Convergence of the numerical error with resolution for the vortex advection test. The error is measured with the L₁-norm (crosses), L₂-norm (dots), and L_∞-norm (square). Results are obtained with the inexact Newton method (η = 10^-6). Three different values of the nonlinear tolerance are investigated: 10^-2,10^-4,10^-6 (from left to right). The labels show the inferred rates of convergence of the different errors.