Hello guys,
Thanks for the tips Marcos, note however that this Pt-pseudo has
never been used before (the guy wants to generate his own because
none is suitable from Siesta's homepage).
Thus I couldn't resist trying, and found out that the problem is
with the L=1 orbital, namely, a too short r_c value causes a
ghost state. Starting from about 2.6 Bohr there are no more
ghosts, though the L=1 pseudo still looks a bit odd.
HTH
Roberto
PS: I used the gen-basis utility to check for ghosts.
On Thu, 27 Oct 2011, Marcos Ver?ssimo Alves wrote:
Guys,
If this pseudo has been used before, it most likely doesn't present ghost
states. If it did, Siesta would have stopped. So one possibility for these
ghost states is that the compiler options are such that wrong results are
being output. This is especially strong in ifort. If it is ifort, I
recommend complete re-compilation of both Siesta ***and*** atom with -O2, in
case the compilation has been done with -O3. Some years ago, one of the
users had this kind of problem - a pseudo that had been already used and
tested presented ghost states after being re-calculated - , and the use of
-O2 instead of -O3 solved the issue.
Hope this helps.
Marcos
On Thu, Oct 27, 2011 at 9:26 AM, R.C.Pasianot <[email protected]> wrote:
Hi Bo Xiao,
So, you mean that when specifying core corrections you get
ghost states but everything's fine if not ?.
You're putting rcore=0.77, may I ask why ?.
My way is letting the code choose, namely, rcore_flag=1.0,
rcore=0.0, which effectively means that rcore will be
such that (core_charge_density = valence_charge_density).
This is quite an accepted criteria. Eventually you can play
manually with rcore later, after seeing what the code gives
in the 1st place.
Goog luck,
Roberto
