[Pw_forum] bfgs error

[Axel Kohlmeyer]

On Sun, 29 Jul 2007 brsahu at physics.utexas.edu wrote:


dear sahu,

BS> Dear PWSCF users,
BS> 
BS> I encountered an error in a calculation (carbon slabs, ultrasoft PP,  
BS> ecutwfc = 30 Ry and ecutrho = 250 Ry)
BS> 
BS> CASE: energy_new > energy_old
BS> 
BS>       new trust radius        =       0.0000495950 bohr
BS> 
BS>       trust_radius < trust_radius_min
BS> 
BS>       resetting bfgs history

[...]

BS> The solution suggested in the mailing list (scf convergence criteria  
BS> can not be greater than the force
BS> criteria) is contrary to my inputs

it is not entirely clear to me, how i should understand this,
but having a tight SCF convergence parameter (conv_thr,  about two 
or three orders of magnitude smaller than forc_conv_thr) is a
pre-requiste for well-converged forces.

what you are seeing is that your forces are still fluctating 
a lot with small displacements (the reason for the BFGS resets).
so you have to consider other sources.

what comes to my mind are: the density cutoff, bad pseudopotentials,
k-point convergence/smearing. your total force is already very
close to the desired accuracy, so it may be something that has
only a smaller impact on forces, hence my suggestion would be
to try with cranking up the density cutoff to 300ry.

cheers,
   axel.

BS> 
BS> as
BS> 
BS>   Total force =     0.003851     Total SCF correction =     0.000054
BS> 
BS> 
BS>       entering subroutine stress ...
BS> 
BS>            total   stress  (Ry/bohr**3)                   (kbar)     P=   -8.86
BS>    -0.00015635   0.00000000   0.00000000        -23.00      0.00      0.00
BS>     0.00000000  -0.00001219  -0.00000281          0.00     -1.79     -0.41
BS>     0.00000000  -0.00000281  -0.00001204          0.00     -0.41     -1.77
BS> 
BS> 
BS>       number of scf cycles    =  61
BS>       number of bfgs steps    =  47
BS> 
BS>       energy old              =    -370.1139086454 Ry
BS> 
BS>       energy new              =    -370.1138810384 Ry
BS> 
BS> and in the input file I have
BS> 
BS> forc_conv_thr = 1.0D-3
BS> conv_thr =  1.0d-9
BS> 
BS> etot_conv_thr is default
BS> 
BS> any suggestions?
BS> 
BS> Sahu
BS> UT-Austin, USA
BS> _______________________________________________
BS> Pw_forum mailing list
BS> Pw_forum at pwscf.org
BS> http://www.democritos.it/mailman/listinfo/pw_forum
BS> 

-- 
=======================================================================
Axel Kohlmeyer   akohlmey at cmm.chem.upenn.edu   http://www.cmm.upenn.edu
   Center for Molecular Modeling   --   University of Pennsylvania
Department of Chemistry, 231 S.34th Street, Philadelphia, PA 19104-6323
tel: 1-215-898-1582,  fax: 1-215-573-6233,  office-tel: 1-215-898-5425
=======================================================================
If you make something idiot-proof, the universe creates a better idiot.