A general solution to the E-B mixing problem

We derive a general ansatz for optimizing pseudo-C_l estimators used to measure CMB anisotropy power spectra, and apply it to the recently-proposed pure pseudo-C_l formalism, to obtain an estimator which achieves near-optimal B-mode power spectrum errors for any specified noise distribution while minimizing leakage from ambiguous modes. Our technique should be relevant for upcoming CMB polarization experiments searching for B-mode polarization. We compare our technique both to the theoretical limits based on a full Fisher matrix calculation and to the standard pseudo-C_l technique. We demonstrate it by applying it to a fiducial survey with realistic inhomogeneous noise, complex boundaries, point source masking, and noise level comparable to what is expected for next generation experiments (~5.75 uK-arcmin). For such an experiment our technique could improve the constraints on the amplitude of a gravity wave background by over a factor of ten compared to what could be obtained using ordinary pseudo-C_l, coming quite close to saturating the theoretical limit. Constraints on the amplitude of the lensing B-modes are improved by about a factor of 3.