Origin of radio polarization in pulsar polar caps

Context. It is crucial to know the polarization properties of coherent radio waves that escape from pulsar polar caps to calculate the radiative transfer through the magnetosphere and to predict observable radio properties. Aims. We describe pair cascades in the pulsar polar cap, and we determine for the first time the Stokes parameters of the escaping radio waves from first-principle kinetic simulations for a pulsar with a magnetic obliquity of 60°. Methods. We present 3D particle-in-cell kinetic simulations that include quantum-electrodynamic pair cascades in a charge-limited flow from the stellar surface. Results. Our model quantitatively and qualitatively explains the observed pulsar radio powers and spectra, the pulse profiles, polarization curves, their temporal variability, the strong Stokes-L and weak Stokes-V polarization components, the decline in the linear polarization with frequency, and the nonexistence of a radius-to-frequency relation. The observable properties of radio emission from the polar cap can vary and include single- or double-peaked profiles. Most of the Stokes V curves from our simulations appear to be antisymmetric, but symmetric curves are also present at some viewing angles. Although the polarization-angle (PA) swing of the radiation from the polar cap fits the rotating vector model (RVM) for most viewing angles, the angles obtained from the RVM do not correspond to the dipole geometry of the magnetic field. Instead, the PA is directly related to the plasma flows in the polar cap. Furthermore, we found that the radiation is associated with escaping plasma bunches and can propagate freely along channels of low plasma density, in addition to being reflected at the channel boundaries. Conclusions. Our simulations demonstrate that pair discharges close to the surface of the polar cap cause the radio emission of pulsars and determine the majority of their typically observed properties. The merits of RVM for estimations of the magnetic field geometry from observations need to be reevaluated.