[CPMD-list] convergence perfect with OLDCODE, fails with NEWCODE

[Juerg Hutter]

Hi

this problem occurs because the two implementation of functionals use
a different approach for the low density and low gradient limit.
The PBE functional was specially affected (see also some older postings).
In the upcomming release the behavior of OLDCODE and NEWCODE should also
for PBE be comparable.
The difference of the two implementation is mainly speed. Newcode
should always be faster (on some computer architectures much faster).

regards

Juerg Hutter

> Dear CPMD developers and users,
>
> I would like just to report that sometime the NEWCODE option (which is the
> default) is not working while the OLDCODE one works fine in CPMD 3.9.1.
> I found it when calculating wavefunctions of both O and S atoms in triplet
> states. I was not able to achieve convergence with the NEWCODE.  I used
> ultrasoft pseudopotentials from David Vanderbilt's web page. Below is my
> NEWCODE input file. The succesful run used the same input file except that
> the NEWCODE keyword was replaced by the OLDCODE one.
>
>
> &CPMD
>   RESTFILE
>     2
>   STORE
>     100
>   OPTIMIZE WAVEFUNCTIONS
>   PCG
>   CONVERGENCE ORBITALS
>     1.0D-7
>   CENTER MOLECULE OFF
>   SPLINE POINTS QFUNCTIONS
>     2001
>   MEMORY BIG
>   LSD
> &END
>
> &DFT
>   NEWCODE
>   GRADIENT CORRECTION PBEX PBEC
>   GC-CUTOFF
>     5.0D-5
> &END
>
> &SYSTEM
>   SYMMETRY
>     0
>   POISSON SOLVER TUCKERMAN
>   ANGSTROM
>   CELL
>     13.0D0  1.0D0  1.0D0  0.0D0  0.0D0  0.0D0
>   TESR
>     4
>   CUTOFF
>     25.0D0
>   MULTIPLICITY
>     3
> &END
>
> &ATOMS
> *O.uspp BINARY NEWF TPSEU
>   LMAX=D
>    1
>    5.8D0  5.5D0  7.1D0
> &END
>
>
> On the output I am getting enormous quantities like
>
>  NFI      GEMAX       CNORM           ETOT        DETOT      TCPU
>    1  4.424E+05   6.544E+04     -15.263055    0.000E+00      3.04
>    2  6.100E+06   1.889E+06      23.493112    3.876E+01      2.98
>    3  4.057E+06   1.120E+06      23.215886   -2.772E-01      2.99
>    4  3.780E+06   1.230E+06      23.008182   -2.077E-01      3.02
>
> It persists when using the default optimization method or starting from
> random wavefunctions.
>
> Thanks for your attention to CPMD development.
> Best regards
> Martin Konopka.
>
>
> ------------------------------------------------------------------------
> Dr. Martin Konopka             http://www.ccms.elf.stuba.sk/konopka.html
> Department of Physics, CCMS                      tel: +421 (0)2 60291714
> Slovak University of Technology (FEI STU)        fax: +421 (0)2 65411483
> Ilkovicova 3, 812 19 Bratislava, Slovakia
> ------------------------------------------------------------------------
> _______________________________________________
> CPMD-list mailing list
> CPMD-list at cpmd.org
> http://cpmd.org/mailman/listinfo/cpmd-list
>