For some reason or the other, the calculation seems to converge much faster for iscf=17 than for iscf=7. When using iscf=17, however, I get warnings about the density going negative. Taking a look at the code, it says:

!! Note that, ignoring model core charge, as long as boxcut>=2

!! the shifted density is derivable from the square of a Fourier

!! interpolated charge density => CANNOT go < 0.

...

!! Fourier interpolated core charge can go < 0 due to Gibbs

!! oscillations;...

Does this mean that the negative densities could arise from the fact that I have a pseudo-core? How come this only happens for iscf=17?

Ecut and boxcut should be OK. Any good suggestions what to do otherwise to prevent this behaviour?