1 Three-Year Data and Constraints on Astrophysical Polarization Rotation

Cosmic Microwave Background (CMB) polarimeters aspire to measure the faint $B$-mode signature predicted to arise from inflationary gravitational waves. They also have the potential to constrain cosmic birefringence which would produce non-zero expectation values for the CMB's $TB$ and $EB$ spectra. However, instrumental systematic effects can also cause these $TB$ and $EB$ correlations to be non-zero. In particular, an overall miscalibration of the polarization orientation of the detectors produces $TB$ and $EB$ spectra which are degenerate with isotropic cosmological birefringence, while also introducing a small but predictable bias on the $BB$ spectrum. The \bicep three-year spectra, which use our standard calibration of detector polarization angles from a dielectric sheet, are consistent with a polarization rotation of $\alpha = -2.77^\circ \pm 0.86^\circ \text{(statistical)} \pm 1.3^\circ \text{(systematic)}$. We revise the estimate of systematic error on the polarization rotation angle from the two-year analysis by comparing multiple calibration methods. We investigate the polarization rotation for the \bicep 100 GHz and 150 GHz bands separately to investigate theoretical models that produce frequency-dependent cosmic birefringence. We find no evidence in the data supporting either these models or Faraday rotation of the CMB polarization by the Milky Way galaxy's magnetic field. If we assume that there is no cosmic birefringence, we can use the $TB$ and $EB$ spectra to calibrate detector polarization orientations, thus reducing bias of the cosmological $B$-mode spectrum from leaked $E$-modes due to possible polarization orientation miscalibration. After applying this "self-calibration" process, we find that the upper limit on the tensor-to-scalar ratio decreases slightly, from $r<0.70$ to $r<0.65$ at $95\%$ confidence.