[CPMD-list] question about 'PROGRAM STOPS IN SUBROUTINE RHOOFR| TOTAL DENSITY SUMS ARE NOT EQUAL'

Fri Dec 1 23:33:57 CET 2006

On Fri, 1 Dec 2006, [gb2312] Åí²© wrote:

CP> Dear list,

dear ???,

CP> I have used CPMD for wavefunction optimization of 
CP> a TiO2 cluster (Ti38O76). At first, it is failed
CP> to converge, so I added 'PCG MINIMIZE' in the &CPMD
CP> section, then, after a few steps, I got the following 
CP> output,
CP> 
CP> ..
CP>  IN FOURIER SPACE:                               912.000000000000
CP>  IN REAL SPACE:                                  912.091989357280
CP>  STOPGM! STACK OF MAIN CALLS:
CP>  STOPGM! CALL    FORCEDR
CP>  STOPGM! CALL     FORCES
CP>  STOPGM! CALL     RHOOFR
CP> 
CP>  PROGRAM STOPS IN SUBROUTINE RHOOFR| TOTAL DENSITY SUMS ARE NOT EQUAL

CP> Is anybody can tell me what's the problem?The whole .inp 
CP> and .out files are attached.

there are four possible reasons for that:
1)  you have a bad pseudopotential:
    did you make any tests on your pseudopotential?
    did you study the output of the pseudopotential
    generation and investigate all warnings.?
    the parameters shipped with the vanderbilt code
    need some tweaking to create a pseudopotential,
    mainly, you need to add a few constraints to 
    avoid negative densities! please try the attached
    one instead.

2)  your atoms are too close:
    please note that in cpmd cell dimensions are
    in atomic units (unless you use the ANGSTROM
    keyword). 
    did you verify the structure as written into
    the GEOMETRY.xyz file? if you check the Ti-O
    distances, you should see that they are _very_
    tight. i guess you have re-calculate the proper
    cell dimension.

3)  your MPI library is corrupting memory
    you are not running in parallel, so this does
    not apply. having to almost double the box in
    every direction will really slow down your
    calculation. so you should start looking for
    a significantly large parallel machine...

4)  you have found a bug in CPMD
    not impossible, but right now 1) and 2) are
    very likely candidates. after switching the
    pseudopotentials, density difference went
    away. changing the cell dimension helped 
    even more.

there are two more remarks about your system:
a) in x- and y-direction your cluster is pretty
close to the supercell boundaries (note you
have to also take the electron density into
account for the size of your system!), so
it is some significant interaction to be expected.
you have to use a poisson solver and/or enlarge
your box in that direction (switching from cubic
to orthorhombix). in z-direction is plenty space.

b) if you use ultra-soft pseudopotentials, especially
in cluster calculations, you need to use a higher
density cutoff than the default of 4x the wavefunction
cutoff, for better convergence. but also you have to
pay attention to the GC-CUTOFF parameter. since you
have more 'noise' in the 'vacuum' area due to the 
lower cutoff. especially, if you are interested in
accurate geometry optimization you should raise the
density cutoff by increasing DUAL to 6.0 or even
higher (some people recommend even 10.0). 
at that level, using norm-conserving pseudopotentials
(where you can stick to the default DUAL of 4.0) may 
be about as fast as ultra-softs. this kind of problem
is less prominents in bulk systems...

CP> I have used the Vanderbilt's code to generated a
CP> uspp file for Ti. (O is obtained from the uspp library) 
CP> >From the previous mailing list, a bad uspp seems to
CP> occur the similar problem, so the uspp file is also 
CP> attached, as I can not be sure whether the uspp I have
CP> generated is good or bad for this situation. 

before using a pseudopotentials, you _always_ have 
to test it with simpler systems. at that stage you
should have already noticed, that there are some
problems with the titanium pseudopotential, if you 
do not add constraints during the pseudopotential
generation to eliminate the negative densities... 
(at least this is how i found it out).

cheers,
   axel.

CP> The structure of the cluster is obtained from anatase
CP> crystal structure.
CP> 
CP> Below is the main body of input
CP> &CPMD
CP> 	OPTIMIZE WAVEFUNCTION
CP> 	INITIALIZE WAVEFUNCTION RANDOM
CP>          (PCG MINIZINE    <<<<<<<<<<<I added it for the second run)
CP> 	CONVERGENCE ORBITALS
CP> 	1.0e-6
CP> 	MAXSTEP
CP> 	100
CP> &END
CP> 
CP> &DFT
CP> 	FUNCTIONAL PBE
CP> 	GC-CUTOFF
CP> 	5.0e-6
CP> &END
CP> 
CP> &SYSTEM
CP> 	SYMMETRY
CP> 	CUBIC
CP> 	SCALE S=1
CP> 	CELL
CP> 	19.044 1.0 1.0 0 0 0
CP> 	CUTOFF
CP> 	25.0
CP> &END
CP> 
CP> &ATOMS
CP> 	ATOMIC CHARGES
CP> 	4.0           
CP> 	-2.0
CP> 
CP> *Ti-gpbe.uspp FORMATTED
CP>   LMAX=D
CP>   38
CP>    ...
CP>   specification
CP>    ...
CP>                       
CP> *O-gpbe.uspp FORMATTED
CP> 	LMAX=P              
CP>   76
CP>    ...               
CP>   specification
CP>    ...
CP> &END  
CP> 
CP> 
CP> 

-- 
=======================================================================
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.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: Ti_VDB_PBE_SEMI.psp.gz
Type: application/x-gzip
Size: 180104 bytes
Desc: 
Url : http://cpmd.org/pipermail/cpmd-list/attachments/20061201/fec13636/attachment.gz 