The ABINIT Users Mailing List ( CLOSED )

[Masayoshi Mikami <mmikami@rc.m-kagaku.co.jp>]

Dear Olivier,

Please let me post my guess:
Which kind of pseudopotential is adapted ?
(from the ABINIT web site ? or the potential you made ?)

In any case, the main difference between Nb and V may
be "4d v.s. 3d states."  The 3d states should be much
more localized than 4d states, so that the necessary ecut
may be much larger than that of Nb (17 Ha), still larger than
26 Ha for RF calculation. (RF calc. may need larger ecut of GS calc...)

If you have FHI98PP code (and/or OPIUM code),
you might want to estimate adequate energy cutoff
by generating the V pseudo as well as Nb pseudo.
You will see the estimated ecut of V is larger than that of Nb.
(i.e. 17 Ha may be enough for Nb, but may not be sufficient for V.
 You might want to increase ecut up to 40Ha ...)
You can refer to the following review to know "how to estimate
necessary ecut from isolated pseudo-atom calculation":
M. Fuchs and M. Scheffler, "Ab initio pseudopotentials for electronic 
structure calculations of poly-atomic systems using density-functional 
theory," Comput. Phys. Commun. 119, 67-98 (1999).

In passing, it might be a good occasion to remember
the old story of the Nb phonon-band issue before v.3.0.3.
Please let me cite the old e-mail that is not archived on the ABINIT 
WEB.
(before the present ML system was adopted ...)
This issue was the motivation to prepare Test_physics ...

Good luck,
Masayoshi

We had a problem with v.2.3.x  ;
From Xavier's mail (Date: Tue, 12 Sep 2000 10:05:32 +0200),
Subject: ABINIT : Problems with response-function ; 41nb.pspnc flawed .
-----------------
        (snip)
- trying other metals, I have observed that everything is OK
for Aluminum, but that for Niobium, the same problem is observed.
Playing with many different parameters does not give stable
acoustic modes. However, the phonon band structure away from gamma
is fine (despite the tortured character of the Nb phonon band 
structure).
        (snip)
-----------------

Then, it was finally solved.
--------------------------------
Date: Tue, 27 Feb 2001 17:29:10 +0100
From: Xavier Gonze <gonze@pcpm.ucl.ac.be>
To: abinit@pcpm.ucl.ac.be
Subject: (ABINIT) v3.0.3 available

Dear Abinitioners,

ABINIT version 3.0.3 is available ... You will find it on the Web site,
as usual.

The complete sources and tests are available, as well as the
binaries for Intel/Linux (seq), Compaq/DEC_EV67_OSF (seq+par),
Compaq/DEC_EV56_OSF (seq), IBM RS6000 44P (seq), HP 8500 (seq).

The list of known problems can be found on the Web, as
well as the modifications with respect to version 3.0.2 .

The problem with the Nb phonon band structure has been solved.
Actually, it was not an ABINIT problem, but a pseudopotential
problem. The Nb phonon band structure is very
sensitive, so that changes in pseudopotentials can lead (or not)
to instabilities ! A careful comparison with the code CPMD
(many thanks to Thierry Deutsch !) has shown that the output of
both codes are the same to better than 1 meV, as soon as the
same pseudopotential is used. For a large elongated supercell,
instabilities were found by both codes.
The instability appears with the
HGH Nb pseudopotential, and not with a FHI98PP pseudopotential,
that Martin Fuchs generated recently (many thanks to him !).

Then, there is a complication to this study, in that there was
nevertheless a bug in ABINIT in the response function part of the code,
related with the treatment of the action of symmetries on dynamical
matrices. It lead to erroneous results when a Nb conventional
cell (with natom=2, and 96 (!) symmetry operations) was used
in the simulation. This bug appeared only in specific structures,
but it is impossible to describe the class of these structures :
if you have doubts about the correctness of pre-v3.0.3 results,
try to redo one calculation with v3.0.3 ...

This bug has been removed. The complete Nb phonon band structure
should be computed in the future, just to be sure that everything is
understood.

One of the bugs causing crash for the Windows OS has also been
eliminated (many thanks to L. Sindic).
Basically, the multi-dataset features are now OK for Windows OS,
while the code might still crash for response-functions.

Then, many modifications, bug removals ... have been done with respect
to v3.0.2 (see the list of changes, below).

So, it is strongly encouraged to switch to ABINITv3.0.3 for production 
work.
It is to be preferred over ABINITv2.3 or earlier versions, especially 
for
the computation of phonon response at non-zero wavevector.

Many thanks to N.Capron, R. Caracas, Mikami-san, J.-Y. Raty, A.Roy,
L. Sindic, M. Verstraete, for bug reports or modifs,
and J.-M. Beuken for the Web site.

Have fun !
Xavier

        (hereafter, release_note)
--------------------------------