[CPMD-list] PBE + NLCC convergence problem

[Oleg Yazyev]

Dear cpmd community,

I've found the strange problem. Probably, somebody already have seen
this and know how to solve it or this observation will help to fix
some bugs.
Generally, when using PBE GGA and pseudopotentials with NLCC SCF
always blows up after few first iterations (see below the example).
The problem is well reproducible (at least all openshell molecular
systems I'm interested in shows this behaviour) but appears only with
PBE GGA (at least LDA, BP and BLYP are OK) and never appears if NLCC
is not used.

 NFI      GEMAX       CNORM           ETOT        DETOT      TCPU
   1  1.994E-02   2.166E-03      -7.223573    0.000E+00     44.19
 LINE SEARCH : LAMBDA=.422     PREDICTED ENERGY = -7.361147976013
   2  1.624E-02   1.836E-03      -7.281170   -5.760E-02     44.01
 LINE SEARCH : LAMBDA=.633     PREDICTED ENERGY = -7.407550491052
   3  9.908E-03   1.222E-03      -7.361091   -7.992E-02     43.86
 LINE SEARCH : LAMBDA=.158     PREDICTED ENERGY =  8.255565606249
   4  1.716E+02   2.868E+01      -7.407496   -4.641E-02     42.93
 LINE SEARCH : LAMBDA=.475     PREDICTED ENERGY =  6.515199589623
   5  5.391E+01   1.544E+01       6.515225    1.392E+01     42.56
 LINE SEARCH : LAMBDA=.712     PREDICTED ENERGY =  6.515174606737
   6  6.666E+01   2.017E+01       6.515202   -2.253E-05     42.59
 LINE SEARCH : LAMBDA=1.07     PREDICTED ENERGY =  6.514689978070
   7  1.564E+01   5.375E+00       6.515178   -2.471E-05     42.64
 LINE SEARCH : LAMBDA=1.60     PREDICTED ENERGY =  6.514698812780
   8  3.005E+01   2.038E+01       6.514744   -4.335E-04     42.87
 LINE SEARCH : LAMBDA=2.40     PREDICTED ENERGY =  6.513880859113
   9  9.822E+00   4.859E+00       6.514704   -4.031E-05     42.47
 LINE SEARCH : LAMBDA=3.60     PREDICTED ENERGY =  6.513401997207
  10  2.021E+01   6.364E+00       6.513972   -7.315E-04     42.84


-- 
Best regards,
 Oleg


______________________________________________________
Oleg Yazyev
Institute of Molecular and Biological Chemistry
Swiss Federal Institute of Technology in Lausanne (EPFL)
EPFL - BCH
CH-1015 Lausanne (Switzerland)
Tel.: +41 21 693 9881
E-mail: oleg.yazyev at epfl.ch
WWW: http://icmbcu001.epfl.ch/yazyev/index.html
_______________________________________________________