[CPMD-list] problem in wavefunction optimization

Thu Jun 16 18:09:09 CEST 2005

On Wed, 15 Jun 2005, Younes Ansari wrote:

dear younes ansari,

in addition to the comments by matthias a few more from me:

  i agree with matthias, your coordinates seem strange.
  i assume you want to simulate bulk caesium,
  but your input is not. you have to realize that
  the CELL parameters in cpmd refer to the _conventional_
  cell, but you need to specify only the coordinates
  for the _primitive_ cell. thus it is generally less
  error prone to use fractional coordinates (SCALE keyword).
  since you have a large cubic system, it is probably 
  easier to use a simple cubic (SYMMETRY 1) supercell.
  note, that for your g(r) calculation you have to apply
  the periodicity according to the unitcell vectors as
  written in the supercell overview section of the cpmd
  output. btw, most standard example codes assume an 
  orthogonal supercell, which is _not_ the case here.

  here is a simple recipe to check it:
  run a wavefunction optimization with CUTOFF set to 1.0,
  MAXSTEP set 1 and RHOOUT. this way you will run a pretty
  pointless but fast calculation and get a DENSITY file.
  now convert the DENSITY file with cpmd2cube.x to DENSITY.cube
  with 'cpmd2cube.x -inbox -rho DENSITY'
  and load this into the VMD program (you need version 1.8.3 or
  newer). now change the default representation from Lines 
  to CPK and you can see your atoms as they are actually placed
  into the supercell, and you'll probably agree that this
  looks strange.  with the periodic tab you can also add
  periodic images of the unitcell and if you add an Isosurface
  representation, you'll also see the shape of the unitcell.

[...]

YA>  ==     NFI=      1      ETOT=     -8.889130   TCPU=    281.29 ==
YA>  == DRHOMAX= 5.938E-03  DETOT=     0.000E+00    THL= 0.000E+00 ==
YA>  ==------------------------------------------------------------==
YA> 
YA>  <<<<<<<<<<<<<<<<<<< LANCZOS DIAGONALIZATION <<<<<<<<<<<<<<<<<<<<
YA>  >> TIME FOR INITIAL SUBSPACE DIAGONALIZATION:              12.19
YA>  >> CYCLE     NCONV        B2MAX        B2MIN     #HPSI      TIME
YA>         1         0    2.428E-08    7.781E-09      6.00     60.74
YA>  !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
YA>  !! UPDRHO| THE NUMBER OF STATES IS TOO SMALL                  !!
YA>  !!         F(  31,  1) =    2.48822E-05                       !!
YA>  !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!

this is a serious warning. when using the free energy functional
you can have fractional occupations, so you have to make sure
that there are enough empty states to accomodate them. CPMD generally
only adds the absolute minimum, but since you have a rather high
electron temperature (ok, it is the default) you need to provide more.
something like 35 would be more appropriate. please note, that when
you switch to the -q9 semicore pseudopotential, you'll need something
like 290 states as there are also more occupied states. on top of
that, there are some useful recommendations in the cpmd manual for
FEMD calculations.

YA>  !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
YA>  ==------------------------------------------------------------==
YA>  ==     NFI=     38      ETOT=     -8.967467   TCPU=    281.74 ==
YA>  == DRHOMAX= 1.082E-03  DETOT=    -1.027E-02    THL= 1.855E-01 ==
YA>  ==------------------------------------------------------------==
YA> 
YA>  <<<<<<<<<<<<<<<<<<< LANCZOS DIAGONALIZATION <<<<<<<<<<<<<<<<<<<<
YA>  FRIESNER| EIGENVECTOR           1  IS VERY BAD!
YA>  FRIESNER| EIGENVECTOR           2  IS VERY BAD!
YA>  FRIESNER| EIGENVECTOR           3  IS VERY BAD!
YA>  FRIESNER| EIGENVECTOR           4  IS VERY BAD!

ok, now you are in deep trouble. but this is just a follow-up 
to the previous warnings. 

YA> 1- I want to khow how can I apply the Kpoint for the input file. 
YA> and how much should be applied.

via the KPOINTS keyword in the &SYSTEM section.
the easiest way is to use a MONKHORST-PACK mesh, and - similar to
the plane wave cutoff - you need to increase the number of k-points 
until the properties, that are of interest to you, are converged.

in summary, there are several parameters that you need to figure
out, before you can start a serious calculation. i'd recommend
you pick a rather small supercell for regular bulk caesium and 
then first check plane wave cutoff convergence with the different
available pseudopotentials, then convergence with respect to k-points
and finally you should check via computing, e.g., the lattice constant
and the bulk modulus which of the available pseudopotentials
fits your needs best.

if you then increase the size of your unitcell to the dimensions,
that you finally want to simulate, you can easily estimate the required
k-point mesh, from the smaller cell: if you double the size of the cell
in one dimension, you can cut down the MP-mesh in that direction
to half the number of points.

you can find some example inputs (unfortunately still without 
proper explanations) at:
http://www.theochem.ruhr-uni-bochum.de/~axel.kohlmeyer/cpmd-tutor/part7.html#chap8_sect1
please compare the total energy per atom for the different to see
the system size vs. k-points relation.

best regards,
	axel kohlmeyer.

YA> In my previous calculation when the trajectory was made by CPMD the
YA> distances between the atoms was about 0.7 to 6 angstroms so the sphere
YA> diameter for Cesium is 4.72 So how it is possible to get a distance of
YA> 0.7 angstrom. Blease tell me if there would be any problem in my input
YA> file.

YA>  
YA> 
YA>  
YA> 
YA> 
YA> 		
YA> ---------------------------------
YA> Discover Yahoo!
YA>  Get on-the-go sports scores, stock quotes, news & more. Check it out!

-- 

=======================================================================
Dr. Axel Kohlmeyer   e-mail: axel.kohlmeyer at theochem.ruhr-uni-bochum.de
Lehrstuhl fuer Theoretische Chemie          Phone: ++49 (0)234/32-26673
Ruhr-Universitaet Bochum - NC 03/53         Fax:   ++49 (0)234/32-14045
D-44780 Bochum  http://www.theochem.ruhr-uni-bochum.de/~axel.kohlmeyer/
=======================================================================
If you make something idiot-proof, the universe creates a better idiot.