[CPMD-list] EWALD SUM not equal in real and Q-space

[Juerg Hutter]

Hi

the problem seems to be realted to the PGI compiler only.
If you replace in rhov.F the REAL by DREAL (line 105)
the problem should disappear

      IF(GEQ0) THEN
        RSUMV=DREAL(DELTAR(1))
      ELSE
        RSUMV=0.0D0
      ENDIF

PS: some tricks to be considered using USPP
    
    SPLINE QFUNCTIONS
    
    may reduce the time for initialization and the
    memory requirements.

    BIG MEMORY

    increases the memory requirements again, but usually
    saves some CPU time in USPP calculations.

PPS: the SURFACE keyword is only effective together
     with SYMMETRY 0 and requires special care of
     orienting the system!

Juerg

----------------------------------------------------------
Juerg Hutter                   Phone : ++41 1 635 4491
Physical Chemistry Institute   FAX   : ++41 1 635 6838
University of Zurich           E-mail: hutter at pci.unizh.ch
Winterthurerstrasse 190
CH-8057 Zurich, Switzerland
----------------------------------------------------------


On Mon, 12 Aug 2002, Shiwu Gao wrote:

> Hi Juerg,
> 
> Attached please find the working directory of pseudopotential 
> generation for Pt, which contains both the input and output 
> for the atomic calculation and ps generation. Note that the 
> case that I sent you does not have any problem, if I use other
> forms of pp. It is only problematic when running with uspp. 
> This might be a problem with uspp then. However I tested with 
> the pseudopotential for 022-Ti, which comes with the uspp 
> package, and was generated by Vanderbildt? The same problems
> occurs. So I am not sure whether the problem is related to 
> my pseudopotential, or uspp implementation in CPMD. 
> 
> I also tested on other machines here. The same problems. So
> it is unlikely to be a fault of computer/OS. 
> 
> Best regards,
> 
> Shiwu
> 
> 
> Juerg Hutter wrote:
> > 
> > Hi
> > 
> > I still can't reproduce your problem. However, as you sent
> > me a binary pseudopotential I was not able to use it.
> > I replaced it by another pp and the charge densities were
> > accurate to machine precision both in real and G space.
> > There are now two possibilities:
> > 1) the problem is connected to your pp
> >    -> please send the input files to the atomic code
> > 2) the problem is connected to your computer/compiler/os etc.
> >    -> try the input on another type of computer
> > 
> > Juerg
> > 
> > PS: send mails also to the mailing list. There might be
> >     other people interested in this problem, or people
> >     that can help.
> > 
> > ----------------------------------------------------------
> > Juerg Hutter                   Phone : ++41 1 635 4491
> > Physical Chemistry Institute   FAX   : ++41 1 635 6838
> > University of Zurich           E-mail: hutter at pci.unizh.ch
> > Winterthurerstrasse 190
> > CH-8057 Zurich, Switzerland
> > ----------------------------------------------------------
> > 
> > On Sun, 11 Aug 2002, Shiwu Gao wrote:
> > 
> > > Dear Juerg,
> > >
> > > Attached please find a test case, where I still have probem.
> > > For smaller systems (than this one), there is no such problems.
> > > In the tar.gz file you also find the output and uspp that
> > > I use. Please let me what you find and where the problem
> > > is.
> > >
> > > Many thanks again,
> > >
> > > Shiwu
> > >
> > > PS
> > >
> > > For you information:
> > >
> > > The test case is a 2-layer slab model of Pt(111) surface in
> > > p(3x3) 2-D symmetry, and the calculation was done my laptop
> > > with RedHat Linux 7.2 as OS.
> > >
> > >
> > > Juerg Hutter wrote:
> > > >
> > > > Hi
> > > >
> > > > I can't reproduce this problem. Could you post an input
> > > > for a system (as small as possible) where this error
> > > > occurs?
> > > >
> > > > Juerg
> > > >
> > > > ----------------------------------------------------------
> > > > Juerg Hutter                   Phone : ++41 1 635 4491
> > > > Physical Chemistry Institute   FAX   : ++41 1 635 6838
> > > > University of Zurich           E-mail: hutter at pci.unizh.ch
> > > > Winterthurerstrasse 190
> > > > CH-8057 Zurich, Switzerland
> > > > ----------------------------------------------------------
> > > >
> > > > On Fri, 9 Aug 2002, Shiwu Gao wrote:
> > > >
> > > > > Hi,
> > > > >
> > > > > I got a trouble in running CPMD with the Vanderbildt uspp,
> > > > > which I generated myself. The program stops and complains that
> > > > > the Ewald sum does not equal in real and Fourier Q-space,
> > > > >
> > > > > Typical error message looks like this
> > > > >
> > > > >   ****************************************************************
> > > > >
> > > > >  CPU TIME FOR WAVEFUNCTION INITIALIZATION:           8.73 SECONDS
> > > > >  ***    RWFOPT| THE NEW SIZE OF THE PROGRAM IS   13844 kBYTES ***
> > > > >  EWALD SUM IN REAL SPACE OVER  5* 5* 5 CELLS
> > > > > IN FOURIER SPACE:                               360.000005713858
> > > > > IN REAL SPACE:                                  359.999999999984
> > > > >
> > > > >  *****************************************************************
> > > > >
> > > > > Note the charge should be 360 in this case. Tested many uspp's including
> > > > > the potential included in the latest uspp package. All have the same
> > > > > problem. Note that the sum in real space is usually OK with an error
> > > > > less than 10^-10, while the error in Q-space is large, 10^-5--10^-6.
> > > > >
> > > > > Has anybody met this problem with uspp? What is the way out of it?
> > > > >
> > > > > Thanks in advance,
> > > > >
> > > > > Shiwu Gao
> > > > > _______________________________________________
> > > > > CPMD-list mailing list
> > > > > CPMD-list at cpmd.org
> > > > > http://www.cpmd.org/mailman/listinfo/cpmd-list
> > > > >
> > > >
> > > > _______________________________________________
> > > > CPMD-list mailing list
> > > > CPMD-list at cpmd.org
> > > > http://www.cpmd.org/mailman/listinfo/cpmd-list