[CPMD-list] help, the problem still exist

[Axel Kohlmeyer]

On Sun, 4 Dec 2005, [gb2312] Shangyi Ma wrote:

SM> Dear apsi:
SM>      Thanks for your reply, but the problem is still exist!

dear shyma,

SM>      I'm troubled in getting the cohesive energy of Al. The cohesive
SM> energy calculation are performed by subtracting the total energy(E1)
SM> of the spin ploarized Al atom from the energy(E2) per atom in the
SM> bulk. So I must get the different energys E1 and E2 firstly.

first off, you should note, that you are digging into parts of the
CPMD code, that are not so frequently used and thus there is less
expertise available and the usage of that part of the code needs you
to do have a deeper understanding of the underlying physics. or to
put it differenty: in the old times people would have written 
'here be dragons' (remember that in europe dragons are usually
fear inspiring creatures).

SM>      In your last letter, you said "The easiest way to calculate the
SM> energy of the single atom is to use MULTIPLICITY of 4 (three
SM> unpaired electrons) and no FREE ENERGY FUNCTIONAL (and you can erase
SM> keyword 'STATES'). Please notice that you have a typo in
SM> 'TEMPERATURE ELCECTRON', however in this case it wouldn't matter
SM> because the default value is the same, but in case of molecular
SM> dynamics this would be interpreted as _ionic_ temperature...", so
SM> need I use the FREE ENERGY FUNCTIONAL if I want to get the
SM> energy(E2) per atom in the bulk?

since you want to compute the _difference_ of energies, you have to
use the same method. IIRC, using FEMD will shift the total energy 
depending on the electron temperature, so you have to use the same 
settings for both calculations. of course you need to run an open-shell
calculation in both cases, too.

SM>      The energy of spin ploarized Al atom I get is -1.95126380 A.U.
SM> and -1.82387924 A.U.  when I specified MULTIPLICITY 2 and 4
SM> respectively.(In my mind, the electron configuration of Al is
SM> 1s(2)2s(2)2p(6)3s(2)3p(1) and one unpair electron, so I think the
SM> multiplicity is 2*(1/2)+1=2.But you specified MULTIPLICITY 4 in your
SM> letter, does it means the electron configuration is
SM> 1s(2)2s(2)2p(6)3s(1)3P(2) and three unpair electrons, so what are
SM> the difference and which multiplicity value should I specified when
SM> I get the energy of spin ploarized Al atom?) The input file for the
SM> spin Al atom calculation is following:

i have to disagree with ari, i think you are correct in using 
a multiplicity of 2.

SM> *******************************************************
SM> &CPMD
SM>   OPTIMIZE WAVEFUNCTION
SM>   CONVERGENCE ORBITALS
SM>   1.0D-6
SM>   PCG MINIMIZE
SM>   TIMESTEP
SM>   20
SM>   LSD
SM>   MIRROR
SM> &END
SM>                                                                                 
SM> &SYSTEM
SM>   SYMMETRY
SM>   0

there is little need to use a poisson solver (symmetry 0)
here. just make sure your box is large enough. 

SM>   MULTIPLICITY
SM>   2                 (4 or 2?)
SM>   POISSON SOLVER TUCKERMAN
SM>   ANGSTROM
SM>   CELL
SM>   10.0  1.0  1.0  0.0  0.0  0.0
SM>   CUTOFF
SM>   30

SM> Then I calculate the energy per atom in bulk with FREE ENERGY
SM> FUNCTIONAL and without it, the E2 are -2.10337731 A.U. and
SM> -2.01913733 A.U. respectively, the following are the input files
SM> with and without free energy funcitonal:

SM> 
SM> ******************************************************
SM> &CPMD
SM>   OPTIMIZE WAVEFUNCTION
SM>   FREE ENERGY FUNCTIONAL
SM>   CONVERGENCE ORBITALS
SM>   1.0D-6
SM>   MIRROR
SM> &END
SM>                                                                                 
SM> &SYSTEM
SM>   SYMMETRY
SM>   FCC
SM>   ANGSTROM
SM>   CELL
SM>   3.95  1.0  1.0  0  0  0   (#I get the equilibration lattice with the energy vs variety lattice)
SM>   CUTOFF
SM>   60

whoa, there! why suddenly a cutoff of 60ry?

SM>   TESR
SM>   3
SM>   STATES
SM>   4
SM>   KPOINTS MONKHORST-PACK
SM>   8 8 8
SM> &END

SM> the cohesive energy I get with the above the values are far away the
SM> experiment values 3.45 ev. Could you help me solve the problem and
SM> get the calculation cohesive value approach the experiment values?

well, you cannot compare two energies with different basis set sizes.
in fact, there are a number of issues that you have to pay attention to, 
many of them related to using a plane wave basis set:
a) you have to make sure, that your energy difference are converged
 with respect to the plain wave cutoff. mind you, the _differences_ not
 the total energy. due to error cancellation, the difference should
 converge much faster. converging the total energy would usually require 
 a higher plane wave cutoff. feel free to do it, but you are wasting 
 your (cpu) time.
b) you have to make sure, that your bulk system's total energy is 
 converged with respect to the number of k-points and the electron 
 temperature. a higher electron temperature should correspond to
 a larger smearing of the electrons, which in turn should facilitate
 an easier k-point convergence, but in turn it will take you further
 away from the 'cold' real minimum. so you have to find a good  
 compromise, that also represents other bulk properties well.
c) you then have to make sure, that all relevant parameters for 
 the two runs (single atom and bulk) are comparable. most importantly,
 plain wave cutoff (and density cutoff), but also the number of grid
 points per volume should be the same (note, that these numbers are
 chosen based on supported FFT-grids, and thus it is most likely that
 you get comparable grids when the lattice constant of the single atom
 box is an integer multiple of the bulk system). you can set the grid
 manually, but you have to double check, because the FFT-will use the
 smallest _supported_ grid, which may be larger that what you chose.

SM>    Another problem about project-dos is that I want to get the
SM> information that the dos vs energy on s,p,d ...orbitals, I check the
SM> LDOS and PROJECT ANALYSIS and don't know how to get the information
SM> I need, could you explain it to me in details?

i am sorry but, no. i know there is a tool in the pwscf package, 
that can do this, and perhaps somebody has dome something similar 
for cpmd. if yes, please speak up. otherwise, you probably have to 
figure it out for yourself.

SM>    I'm poor on the sphere of ab initio calculation and maybe some
SM> questions is stupid and not professional, but I really need your
SM> help thirstly!

well, the best way to deal with it, is to use your thirst and _learn_ 
the method instead of letting others doing the work for you. there is
a ton of textbooks, reviews, lecture notes, and you can learn a lot 
from reading the source code or by just figuring things out, using
common sense and careful double checking of your results. at least
on the last part, you have already come pretty far, so that should be
encouraging. since there are many more people asking about these 
issues, than answering, you may even become the 'CPMD-expert' in
this area pretty fast (all you need is to know a little more and
have a bit experience). as i wrote before, chances are pretty slim
that you'll get anything better than people like me mostly trying 
to guess, what is going on.

with best regards,
     axel.

SM>    Thanks in advace!
SM>    Best regards!
SM>    shyma  
SM> 

-- 
=======================================================================
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.