Bounding the relative errors associated with a complete Stokes polarimeter

In this paper, we propose a criterion for bounding the relative errors associated with the determination of the Stokes vector that describes the state of polarization of a light beam. No assumptions about the magnitude, origin, or statistical behavior of the errors are made. It is shown that figures of merit such as the condition number and the equally weighted variance naturally arise as optimization parameters. Moreover, a third optimization parameter emerges, which takes into account errors associated with the matrix that represents the selected configuration of analyzers. Finally, a new and more general figure of merit is derived from this analysis and is applied in an optimization process of a very well known polarimeter.