Noah Sailer, Kendrick Smith, Yurii Kvasiuk, Alex Laguë, Selim Hotinli

What is the optimal power spectrum estimator on ultra-large scales where the plane-parallel approximation breaks down? Conventional estimators, such as the Yamamoto estimator, are only optimal in the plane-parallel limit, while their associated window functions are typically approximated by truncating a slowly converging infinite series. We address two outstanding challenges in the analysis of wide-angle power spectra. First, we derive the optimal estimator for a broad class of clustering signals and show that it is equivalent to a previously proposed two-$\ell$ generalization of the Yamamoto estimator. Second, we show how to write the exact two-$\ell$ window function as a finite number of terms that can be efficiently evaluated using FFTs. Our results apply to a wide range of observables, including redshift-space distortions (RSDs) and large-scale radial-velocity reconstruction from the kinetic Sunyaev-Zel'dovich effect. Focusing on linear-theory RSDs, we validate the finite window-function representation numerically and show that the two-$\ell$ estimator can yield order-unity improvements in the signal-to-noise ratio of ultra-large-scale power spectrum measurements.