About the 'cutoff' test with GGA

[fangyongxinxi]

Dear,
I have posted one question about the cutoff test with gga before:
http://quantumwise.com/forum/index.php?topic=2110.0
and get a good answer, I increased the cutoff up to 1500 Ry step by step, it still can't convergent.
I thought it might be the reason of the system I used (KFeSe),
 I try the gga with other material, such as NaCl, to test the cutoff
the samething happens, the cutoff set is hard to convergence, even the value is up to 500Ry.

However, this won't happen to LDA (it convergent about ~60 Ry)

So, my question is: what's happens with the cutoff test with GGA ?

Thanks~

[Anders Blom]

It's generally the case that GGA requires a bit higher cut-off since you are evaluating gradients of the density, where in LDA you only need the value of the density. But you should not need crazy high values.

Partly it may be due to the pseudopotential, some of the HGH potentials are very, very hard, so maybe if you can try FHI potentials, unless you are already.

The other question is the definition of "convergence". The total energy is not variational w.r.t. the cut-off in ATK (as it is in a plane-wave code), so it will never saturate smoothly, there will always be a bit of wiggling. So maybe if you start from the other end, and do really, really low cut-offs (like 20 Ry) to see what "bad" energies are like, you may find that your results are reasonably converged already at 100 Ry or so (as they usually are!).