[Pw_forum] Problems using 'relax' with bfgs

[Gabriele Sclauzero]

Dear PW-users,

     while using calculation= 'relax' together with the 'bfgs' algorithm 
I encountered the following problem: the relaxation seems to proceed 
correctly up to a certain point and the reached configuration seems to 
be near the desired minimum position, but in the next bfgs steps the 
systems flips between two configurations and the energy oscillates up 
and down (around the supposed minimum position). At this point the 
algorithm seems not to be able to recognize that we are going back and 
forth and it does not find a configuration that matches the threshold I 
require (i.e. 1.0d-3, that is the default) on the forces.

Shoudn't the algorithm be aware of this situation and do something 
smarter than oscillating back and forth? Or maybe should I adjust any 
parameters with values different from the default ones?

In the attachment is the graph with the energy flips: someone might say 
that the difference is not so much, but the forces on the two atoms I 
let move are still higher than the threshold.
Who might desire the output file (is 400KB large) should ask me, I will 
send it.
Here is the input:
-------
   &control
     calculation='relax',
     restart_mode='from_scratch',
     forc_conv_thr= 1.0d-3,
     tprnfor= .TRUE.,
     pseudo_dir = '/u/cm/sclauzer/Pseudo_/',
     outdir='/local_scratch/sclauzer/tmp/',
     prefix='COat6Ptwire',
   /
   &system
     ibrav = 6,
     celldm(1) = 18,
     celldm(3) = 1.47398638,
     nat= 8,
     ntyp= 3,
     ecutwfc = 29,
     ecutrho = 300,
     occupations= 'smearing',
     smearing='methfessel-paxton',
     degauss=0.01
   /
   &electrons
     conv_thr = 1.0e-8
     mixing_beta = 0.6
   /
   &ions
     ion_dynamics= 'bfgs',
     upscale= 10.d0,
   /
ATOMIC_SPECIES
Pt  195.078  Ptsrnlcc.RRKJ3.UPF
C  12.0107  C.pz-rrkjus.UPF
O  15.9994  O.pz-rrkjus.UPF
K_POINTS AUTOMATIC
1 1 13 0 0 1
ATOMIC_POSITIONS ANGSTROM
C  2.2  0.0  0.0    1 0 0
O  3.343  0.0  0.0    1 0 0
Pt  0.0  0.0  -7.02   0 0 0
Pt  0.0  0.0  -4.68   0 0 0
Pt  0.0  0.0  -2.34   0 0 0
Pt  0.0  0.0  0   0 0 0
Pt  0.0  0.0  2.34   0 0 0
Pt  0.0  0.0  4.68   0 0 0
-------


Gabriele
-------------- next part --------------
A non-text attachment was scrubbed...
Name: Etot_cycle.ps
Type: application/postscript
Size: 16704 bytes
Desc: not available
Url : /pipermail/attachments/20060302/4ec61d0b/attachment.ps