''The problem is [...] in the fact that the exchange and correlation energy is computed in real space on a discrete grid and hence the total energy is invariant (...) only for translation in the FFT grid. Increasing the charge density cutoff increases the grid density thus making the integral more exact thus reducing the problem, unfortunately rather slowly...This problem is usually more severe for GGA than with LDA because the GGA functionals have functional forms that vary more strongly with the position; particularly so for isolated molecules or system with significant portions of "vacuum" because in the exponential tail of the charge density a) the finite cutoff (hence there is an effect due to cutoff) induces oscillations in rho and b) the reduced gradient is diverging.''(info by Stefano de Gironcoli, June 2008)