The ABINIT Users Mailing List ( CLOSED )

[sanix@operamail.com]

Hello,

 As I have mentioned in my previous post, I have troubles when restarting a 
new calculation from an old wavefunction. I did a calculation with 22 SCF 
cycles and had to cancel it, but saved the wavefunction. Now I repeat a 
calculation with exactly the same input file and "irdwfk 1" option. The new 
calculation reads the old wavefunction without any trouble. The only little 
problem is that the occupations are not exactly the same in the new 
calculation, as I use "occopt 4" option and ABINIT does not read the 
occupations from an old wavefunction. I would expect ETOT in the new 
calculation to be very close to the last ETOT in the previous calculation (my 
experience from other codes is that it could vary about 0.01 Ha ). But I get 
a difference of more than 200Ha, what is really a surprise. I enclose the 
input file. I'm using ABINIT 4.3.
If you need a complete input file with positions I can send it, the problem 
is that the calculation needs 3GB of RAM and takes about 10 hours per SCF 
cycle.


-----------------------------------------------------------------

# I-Pt calculation
prtden 1

irdwfk 1

occopt 4
tsmear   0.01

nband 1440
acell 46.249926 46.249926 33.6072

rprim
0.8660254 0.5 0.
0. 1. 0.
0. 0. 1.

ntypat  2

zatnum 78 53
natom 252

typat
1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 2 2 2 2
2 2 2 2 2 2 2 2 2 2 2 2 2 2

nstep 100
toldfe 5.0d-6
ecut 15.0
ixc 11

xred 
..........................
(252 atoms positions follow)

--------------------------------------------------------------------------


Thank you very much for you help,
Alexandre Tkatchenko