Rigorous versus naïve implementation of the Galerkin method for stepped beams